55 skills found · Page 1 of 2
jvm-profiling-tools / Honest ProfilerA sampling JVM profiler without the safepoint sample bias
facebookresearch / BalanceThe balance python package offers a simple workflow and methods for dealing with biased data samples when looking to infer from them to some target population of interest.
sayantann11 / All Classification Templetes For MLClassification - Machine Learning This is ‘Classification’ tutorial which is a part of the Machine Learning course offered by Simplilearn. We will learn Classification algorithms, types of classification algorithms, support vector machines(SVM), Naive Bayes, Decision Tree and Random Forest Classifier in this tutorial. Objectives Let us look at some of the objectives covered under this section of Machine Learning tutorial. Define Classification and list its algorithms Describe Logistic Regression and Sigmoid Probability Explain K-Nearest Neighbors and KNN classification Understand Support Vector Machines, Polynomial Kernel, and Kernel Trick Analyze Kernel Support Vector Machines with an example Implement the Naïve Bayes Classifier Demonstrate Decision Tree Classifier Describe Random Forest Classifier Classification: Meaning Classification is a type of supervised learning. It specifies the class to which data elements belong to and is best used when the output has finite and discrete values. It predicts a class for an input variable as well. There are 2 types of Classification: Binomial Multi-Class Classification: Use Cases Some of the key areas where classification cases are being used: To find whether an email received is a spam or ham To identify customer segments To find if a bank loan is granted To identify if a kid will pass or fail in an examination Classification: Example Social media sentiment analysis has two potential outcomes, positive or negative, as displayed by the chart given below. https://www.simplilearn.com/ice9/free_resources_article_thumb/classification-example-machine-learning.JPG This chart shows the classification of the Iris flower dataset into its three sub-species indicated by codes 0, 1, and 2. https://www.simplilearn.com/ice9/free_resources_article_thumb/iris-flower-dataset-graph.JPG The test set dots represent the assignment of new test data points to one class or the other based on the trained classifier model. Types of Classification Algorithms Let’s have a quick look into the types of Classification Algorithm below. Linear Models Logistic Regression Support Vector Machines Nonlinear models K-nearest Neighbors (KNN) Kernel Support Vector Machines (SVM) Naïve Bayes Decision Tree Classification Random Forest Classification Logistic Regression: Meaning Let us understand the Logistic Regression model below. This refers to a regression model that is used for classification. This method is widely used for binary classification problems. It can also be extended to multi-class classification problems. Here, the dependent variable is categorical: y ϵ {0, 1} A binary dependent variable can have only two values, like 0 or 1, win or lose, pass or fail, healthy or sick, etc In this case, you model the probability distribution of output y as 1 or 0. This is called the sigmoid probability (σ). If σ(θ Tx) > 0.5, set y = 1, else set y = 0 Unlike Linear Regression (and its Normal Equation solution), there is no closed form solution for finding optimal weights of Logistic Regression. Instead, you must solve this with maximum likelihood estimation (a probability model to detect the maximum likelihood of something happening). It can be used to calculate the probability of a given outcome in a binary model, like the probability of being classified as sick or passing an exam. https://www.simplilearn.com/ice9/free_resources_article_thumb/logistic-regression-example-graph.JPG Sigmoid Probability The probability in the logistic regression is often represented by the Sigmoid function (also called the logistic function or the S-curve): https://www.simplilearn.com/ice9/free_resources_article_thumb/sigmoid-function-machine-learning.JPG In this equation, t represents data values * the number of hours studied and S(t) represents the probability of passing the exam. Assume sigmoid function: https://www.simplilearn.com/ice9/free_resources_article_thumb/sigmoid-probability-machine-learning.JPG g(z) tends toward 1 as z -> infinity , and g(z) tends toward 0 as z -> infinity K-nearest Neighbors (KNN) K-nearest Neighbors algorithm is used to assign a data point to clusters based on similarity measurement. It uses a supervised method for classification. The steps to writing a k-means algorithm are as given below: https://www.simplilearn.com/ice9/free_resources_article_thumb/knn-distribution-graph-machine-learning.JPG Choose the number of k and a distance metric. (k = 5 is common) Find k-nearest neighbors of the sample that you want to classify Assign the class label by majority vote. KNN Classification A new input point is classified in the category such that it has the most number of neighbors from that category. For example: https://www.simplilearn.com/ice9/free_resources_article_thumb/knn-classification-machine-learning.JPG Classify a patient as high risk or low risk. Mark email as spam or ham. Keen on learning about Classification Algorithms in Machine Learning? Click here! Support Vector Machine (SVM) Let us understand Support Vector Machine (SVM) in detail below. SVMs are classification algorithms used to assign data to various classes. They involve detecting hyperplanes which segregate data into classes. SVMs are very versatile and are also capable of performing linear or nonlinear classification, regression, and outlier detection. Once ideal hyperplanes are discovered, new data points can be easily classified. https://www.simplilearn.com/ice9/free_resources_article_thumb/support-vector-machines-graph-machine-learning.JPG The optimization objective is to find “maximum margin hyperplane” that is farthest from the closest points in the two classes (these points are called support vectors). In the given figure, the middle line represents the hyperplane. SVM Example Let’s look at this image below and have an idea about SVM in general. Hyperplanes with larger margins have lower generalization error. The positive and negative hyperplanes are represented by: https://www.simplilearn.com/ice9/free_resources_article_thumb/positive-negative-hyperplanes-machine-learning.JPG Classification of any new input sample xtest : If w0 + wTxtest > 1, the sample xtest is said to be in the class toward the right of the positive hyperplane. If w0 + wTxtest < -1, the sample xtest is said to be in the class toward the left of the negative hyperplane. When you subtract the two equations, you get: https://www.simplilearn.com/ice9/free_resources_article_thumb/equation-subtraction-machine-learning.JPG Length of vector w is (L2 norm length): https://www.simplilearn.com/ice9/free_resources_article_thumb/length-of-vector-machine-learning.JPG You normalize with the length of w to arrive at: https://www.simplilearn.com/ice9/free_resources_article_thumb/normalize-equation-machine-learning.JPG SVM: Hard Margin Classification Given below are some points to understand Hard Margin Classification. The left side of equation SVM-1 given above can be interpreted as the distance between the positive (+ve) and negative (-ve) hyperplanes; in other words, it is the margin that can be maximized. Hence the objective of the function is to maximize with the constraint that the samples are classified correctly, which is represented as : https://www.simplilearn.com/ice9/free_resources_article_thumb/hard-margin-classification-machine-learning.JPG This means that you are minimizing ‖w‖. This also means that all positive samples are on one side of the positive hyperplane and all negative samples are on the other side of the negative hyperplane. This can be written concisely as : https://www.simplilearn.com/ice9/free_resources_article_thumb/hard-margin-classification-formula.JPG Minimizing ‖w‖ is the same as minimizing. This figure is better as it is differentiable even at w = 0. The approach listed above is called “hard margin linear SVM classifier.” SVM: Soft Margin Classification Given below are some points to understand Soft Margin Classification. To allow for linear constraints to be relaxed for nonlinearly separable data, a slack variable is introduced. (i) measures how much ith instance is allowed to violate the margin. The slack variable is simply added to the linear constraints. https://www.simplilearn.com/ice9/free_resources_article_thumb/soft-margin-calculation-machine-learning.JPG Subject to the above constraints, the new objective to be minimized becomes: https://www.simplilearn.com/ice9/free_resources_article_thumb/soft-margin-calculation-formula.JPG You have two conflicting objectives now—minimizing slack variable to reduce margin violations and minimizing to increase the margin. The hyperparameter C allows us to define this trade-off. Large values of C correspond to larger error penalties (so smaller margins), whereas smaller values of C allow for higher misclassification errors and larger margins. https://www.simplilearn.com/ice9/free_resources_article_thumb/machine-learning-certification-video-preview.jpg SVM: Regularization The concept of C is the reverse of regularization. Higher C means lower regularization, which increases bias and lowers the variance (causing overfitting). https://www.simplilearn.com/ice9/free_resources_article_thumb/concept-of-c-graph-machine-learning.JPG IRIS Data Set The Iris dataset contains measurements of 150 IRIS flowers from three different species: Setosa Versicolor Viriginica Each row represents one sample. Flower measurements in centimeters are stored as columns. These are called features. IRIS Data Set: SVM Let’s train an SVM model using sci-kit-learn for the Iris dataset: https://www.simplilearn.com/ice9/free_resources_article_thumb/svm-model-graph-machine-learning.JPG Nonlinear SVM Classification There are two ways to solve nonlinear SVMs: by adding polynomial features by adding similarity features Polynomial features can be added to datasets; in some cases, this can create a linearly separable dataset. https://www.simplilearn.com/ice9/free_resources_article_thumb/nonlinear-classification-svm-machine-learning.JPG In the figure on the left, there is only 1 feature x1. This dataset is not linearly separable. If you add x2 = (x1)2 (figure on the right), the data becomes linearly separable. Polynomial Kernel In sci-kit-learn, one can use a Pipeline class for creating polynomial features. Classification results for the Moons dataset are shown in the figure. https://www.simplilearn.com/ice9/free_resources_article_thumb/polynomial-kernel-machine-learning.JPG Polynomial Kernel with Kernel Trick Let us look at the image below and understand Kernel Trick in detail. https://www.simplilearn.com/ice9/free_resources_article_thumb/polynomial-kernel-with-kernel-trick.JPG For large dimensional datasets, adding too many polynomial features can slow down the model. You can apply a kernel trick with the effect of polynomial features without actually adding them. The code is shown (SVC class) below trains an SVM classifier using a 3rd-degree polynomial kernel but with a kernel trick. https://www.simplilearn.com/ice9/free_resources_article_thumb/polynomial-kernel-equation-machine-learning.JPG The hyperparameter coefθ controls the influence of high-degree polynomials. Kernel SVM Let us understand in detail about Kernel SVM. Kernel SVMs are used for classification of nonlinear data. In the chart, nonlinear data is projected into a higher dimensional space via a mapping function where it becomes linearly separable. https://www.simplilearn.com/ice9/free_resources_article_thumb/kernel-svm-machine-learning.JPG In the higher dimension, a linear separating hyperplane can be derived and used for classification. A reverse projection of the higher dimension back to original feature space takes it back to nonlinear shape. As mentioned previously, SVMs can be kernelized to solve nonlinear classification problems. You can create a sample dataset for XOR gate (nonlinear problem) from NumPy. 100 samples will be assigned the class sample 1, and 100 samples will be assigned the class label -1. https://www.simplilearn.com/ice9/free_resources_article_thumb/kernel-svm-graph-machine-learning.JPG As you can see, this data is not linearly separable. https://www.simplilearn.com/ice9/free_resources_article_thumb/kernel-svm-non-separable.JPG You now use the kernel trick to classify XOR dataset created earlier. https://www.simplilearn.com/ice9/free_resources_article_thumb/kernel-svm-xor-machine-learning.JPG Naïve Bayes Classifier What is Naive Bayes Classifier? Have you ever wondered how your mail provider implements spam filtering or how online news channels perform news text classification or even how companies perform sentiment analysis of their audience on social media? All of this and more are done through a machine learning algorithm called Naive Bayes Classifier. Naive Bayes Named after Thomas Bayes from the 1700s who first coined this in the Western literature. Naive Bayes classifier works on the principle of conditional probability as given by the Bayes theorem. Advantages of Naive Bayes Classifier Listed below are six benefits of Naive Bayes Classifier. Very simple and easy to implement Needs less training data Handles both continuous and discrete data Highly scalable with the number of predictors and data points As it is fast, it can be used in real-time predictions Not sensitive to irrelevant features Bayes Theorem We will understand Bayes Theorem in detail from the points mentioned below. According to the Bayes model, the conditional probability P(Y|X) can be calculated as: P(Y|X) = P(X|Y)P(Y) / P(X) This means you have to estimate a very large number of P(X|Y) probabilities for a relatively small vector space X. For example, for a Boolean Y and 30 possible Boolean attributes in the X vector, you will have to estimate 3 billion probabilities P(X|Y). To make it practical, a Naïve Bayes classifier is used, which assumes conditional independence of P(X) to each other, with a given value of Y. This reduces the number of probability estimates to 2*30=60 in the above example. Naïve Bayes Classifier for SMS Spam Detection Consider a labeled SMS database having 5574 messages. It has messages as given below: https://www.simplilearn.com/ice9/free_resources_article_thumb/naive-bayes-spam-machine-learning.JPG Each message is marked as spam or ham in the data set. Let’s train a model with Naïve Bayes algorithm to detect spam from ham. The message lengths and their frequency (in the training dataset) are as shown below: https://www.simplilearn.com/ice9/free_resources_article_thumb/naive-bayes-spam-spam-detection.JPG Analyze the logic you use to train an algorithm to detect spam: Split each message into individual words/tokens (bag of words). Lemmatize the data (each word takes its base form, like “walking” or “walked” is replaced with “walk”). Convert data to vectors using scikit-learn module CountVectorizer. Run TFIDF to remove common words like “is,” “are,” “and.” Now apply scikit-learn module for Naïve Bayes MultinomialNB to get the Spam Detector. This spam detector can then be used to classify a random new message as spam or ham. Next, the accuracy of the spam detector is checked using the Confusion Matrix. For the SMS spam example above, the confusion matrix is shown on the right. Accuracy Rate = Correct / Total = (4827 + 592)/5574 = 97.21% Error Rate = Wrong / Total = (155 + 0)/5574 = 2.78% https://www.simplilearn.com/ice9/free_resources_article_thumb/confusion-matrix-machine-learning.JPG Although confusion Matrix is useful, some more precise metrics are provided by Precision and Recall. https://www.simplilearn.com/ice9/free_resources_article_thumb/precision-recall-matrix-machine-learning.JPG Precision refers to the accuracy of positive predictions. https://www.simplilearn.com/ice9/free_resources_article_thumb/precision-formula-machine-learning.JPG Recall refers to the ratio of positive instances that are correctly detected by the classifier (also known as True positive rate or TPR). https://www.simplilearn.com/ice9/free_resources_article_thumb/recall-formula-machine-learning.JPG Precision/Recall Trade-off To detect age-appropriate videos for kids, you need high precision (low recall) to ensure that only safe videos make the cut (even though a few safe videos may be left out). The high recall is needed (low precision is acceptable) in-store surveillance to catch shoplifters; a few false alarms are acceptable, but all shoplifters must be caught. Learn about Naive Bayes in detail. Click here! Decision Tree Classifier Some aspects of the Decision Tree Classifier mentioned below are. Decision Trees (DT) can be used both for classification and regression. The advantage of decision trees is that they require very little data preparation. They do not require feature scaling or centering at all. They are also the fundamental components of Random Forests, one of the most powerful ML algorithms. Unlike Random Forests and Neural Networks (which do black-box modeling), Decision Trees are white box models, which means that inner workings of these models are clearly understood. In the case of classification, the data is segregated based on a series of questions. Any new data point is assigned to the selected leaf node. https://www.simplilearn.com/ice9/free_resources_article_thumb/decision-tree-classifier-machine-learning.JPG Start at the tree root and split the data on the feature using the decision algorithm, resulting in the largest information gain (IG). This splitting procedure is then repeated in an iterative process at each child node until the leaves are pure. This means that the samples at each node belonging to the same class. In practice, you can set a limit on the depth of the tree to prevent overfitting. The purity is compromised here as the final leaves may still have some impurity. The figure shows the classification of the Iris dataset. https://www.simplilearn.com/ice9/free_resources_article_thumb/decision-tree-classifier-graph.JPG IRIS Decision Tree Let’s build a Decision Tree using scikit-learn for the Iris flower dataset and also visualize it using export_graphviz API. https://www.simplilearn.com/ice9/free_resources_article_thumb/iris-decision-tree-machine-learning.JPG The output of export_graphviz can be converted into png format: https://www.simplilearn.com/ice9/free_resources_article_thumb/iris-decision-tree-output.JPG Sample attribute stands for the number of training instances the node applies to. Value attribute stands for the number of training instances of each class the node applies to. Gini impurity measures the node’s impurity. A node is “pure” (gini=0) if all training instances it applies to belong to the same class. https://www.simplilearn.com/ice9/free_resources_article_thumb/impurity-formula-machine-learning.JPG For example, for Versicolor (green color node), the Gini is 1-(0/54)2 -(49/54)2 -(5/54) 2 ≈ 0.168 https://www.simplilearn.com/ice9/free_resources_article_thumb/iris-decision-tree-sample.JPG Decision Boundaries Let us learn to create decision boundaries below. For the first node (depth 0), the solid line splits the data (Iris-Setosa on left). Gini is 0 for Setosa node, so no further split is possible. The second node (depth 1) splits the data into Versicolor and Virginica. If max_depth were set as 3, a third split would happen (vertical dotted line). https://www.simplilearn.com/ice9/free_resources_article_thumb/decision-tree-boundaries.JPG For a sample with petal length 5 cm and petal width 1.5 cm, the tree traverses to depth 2 left node, so the probability predictions for this sample are 0% for Iris-Setosa (0/54), 90.7% for Iris-Versicolor (49/54), and 9.3% for Iris-Virginica (5/54) CART Training Algorithm Scikit-learn uses Classification and Regression Trees (CART) algorithm to train Decision Trees. CART algorithm: Split the data into two subsets using a single feature k and threshold tk (example, petal length < “2.45 cm”). This is done recursively for each node. k and tk are chosen such that they produce the purest subsets (weighted by their size). The objective is to minimize the cost function as given below: https://www.simplilearn.com/ice9/free_resources_article_thumb/cart-training-algorithm-machine-learning.JPG The algorithm stops executing if one of the following situations occurs: max_depth is reached No further splits are found for each node Other hyperparameters may be used to stop the tree: min_samples_split min_samples_leaf min_weight_fraction_leaf max_leaf_nodes Gini Impurity or Entropy Entropy is one more measure of impurity and can be used in place of Gini. https://www.simplilearn.com/ice9/free_resources_article_thumb/gini-impurity-entrophy.JPG It is a degree of uncertainty, and Information Gain is the reduction that occurs in entropy as one traverses down the tree. Entropy is zero for a DT node when the node contains instances of only one class. Entropy for depth 2 left node in the example given above is: https://www.simplilearn.com/ice9/free_resources_article_thumb/entrophy-for-depth-2.JPG Gini and Entropy both lead to similar trees. DT: Regularization The following figure shows two decision trees on the moons dataset. https://www.simplilearn.com/ice9/free_resources_article_thumb/dt-regularization-machine-learning.JPG The decision tree on the right is restricted by min_samples_leaf = 4. The model on the left is overfitting, while the model on the right generalizes better. Random Forest Classifier Let us have an understanding of Random Forest Classifier below. A random forest can be considered an ensemble of decision trees (Ensemble learning). Random Forest algorithm: Draw a random bootstrap sample of size n (randomly choose n samples from the training set). Grow a decision tree from the bootstrap sample. At each node, randomly select d features. Split the node using the feature that provides the best split according to the objective function, for instance by maximizing the information gain. Repeat the steps 1 to 2 k times. (k is the number of trees you want to create, using a subset of samples) Aggregate the prediction by each tree for a new data point to assign the class label by majority vote (pick the group selected by the most number of trees and assign new data point to that group). Random Forests are opaque, which means it is difficult to visualize their inner workings. https://www.simplilearn.com/ice9/free_resources_article_thumb/random-forest-classifier-graph.JPG However, the advantages outweigh their limitations since you do not have to worry about hyperparameters except k, which stands for the number of decision trees to be created from a subset of samples. RF is quite robust to noise from the individual decision trees. Hence, you need not prune individual decision trees. The larger the number of decision trees, the more accurate the Random Forest prediction is. (This, however, comes with higher computation cost). Key Takeaways Let us quickly run through what we have learned so far in this Classification tutorial. Classification algorithms are supervised learning methods to split data into classes. They can work on Linear Data as well as Nonlinear Data. Logistic Regression can classify data based on weighted parameters and sigmoid conversion to calculate the probability of classes. K-nearest Neighbors (KNN) algorithm uses similar features to classify data. Support Vector Machines (SVMs) classify data by detecting the maximum margin hyperplane between data classes. Naïve Bayes, a simplified Bayes Model, can help classify data using conditional probability models. Decision Trees are powerful classifiers and use tree splitting logic until pure or somewhat pure leaf node classes are attained. Random Forests apply Ensemble Learning to Decision Trees for more accurate classification predictions. Conclusion This completes ‘Classification’ tutorial. In the next tutorial, we will learn 'Unsupervised Learning with Clustering.'
himanshub1007 / Alzhimers Disease Prediction Using Deep Learning# AD-Prediction Convolutional Neural Networks for Alzheimer's Disease Prediction Using Brain MRI Image ## Abstract Alzheimers disease (AD) is characterized by severe memory loss and cognitive impairment. It associates with significant brain structure changes, which can be measured by magnetic resonance imaging (MRI) scan. The observable preclinical structure changes provides an opportunity for AD early detection using image classification tools, like convolutional neural network (CNN). However, currently most AD related studies were limited by sample size. Finding an efficient way to train image classifier on limited data is critical. In our project, we explored different transfer-learning methods based on CNN for AD prediction brain structure MRI image. We find that both pretrained 2D AlexNet with 2D-representation method and simple neural network with pretrained 3D autoencoder improved the prediction performance comparing to a deep CNN trained from scratch. The pretrained 2D AlexNet performed even better (**86%**) than the 3D CNN with autoencoder (**77%**). ## Method #### 1. Data In this project, we used public brain MRI data from **Alzheimers Disease Neuroimaging Initiative (ADNI)** Study. ADNI is an ongoing, multicenter cohort study, started from 2004. It focuses on understanding the diagnostic and predictive value of Alzheimers disease specific biomarkers. The ADNI study has three phases: ADNI1, ADNI-GO, and ADNI2. Both ADNI1 and ADNI2 recruited new AD patients and normal control as research participants. Our data included a total of 686 structure MRI scans from both ADNI1 and ADNI2 phases, with 310 AD cases and 376 normal controls. We randomly derived the total sample into training dataset (n = 519), validation dataset (n = 100), and testing dataset (n = 67). #### 2. Image preprocessing Image preprocessing were conducted using Statistical Parametric Mapping (SPM) software, version 12. The original MRI scans were first skull-stripped and segmented using segmentation algorithm based on 6-tissue probability mapping and then normalized to the International Consortium for Brain Mapping template of European brains using affine registration. Other configuration includes: bias, noise, and global intensity normalization. The standard preprocessing process output 3D image files with an uniform size of 121x145x121. Skull-stripping and normalization ensured the comparability between images by transforming the original brain image into a standard image space, so that same brain substructures can be aligned at same image coordinates for different participants. Diluted or enhanced intensity was used to compensate the structure changes. the In our project, we used both whole brain (including both grey matter and white matter) and grey matter only. #### 3. AlexNet and Transfer Learning Convolutional Neural Networks (CNN) are very similar to ordinary Neural Networks. A CNN consists of an input and an output layer, as well as multiple hidden layers. The hidden layers are either convolutional, pooling or fully connected. ConvNet architectures make the explicit assumption that the inputs are images, which allows us to encode certain properties into the architecture. These then make the forward function more efficient to implement and vastly reduce the amount of parameters in the network. #### 3.1. AlexNet The net contains eight layers with weights; the first five are convolutional and the remaining three are fully connected. The overall architecture is shown in Figure 1. The output of the last fully-connected layer is fed to a 1000-way softmax which produces a distribution over the 1000 class labels. AlexNet maximizes the multinomial logistic regression objective, which is equivalent to maximizing the average across training cases of the log-probability of the correct label under the prediction distribution. The kernels of the second, fourth, and fifth convolutional layers are connected only to those kernel maps in the previous layer which reside on the same GPU (as shown in Figure1). The kernels of the third convolutional layer are connected to all kernel maps in the second layer. The neurons in the fully connected layers are connected to all neurons in the previous layer. Response-normalization layers follow the first and second convolutional layers. Max-pooling layers follow both response-normalization layers as well as the fifth convolutional layer. The ReLU non-linearity is applied to the output of every convolutional and fully-connected layer.  The first convolutional layer filters the 224x224x3 input image with 96 kernels of size 11x11x3 with a stride of 4 pixels (this is the distance between the receptive field centers of neighboring neurons in a kernel map). The second convolutional layer takes as input the (response-normalized and pooled) output of the first convolutional layer and filters it with 256 kernels of size 5x5x48. The third, fourth, and fifth convolutional layers are connected to one another without any intervening pooling or normalization layers. The third convolutional layer has 384 kernels of size 3x3x256 connected to the (normalized, pooled) outputs of the second convolutional layer. The fourth convolutional layer has 384 kernels of size 3x3x192 , and the fifth convolutional layer has 256 kernels of size 3x3x192. The fully-connected layers have 4096 neurons each. #### 3.2. Transfer Learning Training an entire Convolutional Network from scratch (with random initialization) is impractical[14] because it is relatively rare to have a dataset of sufficient size. An alternative is to pretrain a Conv-Net on a very large dataset (e.g. ImageNet), and then use the ConvNet either as an initialization or a fixed feature extractor for the task of interest. Typically, there are three major transfer learning scenarios: **ConvNet as fixed feature extractor:** We can take a ConvNet pretrained on ImageNet, and remove the last fully-connected layer, then treat the rest structure as a fixed feature extractor for the target dataset. In AlexNet, this would be a 4096-D vector. Usually, we call these features as CNN codes. Once we get these features, we can train a linear classifier (e.g. linear SVM or Softmax classifier) for our target dataset. **Fine-tuning the ConvNet:** Another idea is not only replace the last fully-connected layer in the classifier, but to also fine-tune the parameters of the pretrained network. Due to overfitting concerns, we can only fine-tune some higher-level part of the network. This suggestion is motivated by the observation that earlier features in a ConvNet contains more generic features (e.g. edge detectors or color blob detectors) that can be useful for many kind of tasks. But the later layer of the network becomes progressively more specific to the details of the classes contained in the original dataset. **Pretrained models:** The released pretrained model is usually the final ConvNet checkpoint. So it is common to see people use the network for fine-tuning. #### 4. 3D Autoencoder and Convolutional Neural Network We take a two-stage approach where we first train a 3D sparse autoencoder to learn filters for convolution operations, and then build a convolutional neural network whose first layer uses the filters learned with the autoencoder.  #### 4.1. Sparse Autoencoder An autoencoder is a 3-layer neural network that is used to extract features from an input such as an image. Sparse representations can provide a simple interpretation of the input data in terms of a small number of \parts by extracting the structure hidden in the data. The autoencoder has an input layer, a hidden layer and an output layer, and the input and output layers have same number of units, while the hidden layer contains more units for a sparse and overcomplete representation. The encoder function maps input x to representation h, and the decoder function maps the representation h to the output x. In our problem, we extract 3D patches from scans as the input to the network. The decoder function aims to reconstruct the input form the hidden representation h. #### 4.2. 3D Convolutional Neural Network Training the 3D convolutional neural network(CNN) is the second stage. The CNN we use in this project has one convolutional layer, one pooling layer, two linear layers, and finally a log softmax layer. After training the sparse autoencoder, we take the weights and biases of the encoder from trained model, and use them a 3D filter of a 3D convolutional layer of the 1-layer convolutional neural network. Figure 2 shows the architecture of the network. #### 5. Tools In this project, we used Nibabel for MRI image processing and PyTorch Neural Networks implementation.
automl / DeepCAVEAn interactive framework to visualize and analyze your AutoML process in real-time.
kboone / AvocadoPhotometric Classification of Astronomical Transients and Variables With Biased Spectroscopic Samples
Aryia-Behroziuan / NeuronsAn ANN is a model based on a collection of connected units or nodes called "artificial neurons", which loosely model the neurons in a biological brain. Each connection, like the synapses in a biological brain, can transmit information, a "signal", from one artificial neuron to another. An artificial neuron that receives a signal can process it and then signal additional artificial neurons connected to it. In common ANN implementations, the signal at a connection between artificial neurons is a real number, and the output of each artificial neuron is computed by some non-linear function of the sum of its inputs. The connections between artificial neurons are called "edges". Artificial neurons and edges typically have a weight that adjusts as learning proceeds. The weight increases or decreases the strength of the signal at a connection. Artificial neurons may have a threshold such that the signal is only sent if the aggregate signal crosses that threshold. Typically, artificial neurons are aggregated into layers. Different layers may perform different kinds of transformations on their inputs. Signals travel from the first layer (the input layer) to the last layer (the output layer), possibly after traversing the layers multiple times. The original goal of the ANN approach was to solve problems in the same way that a human brain would. However, over time, attention moved to performing specific tasks, leading to deviations from biology. Artificial neural networks have been used on a variety of tasks, including computer vision, speech recognition, machine translation, social network filtering, playing board and video games and medical diagnosis. Deep learning consists of multiple hidden layers in an artificial neural network. This approach tries to model the way the human brain processes light and sound into vision and hearing. Some successful applications of deep learning are computer vision and speech recognition.[68] Decision trees Main article: Decision tree learning Decision tree learning uses a decision tree as a predictive model to go from observations about an item (represented in the branches) to conclusions about the item's target value (represented in the leaves). It is one of the predictive modeling approaches used in statistics, data mining, and machine learning. Tree models where the target variable can take a discrete set of values are called classification trees; in these tree structures, leaves represent class labels and branches represent conjunctions of features that lead to those class labels. Decision trees where the target variable can take continuous values (typically real numbers) are called regression trees. In decision analysis, a decision tree can be used to visually and explicitly represent decisions and decision making. In data mining, a decision tree describes data, but the resulting classification tree can be an input for decision making. Support vector machines Main article: Support vector machines Support vector machines (SVMs), also known as support vector networks, are a set of related supervised learning methods used for classification and regression. Given a set of training examples, each marked as belonging to one of two categories, an SVM training algorithm builds a model that predicts whether a new example falls into one category or the other.[69] An SVM training algorithm is a non-probabilistic, binary, linear classifier, although methods such as Platt scaling exist to use SVM in a probabilistic classification setting. In addition to performing linear classification, SVMs can efficiently perform a non-linear classification using what is called the kernel trick, implicitly mapping their inputs into high-dimensional feature spaces. Illustration of linear regression on a data set. Regression analysis Main article: Regression analysis Regression analysis encompasses a large variety of statistical methods to estimate the relationship between input variables and their associated features. Its most common form is linear regression, where a single line is drawn to best fit the given data according to a mathematical criterion such as ordinary least squares. The latter is often extended by regularization (mathematics) methods to mitigate overfitting and bias, as in ridge regression. When dealing with non-linear problems, go-to models include polynomial regression (for example, used for trendline fitting in Microsoft Excel[70]), logistic regression (often used in statistical classification) or even kernel regression, which introduces non-linearity by taking advantage of the kernel trick to implicitly map input variables to higher-dimensional space. Bayesian networks Main article: Bayesian network A simple Bayesian network. Rain influences whether the sprinkler is activated, and both rain and the sprinkler influence whether the grass is wet. A Bayesian network, belief network, or directed acyclic graphical model is a probabilistic graphical model that represents a set of random variables and their conditional independence with a directed acyclic graph (DAG). For example, a Bayesian network could represent the probabilistic relationships between diseases and symptoms. Given symptoms, the network can be used to compute the probabilities of the presence of various diseases. Efficient algorithms exist that perform inference and learning. Bayesian networks that model sequences of variables, like speech signals or protein sequences, are called dynamic Bayesian networks. Generalizations of Bayesian networks that can represent and solve decision problems under uncertainty are called influence diagrams. Genetic algorithms Main article: Genetic algorithm A genetic algorithm (GA) is a search algorithm and heuristic technique that mimics the process of natural selection, using methods such as mutation and crossover to generate new genotypes in the hope of finding good solutions to a given problem. In machine learning, genetic algorithms were used in the 1980s and 1990s.[71][72] Conversely, machine learning techniques have been used to improve the performance of genetic and evolutionary algorithms.[73] Training models Usually, machine learning models require a lot of data in order for them to perform well. Usually, when training a machine learning model, one needs to collect a large, representative sample of data from a training set. Data from the training set can be as varied as a corpus of text, a collection of images, and data collected from individual users of a service. Overfitting is something to watch out for when training a machine learning model. Federated learning Main article: Federated learning Federated learning is an adapted form of distributed artificial intelligence to training machine learning models that decentralizes the training process, allowing for users' privacy to be maintained by not needing to send their data to a centralized server. This also increases efficiency by decentralizing the training process to many devices. For example, Gboard uses federated machine learning to train search query prediction models on users' mobile phones without having to send individual searches back to Google.[74] Applications There are many applications for machine learning, including: Agriculture Anatomy Adaptive websites Affective computing Banking Bioinformatics Brain–machine interfaces Cheminformatics Citizen science Computer networks Computer vision Credit-card fraud detection Data quality DNA sequence classification Economics Financial market analysis[75] General game playing Handwriting recognition Information retrieval Insurance Internet fraud detection Linguistics Machine learning control Machine perception Machine translation Marketing Medical diagnosis Natural language processing Natural language understanding Online advertising Optimization Recommender systems Robot locomotion Search engines Sentiment analysis Sequence mining Software engineering Speech recognition Structural health monitoring Syntactic pattern recognition Telecommunication Theorem proving Time series forecasting User behavior analytics In 2006, the media-services provider Netflix held the first "Netflix Prize" competition to find a program to better predict user preferences and improve the accuracy of its existing Cinematch movie recommendation algorithm by at least 10%. A joint team made up of researchers from AT&T Labs-Research in collaboration with the teams Big Chaos and Pragmatic Theory built an ensemble model to win the Grand Prize in 2009 for $1 million.[76] Shortly after the prize was awarded, Netflix realized that viewers' ratings were not the best indicators of their viewing patterns ("everything is a recommendation") and they changed their recommendation engine accordingly.[77] In 2010 The Wall Street Journal wrote about the firm Rebellion Research and their use of machine learning to predict the financial crisis.[78] In 2012, co-founder of Sun Microsystems, Vinod Khosla, predicted that 80% of medical doctors' jobs would be lost in the next two decades to automated machine learning medical diagnostic software.[79] In 2014, it was reported that a machine learning algorithm had been applied in the field of art history to study fine art paintings and that it may have revealed previously unrecognized influences among artists.[80] In 2019 Springer Nature published the first research book created using machine learning.[81] Limitations Although machine learning has been transformative in some fields, machine-learning programs often fail to deliver expected results.[82][83][84] Reasons for this are numerous: lack of (suitable) data, lack of access to the data, data bias, privacy problems, badly chosen tasks and algorithms, wrong tools and people, lack of resources, and evaluation problems.[85] In 2018, a self-driving car from Uber failed to detect a pedestrian, who was killed after a collision.[86] Attempts to use machine learning in healthcare with the IBM Watson system failed to deliver even after years of time and billions of dollars invested.[87][88] Bias Main article: Algorithmic bias Machine learning approaches in particular can suffer from different data biases. A machine learning system trained on current customers only may not be able to predict the needs of new customer groups that are not represented in the training data. When trained on man-made data, machine learning is likely to pick up the same constitutional and unconscious biases already present in society.[89] Language models learned from data have been shown to contain human-like biases.[90][91] Machine learning systems used for criminal risk assessment have been found to be biased against black people.[92][93] In 2015, Google photos would often tag black people as gorillas,[94] and in 2018 this still was not well resolved, but Google reportedly was still using the workaround to remove all gorillas from the training data, and thus was not able to recognize real gorillas at all.[95] Similar issues with recognizing non-white people have been found in many other systems.[96] In 2016, Microsoft tested a chatbot that learned from Twitter, and it quickly picked up racist and sexist language.[97] Because of such challenges, the effective use of machine learning may take longer to be adopted in other domains.[98] Concern for fairness in machine learning, that is, reducing bias in machine learning and propelling its use for human good is increasingly expressed by artificial intelligence scientists, including Fei-Fei Li, who reminds engineers that "There’s nothing artificial about AI...It’s inspired by people, it’s created by people, and—most importantly—it impacts people. It is a powerful tool we are only just beginning to understand, and that is a profound responsibility.”[99] Model assessments Classification of machine learning models can be validated by accuracy estimation techniques like the holdout method, which splits the data in a training and test set (conventionally 2/3 training set and 1/3 test set designation) and evaluates the performance of the training model on the test set. In comparison, the K-fold-cross-validation method randomly partitions the data into K subsets and then K experiments are performed each respectively considering 1 subset for evaluation and the remaining K-1 subsets for training the model. In addition to the holdout and cross-validation methods, bootstrap, which samples n instances with replacement from the dataset, can be used to assess model accuracy.[100] In addition to overall accuracy, investigators frequently report sensitivity and specificity meaning True Positive Rate (TPR) and True Negative Rate (TNR) respectively. Similarly, investigators sometimes report the false positive rate (FPR) as well as the false negative rate (FNR). However, these rates are ratios that fail to reveal their numerators and denominators. The total operating characteristic (TOC) is an effective method to express a model's diagnostic ability. TOC shows the numerators and denominators of the previously mentioned rates, thus TOC provides more information than the commonly used receiver operating characteristic (ROC) and ROC's associated area under the curve (AUC).[101] Ethics Machine learning poses a host of ethical questions. Systems which are trained on datasets collected with biases may exhibit these biases upon use (algorithmic bias), thus digitizing cultural prejudices.[102] For example, using job hiring data from a firm with racist hiring policies may lead to a machine learning system duplicating the bias by scoring job applicants against similarity to previous successful applicants.[103][104] Responsible collection of data and documentation of algorithmic rules used by a system thus is a critical part of machine learning. Because human languages contain biases, machines trained on language corpora will necessarily also learn these biases.[105][106] Other forms of ethical challenges, not related to personal biases, are more seen in health care. There are concerns among health care professionals that these systems might not be designed in the public's interest but as income-generating machines. This is especially true in the United States where there is a long-standing ethical dilemma of improving health care, but also increasing profits. For example, the algorithms could be designed to provide patients with unnecessary tests or medication in which the algorithm's proprietary owners hold stakes. There is huge potential for machine learning in health care to provide professionals a great tool to diagnose, medicate, and even plan recovery paths for patients, but this will not happen until the personal biases mentioned previously, and these "greed" biases are addressed.[107] Hardware Since the 2010s, advances in both machine learning algorithms and computer hardware have led to more efficient methods for training deep neural networks (a particular narrow subdomain of machine learning) that contain many layers of non-linear hidden units.[108] By 2019, graphic processing units (GPUs), often with AI-specific enhancements, had displaced CPUs as the dominant method of training large-scale commercial cloud AI.[109] OpenAI estimated the hardware compute used in the largest deep learning projects from AlexNet (2012) to AlphaZero (2017), and found a 300,000-fold increase in the amount of compute required, with a doubling-time trendline of 3.4 months.[110][111] Software Software suites containing a variety of machine learning algorithms include the following: Free and open-source so
InhwanBae / NPSNOfficial Code for "Non-Probability Sampling Network for Stochastic Human Trajectory Prediction (CVPR 2022)"
UW-Madison-Lee-Lab / LLM Judge ReportingA simple plug-in framework that corrects bias and computes confidence intervals in reporting LLM-as-a-judge evaluation, and an adaptive algorithm that efficiently allocates calibration samples to reduce uncertainty in estimates.
jonrau1 / AWS DevSecOps FactorySample DevSecOps pipelines (heavily biased on the "Sec") for various stacks and tools using open-source security tools and AWS native services
Aryia-Behroziuan / ReferencesPoole, Mackworth & Goebel 1998, p. 1. Russell & Norvig 2003, p. 55. Definition of AI as the study of intelligent agents: Poole, Mackworth & Goebel (1998), which provides the version that is used in this article. These authors use the term "computational intelligence" as a synonym for artificial intelligence.[1] Russell & Norvig (2003) (who prefer the term "rational agent") and write "The whole-agent view is now widely accepted in the field".[2] Nilsson 1998 Legg & Hutter 2007 Russell & Norvig 2009, p. 2. McCorduck 2004, p. 204 Maloof, Mark. "Artificial Intelligence: An Introduction, p. 37" (PDF). georgetown.edu. Archived (PDF) from the original on 25 August 2018. "How AI Is Getting Groundbreaking Changes In Talent Management And HR Tech". Hackernoon. Archived from the original on 11 September 2019. Retrieved 14 February 2020. Schank, Roger C. (1991). "Where's the AI". AI magazine. Vol. 12 no. 4. p. 38. Russell & Norvig 2009. "AlphaGo – Google DeepMind". Archived from the original on 10 March 2016. Allen, Gregory (April 2020). "Department of Defense Joint AI Center - Understanding AI Technology" (PDF). AI.mil - The official site of the Department of Defense Joint Artificial Intelligence Center. Archived (PDF) from the original on 21 April 2020. Retrieved 25 April 2020. Optimism of early AI: * Herbert Simon quote: Simon 1965, p. 96 quoted in Crevier 1993, p. 109. * Marvin Minsky quote: Minsky 1967, p. 2 quoted in Crevier 1993, p. 109. Boom of the 1980s: rise of expert systems, Fifth Generation Project, Alvey, MCC, SCI: * McCorduck 2004, pp. 426–441 * Crevier 1993, pp. 161–162,197–203, 211, 240 * Russell & Norvig 2003, p. 24 * NRC 1999, pp. 210–211 * Newquist 1994, pp. 235–248 First AI Winter, Mansfield Amendment, Lighthill report * Crevier 1993, pp. 115–117 * Russell & Norvig 2003, p. 22 * NRC 1999, pp. 212–213 * Howe 1994 * Newquist 1994, pp. 189–201 Second AI winter: * McCorduck 2004, pp. 430–435 * Crevier 1993, pp. 209–210 * NRC 1999, pp. 214–216 * Newquist 1994, pp. 301–318 AI becomes hugely successful in the early 21st century * Clark 2015 Pamela McCorduck (2004, p. 424) writes of "the rough shattering of AI in subfields—vision, natural language, decision theory, genetic algorithms, robotics ... and these with own sub-subfield—that would hardly have anything to say to each other." This list of intelligent traits is based on the topics covered by the major AI textbooks, including: * Russell & Norvig 2003 * Luger & Stubblefield 2004 * Poole, Mackworth & Goebel 1998 * Nilsson 1998 Kolata 1982. Maker 2006. Biological intelligence vs. intelligence in general: Russell & Norvig 2003, pp. 2–3, who make the analogy with aeronautical engineering. McCorduck 2004, pp. 100–101, who writes that there are "two major branches of artificial intelligence: one aimed at producing intelligent behavior regardless of how it was accomplished, and the other aimed at modeling intelligent processes found in nature, particularly human ones." Kolata 1982, a paper in Science, which describes McCarthy's indifference to biological models. Kolata quotes McCarthy as writing: "This is AI, so we don't care if it's psychologically real".[19] McCarthy recently reiterated his position at the AI@50 conference where he said "Artificial intelligence is not, by definition, simulation of human intelligence".[20]. Neats vs. scruffies: * McCorduck 2004, pp. 421–424, 486–489 * Crevier 1993, p. 168 * Nilsson 1983, pp. 10–11 Symbolic vs. sub-symbolic AI: * Nilsson (1998, p. 7), who uses the term "sub-symbolic". General intelligence (strong AI) is discussed in popular introductions to AI: * Kurzweil 1999 and Kurzweil 2005 See the Dartmouth proposal, under Philosophy, below. McCorduck 2004, p. 34. McCorduck 2004, p. xviii. McCorduck 2004, p. 3. McCorduck 2004, pp. 340–400. This is a central idea of Pamela McCorduck's Machines Who Think. She writes: "I like to think of artificial intelligence as the scientific apotheosis of a venerable cultural tradition."[26] "Artificial intelligence in one form or another is an idea that has pervaded Western intellectual history, a dream in urgent need of being realized."[27] "Our history is full of attempts—nutty, eerie, comical, earnest, legendary and real—to make artificial intelligences, to reproduce what is the essential us—bypassing the ordinary means. Back and forth between myth and reality, our imaginations supplying what our workshops couldn't, we have engaged for a long time in this odd form of self-reproduction."[28] She traces the desire back to its Hellenistic roots and calls it the urge to "forge the Gods."[29] "Stephen Hawking believes AI could be mankind's last accomplishment". BetaNews. 21 October 2016. Archived from the original on 28 August 2017. Lombardo P, Boehm I, Nairz K (2020). "RadioComics – Santa Claus and the future of radiology". Eur J Radiol. 122 (1): 108771. doi:10.1016/j.ejrad.2019.108771. PMID 31835078. Ford, Martin; Colvin, Geoff (6 September 2015). "Will robots create more jobs than they destroy?". The Guardian. Archived from the original on 16 June 2018. Retrieved 13 January 2018. AI applications widely used behind the scenes: * Russell & Norvig 2003, p. 28 * Kurzweil 2005, p. 265 * NRC 1999, pp. 216–222 * Newquist 1994, pp. 189–201 AI in myth: * McCorduck 2004, pp. 4–5 * Russell & Norvig 2003, p. 939 AI in early science fiction. * McCorduck 2004, pp. 17–25 Formal reasoning: * Berlinski, David (2000). The Advent of the Algorithm. Harcourt Books. ISBN 978-0-15-601391-8. OCLC 46890682. Archived from the original on 26 July 2020. Retrieved 22 August 2020. Turing, Alan (1948), "Machine Intelligence", in Copeland, B. Jack (ed.), The Essential Turing: The ideas that gave birth to the computer age, Oxford: Oxford University Press, p. 412, ISBN 978-0-19-825080-7 Russell & Norvig 2009, p. 16. Dartmouth conference: * McCorduck 2004, pp. 111–136 * Crevier 1993, pp. 47–49, who writes "the conference is generally recognized as the official birthdate of the new science." * Russell & Norvig 2003, p. 17, who call the conference "the birth of artificial intelligence." * NRC 1999, pp. 200–201 McCarthy, John (1988). "Review of The Question of Artificial Intelligence". Annals of the History of Computing. 10 (3): 224–229., collected in McCarthy, John (1996). "10. Review of The Question of Artificial Intelligence". Defending AI Research: A Collection of Essays and Reviews. CSLI., p. 73, "[O]ne of the reasons for inventing the term "artificial intelligence" was to escape association with "cybernetics". Its concentration on analog feedback seemed misguided, and I wished to avoid having either to accept Norbert (not Robert) Wiener as a guru or having to argue with him." Hegemony of the Dartmouth conference attendees: * Russell & Norvig 2003, p. 17, who write "for the next 20 years the field would be dominated by these people and their students." * McCorduck 2004, pp. 129–130 Russell & Norvig 2003, p. 18. Schaeffer J. (2009) Didn't Samuel Solve That Game?. In: One Jump Ahead. Springer, Boston, MA Samuel, A. L. (July 1959). "Some Studies in Machine Learning Using the Game of Checkers". IBM Journal of Research and Development. 3 (3): 210–229. CiteSeerX 10.1.1.368.2254. doi:10.1147/rd.33.0210. "Golden years" of AI (successful symbolic reasoning programs 1956–1973): * McCorduck 2004, pp. 243–252 * Crevier 1993, pp. 52–107 * Moravec 1988, p. 9 * Russell & Norvig 2003, pp. 18–21 The programs described are Arthur Samuel's checkers program for the IBM 701, Daniel Bobrow's STUDENT, Newell and Simon's Logic Theorist and Terry Winograd's SHRDLU. DARPA pours money into undirected pure research into AI during the 1960s: * McCorduck 2004, p. 131 * Crevier 1993, pp. 51, 64–65 * NRC 1999, pp. 204–205 AI in England: * Howe 1994 Lighthill 1973. Expert systems: * ACM 1998, I.2.1 * Russell & Norvig 2003, pp. 22–24 * Luger & Stubblefield 2004, pp. 227–331 * Nilsson 1998, chpt. 17.4 * McCorduck 2004, pp. 327–335, 434–435 * Crevier 1993, pp. 145–62, 197–203 * Newquist 1994, pp. 155–183 Mead, Carver A.; Ismail, Mohammed (8 May 1989). Analog VLSI Implementation of Neural Systems (PDF). The Kluwer International Series in Engineering and Computer Science. 80. Norwell, MA: Kluwer Academic Publishers. doi:10.1007/978-1-4613-1639-8. ISBN 978-1-4613-1639-8. Archived from the original (PDF) on 6 November 2019. Retrieved 24 January 2020. Formal methods are now preferred ("Victory of the neats"): * Russell & Norvig 2003, pp. 25–26 * McCorduck 2004, pp. 486–487 McCorduck 2004, pp. 480–483. Markoff 2011. "Ask the AI experts: What's driving today's progress in AI?". McKinsey & Company. Archived from the original on 13 April 2018. Retrieved 13 April 2018. Administrator. "Kinect's AI breakthrough explained". i-programmer.info. Archived from the original on 1 February 2016. Rowinski, Dan (15 January 2013). "Virtual Personal Assistants & The Future Of Your Smartphone [Infographic]". ReadWrite. Archived from the original on 22 December 2015. "Artificial intelligence: Google's AlphaGo beats Go master Lee Se-dol". BBC News. 12 March 2016. Archived from the original on 26 August 2016. Retrieved 1 October 2016. Metz, Cade (27 May 2017). "After Win in China, AlphaGo's Designers Explore New AI". Wired. Archived from the original on 2 June 2017. "World's Go Player Ratings". May 2017. Archived from the original on 1 April 2017. "柯洁迎19岁生日 雄踞人类世界排名第一已两年" (in Chinese). May 2017. Archived from the original on 11 August 2017. Clark, Jack (8 December 2015). "Why 2015 Was a Breakthrough Year in Artificial Intelligence". Bloomberg News. Archived from the original on 23 November 2016. Retrieved 23 November 2016. After a half-decade of quiet breakthroughs in artificial intelligence, 2015 has been a landmark year. Computers are smarter and learning faster than ever. "Reshaping Business With Artificial Intelligence". MIT Sloan Management Review. Archived from the original on 19 May 2018. Retrieved 2 May 2018. Lorica, Ben (18 December 2017). "The state of AI adoption". O'Reilly Media. Archived from the original on 2 May 2018. Retrieved 2 May 2018. Allen, Gregory (6 February 2019). "Understanding China's AI Strategy". Center for a New American Security. Archived from the original on 17 March 2019. "Review | How two AI superpowers – the U.S. and China – battle for supremacy in the field". Washington Post. 2 November 2018. Archived from the original on 4 November 2018. Retrieved 4 November 2018. at 10:11, Alistair Dabbs 22 Feb 2019. "Artificial Intelligence: You know it isn't real, yeah?". www.theregister.co.uk. Archived from the original on 21 May 2020. Retrieved 22 August 2020. "Stop Calling it Artificial Intelligence". Archived from the original on 2 December 2019. Retrieved 1 December 2019. "AI isn't taking over the world – it doesn't exist yet". GBG Global website. Archived from the original on 11 August 2020. Retrieved 22 August 2020. Kaplan, Andreas; Haenlein, Michael (1 January 2019). "Siri, Siri, in my hand: Who's the fairest in the land? On the interpretations, illustrations, and implications of artificial intelligence". Business Horizons. 62 (1): 15–25. doi:10.1016/j.bushor.2018.08.004. Domingos 2015, Chapter 5. Domingos 2015, Chapter 7. Lindenbaum, M., Markovitch, S., & Rusakov, D. (2004). Selective sampling for nearest neighbor classifiers. Machine learning, 54(2), 125–152. Domingos 2015, Chapter 1. Intractability and efficiency and the combinatorial explosion: * Russell & Norvig 2003, pp. 9, 21–22 Domingos 2015, Chapter 2, Chapter 3. Hart, P. E.; Nilsson, N. J.; Raphael, B. (1972). "Correction to "A Formal Basis for the Heuristic Determination of Minimum Cost Paths"". SIGART Newsletter (37): 28–29. doi:10.1145/1056777.1056779. S2CID 6386648. Domingos 2015, Chapter 2, Chapter 4, Chapter 6. "Can neural network computers learn from experience, and if so, could they ever become what we would call 'smart'?". Scientific American. 2018. Archived from the original on 25 March 2018. Retrieved 24 March 2018. Domingos 2015, Chapter 6, Chapter 7. Domingos 2015, p. 286. "Single pixel change fools AI programs". BBC News. 3 November 2017. Archived from the original on 22 March 2018. Retrieved 12 March 2018. "AI Has a Hallucination Problem That's Proving Tough to Fix". WIRED. 2018. Archived from the original on 12 March 2018. Retrieved 12 March 2018. Matti, D.; Ekenel, H. K.; Thiran, J. P. (2017). Combining LiDAR space clustering and convolutional neural networks for pedestrian detection. 2017 14th IEEE International Conference on Advanced Video and Signal Based Surveillance (AVSS). pp. 1–6. arXiv:1710.06160. doi:10.1109/AVSS.2017.8078512. ISBN 978-1-5386-2939-0. S2CID 2401976. Ferguson, Sarah; Luders, Brandon; Grande, Robert C.; How, Jonathan P. (2015). Real-Time Predictive Modeling and Robust Avoidance of Pedestrians with Uncertain, Changing Intentions. Algorithmic Foundations of Robotics XI. Springer Tracts in Advanced Robotics. 107. Springer, Cham. pp. 161–177. arXiv:1405.5581. doi:10.1007/978-3-319-16595-0_10. ISBN 978-3-319-16594-3. S2CID 8681101. "Cultivating Common Sense | DiscoverMagazine.com". Discover Magazine. 2017. Archived from the original on 25 March 2018. Retrieved 24 March 2018. Davis, Ernest; Marcus, Gary (24 August 2015). "Commonsense reasoning and commonsense knowledge in artificial intelligence". Communications of the ACM. 58 (9): 92–103. doi:10.1145/2701413. S2CID 13583137. Archived from the original on 22 August 2020. Retrieved 6 April 2020. Winograd, Terry (January 1972). "Understanding natural language". Cognitive Psychology. 3 (1): 1–191. doi:10.1016/0010-0285(72)90002-3. "Don't worry: Autonomous cars aren't coming tomorrow (or next year)". Autoweek. 2016. Archived from the original on 25 March 2018. Retrieved 24 March 2018. Knight, Will (2017). "Boston may be famous for bad drivers, but it's the testing ground for a smarter self-driving car". MIT Technology Review. Archived from the original on 22 August 2020. Retrieved 27 March 2018. Prakken, Henry (31 August 2017). "On the problem of making autonomous vehicles conform to traffic law". Artificial Intelligence and Law. 25 (3): 341–363. doi:10.1007/s10506-017-9210-0. Lieto, Antonio (May 2018). "The knowledge level in cognitive architectures: Current limitations and possible developments". Cognitive Systems Research. 48: 39–55. doi:10.1016/j.cogsys.2017.05.001. hdl:2318/1665207. S2CID 206868967. Problem solving, puzzle solving, game playing and deduction: * Russell & Norvig 2003, chpt. 3–9, * Poole, Mackworth & Goebel 1998, chpt. 2,3,7,9, * Luger & Stubblefield 2004, chpt. 3,4,6,8, * Nilsson 1998, chpt. 7–12 Uncertain reasoning: * Russell & Norvig 2003, pp. 452–644, * Poole, Mackworth & Goebel 1998, pp. 345–395, * Luger & Stubblefield 2004, pp. 333–381, * Nilsson 1998, chpt. 19 Psychological evidence of sub-symbolic reasoning: * Wason & Shapiro (1966) showed that people do poorly on completely abstract problems, but if the problem is restated to allow the use of intuitive social intelligence, performance dramatically improves. (See Wason selection task) * Kahneman, Slovic & Tversky (1982) have shown that people are terrible at elementary problems that involve uncertain reasoning. (See list of cognitive biases for several examples). * Lakoff & Núñez (2000) have controversially argued that even our skills at mathematics depend on knowledge and skills that come from "the body", i.e. sensorimotor and perceptual skills. (See Where Mathematics Comes From) Knowledge representation: * ACM 1998, I.2.4, * Russell & Norvig 2003, pp. 320–363, * Poole, Mackworth & Goebel 1998, pp. 23–46, 69–81, 169–196, 235–277, 281–298, 319–345, * Luger & Stubblefield 2004, pp. 227–243, * Nilsson 1998, chpt. 18 Knowledge engineering: * Russell & Norvig 2003, pp. 260–266, * Poole, Mackworth & Goebel 1998, pp. 199–233, * Nilsson 1998, chpt. ≈17.1–17.4 Representing categories and relations: Semantic networks, description logics, inheritance (including frames and scripts): * Russell & Norvig 2003, pp. 349–354, * Poole, Mackworth & Goebel 1998, pp. 174–177, * Luger & Stubblefield 2004, pp. 248–258, * Nilsson 1998, chpt. 18.3 Representing events and time:Situation calculus, event calculus, fluent calculus (including solving the frame problem): * Russell & Norvig 2003, pp. 328–341, * Poole, Mackworth & Goebel 1998, pp. 281–298, * Nilsson 1998, chpt. 18.2 Causal calculus: * Poole, Mackworth & Goebel 1998, pp. 335–337 Representing knowledge about knowledge: Belief calculus, modal logics: * Russell & Norvig 2003, pp. 341–344, * Poole, Mackworth & Goebel 1998, pp. 275–277 Sikos, Leslie F. (June 2017). Description Logics in Multimedia Reasoning. Cham: Springer. doi:10.1007/978-3-319-54066-5. ISBN 978-3-319-54066-5. S2CID 3180114. Archived from the original on 29 August 2017. Ontology: * Russell & Norvig 2003, pp. 320–328 Smoliar, Stephen W.; Zhang, HongJiang (1994). "Content based video indexing and retrieval". IEEE Multimedia. 1 (2): 62–72. doi:10.1109/93.311653. S2CID 32710913. Neumann, Bernd; Möller, Ralf (January 2008). "On scene interpretation with description logics". Image and Vision Computing. 26 (1): 82–101. doi:10.1016/j.imavis.2007.08.013. Kuperman, G. J.; Reichley, R. M.; Bailey, T. C. (1 July 2006). "Using Commercial Knowledge Bases for Clinical Decision Support: Opportunities, Hurdles, and Recommendations". Journal of the American Medical Informatics Association. 13 (4): 369–371. doi:10.1197/jamia.M2055. PMC 1513681. PMID 16622160. MCGARRY, KEN (1 December 2005). "A survey of interestingness measures for knowledge discovery". The Knowledge Engineering Review. 20 (1): 39–61. doi:10.1017/S0269888905000408. S2CID 14987656. Bertini, M; Del Bimbo, A; Torniai, C (2006). "Automatic annotation and semantic retrieval of video sequences using multimedia ontologies". MM '06 Proceedings of the 14th ACM international conference on Multimedia. 14th ACM international conference on Multimedia. Santa Barbara: ACM. pp. 679–682. Qualification problem: * McCarthy & Hayes 1969 * Russell & Norvig 2003[page needed] While McCarthy was primarily concerned with issues in the logical representation of actions, Russell & Norvig 2003 apply the term to the more general issue of default reasoning in the vast network of assumptions underlying all our commonsense knowledge. Default reasoning and default logic, non-monotonic logics, circumscription, closed world assumption, abduction (Poole et al. places abduction under "default reasoning". Luger et al. places this under "uncertain reasoning"): * Russell & Norvig 2003, pp. 354–360, * Poole, Mackworth & Goebel 1998, pp. 248–256, 323–335, * Luger & Stubblefield 2004, pp. 335–363, * Nilsson 1998, ~18.3.3 Breadth of commonsense knowledge: * Russell & Norvig 2003, p. 21, * Crevier 1993, pp. 113–114, * Moravec 1988, p. 13, * Lenat & Guha 1989 (Introduction) Dreyfus & Dreyfus 1986. Gladwell 2005. Expert knowledge as embodied intuition: * Dreyfus & Dreyfus 1986 (Hubert Dreyfus is a philosopher and critic of AI who was among the first to argue that most useful human knowledge was encoded sub-symbolically. See Dreyfus' critique of AI) * Gladwell 2005 (Gladwell's Blink is a popular introduction to sub-symbolic reasoning and knowledge.) * Hawkins & Blakeslee 2005 (Hawkins argues that sub-symbolic knowledge should be the primary focus of AI research.) Planning: * ACM 1998, ~I.2.8, * Russell & Norvig 2003, pp. 375–459, * Poole, Mackworth & Goebel 1998, pp. 281–316, * Luger & Stubblefield 2004, pp. 314–329, * Nilsson 1998, chpt. 10.1–2, 22 Information value theory: * Russell & Norvig 2003, pp. 600–604 Classical planning: * Russell & Norvig 2003, pp. 375–430, * Poole, Mackworth & Goebel 1998, pp. 281–315, * Luger & Stubblefield 2004, pp. 314–329, * Nilsson 1998, chpt. 10.1–2, 22 Planning and acting in non-deterministic domains: conditional planning, execution monitoring, replanning and continuous planning: * Russell & Norvig 2003, pp. 430–449 Multi-agent planning and emergent behavior: * Russell & Norvig 2003, pp. 449–455 Turing 1950. Solomonoff 1956. Alan Turing discussed the centrality of learning as early as 1950, in his classic paper "Computing Machinery and Intelligence".[120] In 1956, at the original Dartmouth AI summer conference, Ray Solomonoff wrote a report on unsupervised probabilistic machine learning: "An Inductive Inference Machine".[121] This is a form of Tom Mitchell's widely quoted definition of machine learning: "A computer program is set to learn from an experience E with respect to some task T and some performance measure P if its performance on T as measured by P improves with experience E." Learning: * ACM 1998, I.2.6, * Russell & Norvig 2003, pp. 649–788, * Poole, Mackworth & Goebel 1998, pp. 397–438, * Luger & Stubblefield 2004, pp. 385–542, * Nilsson 1998, chpt. 3.3, 10.3, 17.5, 20 Jordan, M. I.; Mitchell, T. M. (16 July 2015). "Machine learning: Trends, perspectives, and prospects". Science. 349 (6245): 255–260. Bibcode:2015Sci...349..255J. doi:10.1126/science.aaa8415. PMID 26185243. S2CID 677218. Reinforcement learning: * Russell & Norvig 2003, pp. 763–788 * Luger & Stubblefield 2004, pp. 442–449 Natural language processing: * ACM 1998, I.2.7 * Russell & Norvig 2003, pp. 790–831 * Poole, Mackworth & Goebel 1998, pp. 91–104 * Luger & Stubblefield 2004, pp. 591–632 "Versatile question answering systems: seeing in synthesis" Archived 1 February 2016 at the Wayback Machine, Mittal et al., IJIIDS, 5(2), 119–142, 2011 Applications of natural language processing, including information retrieval (i.e. text mining) and machine translation: * Russell & Norvig 2003, pp. 840–857, * Luger & Stubblefield 2004, pp. 623–630 Cambria, Erik; White, Bebo (May 2014). "Jumping NLP Curves: A Review of Natural Language Processing Research [Review Article]". IEEE Computational Intelligence Magazine. 9 (2): 48–57. doi:10.1109/MCI.2014.2307227. S2CID 206451986. Vincent, James (7 November 2019). "OpenAI has published the text-generating AI it said was too dangerous to share". The Verge. Archived from the original on 11 June 2020. Retrieved 11 June 2020. Machine perception: * Russell & Norvig 2003, pp. 537–581, 863–898 * Nilsson 1998, ~chpt. 6 Speech recognition: * ACM 1998, ~I.2.7 * Russell & Norvig 2003, pp. 568–578 Object recognition: * Russell & Norvig 2003, pp. 885–892 Computer vision: * ACM 1998, I.2.10 * Russell & Norvig 2003, pp. 863–898 * Nilsson 1998, chpt. 6 Robotics: * ACM 1998, I.2.9, * Russell & Norvig 2003, pp. 901–942, * Poole, Mackworth & Goebel 1998, pp. 443–460 Moving and configuration space: * Russell & Norvig 2003, pp. 916–932 Tecuci 2012. Robotic mapping (localization, etc): * Russell & Norvig 2003, pp. 908–915 Cadena, Cesar; Carlone, Luca; Carrillo, Henry; Latif, Yasir; Scaramuzza, Davide; Neira, Jose; Reid, Ian; Leonard, John J. (December 2016). "Past, Present, and Future of Simultaneous Localization and Mapping: Toward the Robust-Perception Age". IEEE Transactions on Robotics. 32 (6): 1309–1332. arXiv:1606.05830. Bibcode:2016arXiv160605830C. doi:10.1109/TRO.2016.2624754. S2CID 2596787. Moravec, Hans (1988). Mind Children. Harvard University Press. p. 15. Chan, Szu Ping (15 November 2015). "This is what will happen when robots take over the world". Archived from the original on 24 April 2018. Retrieved 23 April 2018. "IKEA furniture and the limits of AI". The Economist. 2018. Archived from the original on 24 April 2018. Retrieved 24 April 2018. Kismet. Thompson, Derek (2018). "What Jobs Will the Robots Take?". The Atlantic. Archived from the original on 24 April 2018. Retrieved 24 April 2018. Scassellati, Brian (2002). "Theory of mind for a humanoid robot". Autonomous Robots. 12 (1): 13–24. doi:10.1023/A:1013298507114. S2CID 1979315. Cao, Yongcan; Yu, Wenwu; Ren, Wei; Chen, Guanrong (February 2013). "An Overview of Recent Progress in the Study of Distributed Multi-Agent Coordination". IEEE Transactions on Industrial Informatics. 9 (1): 427–438. arXiv:1207.3231. doi:10.1109/TII.2012.2219061. S2CID 9588126. Thro 1993. Edelson 1991. Tao & Tan 2005. Poria, Soujanya; Cambria, Erik; Bajpai, Rajiv; Hussain, Amir (September 2017). "A review of affective computing: From unimodal analysis to multimodal fusion". Information Fusion. 37: 98–125. doi:10.1016/j.inffus.2017.02.003. hdl:1893/25490. Emotion and affective computing: * Minsky 2006 Waddell, Kaveh (2018). "Chatbots Have Entered the Uncanny Valley". The Atlantic. Archived from the original on 24 April 2018. Retrieved 24 April 2018. Pennachin, C.; Goertzel, B. (2007). Contemporary Approaches to Artificial General Intelligence. Artificial General Intelligence. Cognitive Technologies. Cognitive Technologies. Berlin, Heidelberg: Springer. doi:10.1007/978-3-540-68677-4_1. ISBN 978-3-540-23733-4. Roberts, Jacob (2016). "Thinking Machines: The Search for Artificial Intelligence". Distillations. Vol. 2 no. 2. pp. 14–23. Archived from the original on 19 August 2018. Retrieved 20 March 2018. "The superhero of artificial intelligence: can this genius keep it in check?". the Guardian. 16 February 2016. Archived from the original on 23 April 2018. Retrieved 26 April 2018. Mnih, Volodymyr; Kavukcuoglu, Koray; Silver, David; Rusu, Andrei A.; Veness, Joel; Bellemare, Marc G.; Graves, Alex; Riedmiller, Martin; Fidjeland, Andreas K.; Ostrovski, Georg; Petersen, Stig; Beattie, Charles; Sadik, Amir; Antonoglou, Ioannis; King, Helen; Kumaran, Dharshan; Wierstra, Daan; Legg, Shane; Hassabis, Demis (26 February 2015). "Human-level control through deep reinforcement learning". Nature. 518 (7540): 529–533. Bibcode:2015Natur.518..529M. doi:10.1038/nature14236. PMID 25719670. S2CID 205242740. Sample, Ian (14 March 2017). "Google's DeepMind makes AI program that can learn like a human". the Guardian. Archived from the original on 26 April 2018. Retrieved 26 April 2018. "From not working to neural networking". The Economist. 2016. Archived from the original on 31 December 2016. Retrieved 26 April 2018. Domingos 2015. Artificial brain arguments: AI requires a simulation of the operation of the human brain * Russell & Norvig 2003, p. 957 * Crevier 1993, pp. 271 and 279 A few of the people who make some form of the argument: * Moravec 1988 * Kurzweil 2005, p. 262 * Hawkins & Blakeslee 2005 The most extreme form of this argument (the brain replacement scenario) was put forward by Clark Glymour in the mid-1970s and was touched on by Zenon Pylyshyn and John Searle in 1980. Goertzel, Ben; Lian, Ruiting; Arel, Itamar; de Garis, Hugo; Chen, Shuo (December 2010). "A world survey of artificial brain projects, Part II: Biologically inspired cognitive architectures". Neurocomputing. 74 (1–3): 30–49. doi:10.1016/j.neucom.2010.08.012. Nilsson 1983, p. 10. Nils Nilsson writes: "Simply put, there is wide disagreement in the field about what AI is all about."[163] AI's immediate precursors: * McCorduck 2004, pp. 51–107 * Crevier 1993, pp. 27–32 * Russell & Norvig 2003, pp. 15, 940 * Moravec 1988, p. 3 Haugeland 1985, pp. 112–117 The most dramatic case of sub-symbolic AI being pushed into the background was the devastating critique of perceptrons by Marvin Minsky and Seymour Papert in 1969. See History of AI, AI winter, or Frank Rosenblatt. Cognitive simulation, Newell and Simon, AI at CMU (then called Carnegie Tech): * McCorduck 2004, pp. 139–179, 245–250, 322–323 (EPAM) * Crevier 1993, pp. 145–149 Soar (history): * McCorduck 2004, pp. 450–451 * Crevier 1993, pp. 258–263 McCarthy and AI research at SAIL and SRI International: * McCorduck 2004, pp. 251–259 * Crevier 1993 AI research at Edinburgh and in France, birth of Prolog: * Crevier 1993, pp. 193–196 * Howe 1994 AI at MIT under Marvin Minsky in the 1960s : * McCorduck 2004, pp. 259–305 * Crevier 1993, pp. 83–102, 163–176 * Russell & Norvig 2003, p. 19 Cyc: * McCorduck 2004, p. 489, who calls it "a determinedly scruffy enterprise" * Crevier 1993, pp. 239–243 * Russell & Norvig 2003, p. 363−365 * Lenat & Guha 1989 Knowledge revolution: * McCorduck 2004, pp. 266–276, 298–300, 314, 421 * Russell & Norvig 2003, pp. 22–23 Frederick, Hayes-Roth; William, Murray; Leonard, Adelman. "Expert systems". AccessScience. doi:10.1036/1097-8542.248550. Embodied approaches to AI: * McCorduck 2004, pp. 454–462 * Brooks 1990 * Moravec 1988 Weng et al. 2001. Lungarella et al. 2003. Asada et al. 2009. Oudeyer 2010. Revival of connectionism: * Crevier 1993, pp. 214–215 * Russell & Norvig 2003, p. 25 Computational intelligence * IEEE Computational Intelligence Society Archived 9 May 2008 at the Wayback Machine Hutson, Matthew (16 February 2018). "Artificial intelligence faces reproducibility crisis". Science. pp. 725–726. Bibcode:2018Sci...359..725H. doi:10.1126/science.359.6377.725. Archived from the original on 29 April 2018. Retrieved 28 April 2018. Norvig 2012. Langley 2011. Katz 2012. The intelligent agent paradigm: * Russell & Norvig 2003, pp. 27, 32–58, 968–972 * Poole, Mackworth & Goebel 1998, pp. 7–21 * Luger & Stubblefield 2004, pp. 235–240 * Hutter 2005, pp. 125–126 The definition used in this article, in terms of goals, actions, perception and environment, is due to Russell & Norvig (2003). Other definitions also include knowledge and learning as additional criteria. Agent architectures, hybrid intelligent systems: * Russell & Norvig (2003, pp. 27, 932, 970–972) * Nilsson (1998, chpt. 25) Hierarchical control system: * Albus 2002 Lieto, Antonio; Lebiere, Christian; Oltramari, Alessandro (May 2018). "The knowledge level in cognitive architectures: Current limitations and possibile developments". Cognitive Systems Research. 48: 39–55. doi:10.1016/j.cogsys.2017.05.001. hdl:2318/1665207. S2CID 206868967. Lieto, Antonio; Bhatt, Mehul; Oltramari, Alessandro; Vernon, David (May 2018). "The role of cognitive architectures in general artificial intelligence". Cognitive Systems Research. 48: 1–3. doi:10.1016/j.cogsys.2017.08.003. hdl:2318/1665249. S2CID 36189683. Russell & Norvig 2009, p. 1. White Paper: On Artificial Intelligence - A European approach to excellence and trust (PDF). Brussels: European Commission. 2020. p. 1. Archived (PDF) from the original on 20 February 2020. Retrieved 20 February 2020. CNN 2006. Using AI to predict flight delays Archived 20 November 2018 at the Wayback Machine, Ishti.org. N. Aletras; D. Tsarapatsanis; D. Preotiuc-Pietro; V. Lampos (2016). "Predicting judicial decisions of the European Court of Human Rights: a Natural Language Processing perspective". PeerJ Computer Science. 2: e93. doi:10.7717/peerj-cs.93. "The Economist Explains: Why firms are piling into artificial intelligence". The Economist. 31 March 2016. Archived from the original on 8 May 2016. Retrieved 19 May 2016. Lohr, Steve (28 February 2016). "The Promise of Artificial Intelligence Unfolds in Small Steps". The New York Times. Archived from the original on 29 February 2016. Retrieved 29 February 2016. Frangoul, Anmar (14 June 2019). "A Californian business is using A.I. to change the way we think about energy storage". CNBC. Archived from the original on 25 July 2020. Retrieved 5 November 2019. Wakefield, Jane (15 June 2016). "Social media 'outstrips TV' as news source for young people". BBC News. Archived from the original on 24 June 2016. Smith, Mark (22 July 2016). "So you think you chose to read this article?". BBC News. Archived from the original on 25 July 2016. Brown, Eileen. "Half of Americans do not believe deepfake news could target them online". ZDNet. Archived from the original on 6 November 2019. Retrieved 3 December 2019. The Turing test: Turing's original publication: * Turing 1950 Historical influence and philosophical implications: * Haugeland 1985, pp. 6–9 * Crevier 1993, p. 24 * McCorduck 2004, pp. 70–71 * Russell & Norvig 2003, pp. 2–3 and 948 Dartmouth proposal: * McCarthy et al. 1955 (the original proposal) * Crevier 1993, p. 49 (historical significance) The physical symbol systems hypothesis: * Newell & Simon 1976, p. 116 * McCorduck 2004, p. 153 * Russell & Norvig 2003, p. 18 Dreyfus 1992, p. 156. Dreyfus criticized the necessary condition of the physical symbol system hypothesis, which he called the "psychological assumption": "The mind can be viewed as a device operating on bits of information according to formal rules."[206] Dreyfus' critique of artificial intelligence: * Dreyfus 1972, Dreyfus & Dreyfus 1986 * Crevier 1993, pp. 120–132 * McCorduck 2004, pp. 211–239 * Russell & Norvig 2003, pp. 950–952, Gödel 1951: in this lecture, Kurt Gödel uses the incompleteness theorem to arrive at the following disjunction: (a) the human mind is not a consistent finite machine, or (b) there exist Diophantine equations for which it cannot decide whether solutions exist. Gödel finds (b) implausible, and thus seems to have believed the human mind was not equivalent to a finite machine, i.e., its power exceeded that of any finite machine. He recognized that this was only a conjecture, since one could never disprove (b). Yet he considered the disjunctive conclusion to be a "certain fact". The Mathematical Objection: * Russell & Norvig 2003, p. 949 * McCorduck 2004, pp. 448–449 Making the Mathematical Objection: * Lucas 1961 * Penrose 1989 Refuting Mathematical Objection: * Turing 1950 under "(2) The Mathematical Objection" * Hofstadter 1979 Background: * Gödel 1931, Church 1936, Kleene 1935, Turing 1937 Graham Oppy (20 January 2015). "Gödel's Incompleteness Theorems". Stanford Encyclopedia of Philosophy. Archived from the original on 22 April 2016. Retrieved 27 April 2016. These Gödelian anti-mechanist arguments are, however, problematic, and there is wide consensus that they fail. Stuart J. Russell; Peter Norvig (2010). "26.1.2: Philosophical Foundations/Weak AI: Can Machines Act Intelligently?/The mathematical objection". Artificial Intelligence: A Modern Approach (3rd ed.). Upper Saddle River, NJ: Prentice Hall. ISBN 978-0-13-604259-4. even if we grant that computers have limitations on what they can prove, there is no evidence that humans are immune from those limitations. Mark Colyvan. An introduction to the philosophy of mathematics. Cambridge University Press, 2012. From 2.2.2, 'Philosophical significance of Gödel's incompleteness results': "The accepted wisdom (with which I concur) is that the Lucas-Penrose arguments fail." Iphofen, Ron; Kritikos, Mihalis (3 January 2019). "Regulating artificial intelligence and robotics: ethics by design in a digital society". Contemporary Social Science: 1–15. doi:10.1080/21582041.2018.1563803. ISSN 2158-2041. "Ethical AI Learns Human Rights Framework". Voice of America. Archived from the original on 11 November 2019. Retrieved 10 November 2019. Crevier 1993, pp. 132–144. In the early 1970s, Kenneth Colby presented a version of Weizenbaum's ELIZA known as DOCTOR which he promoted as a serious therapeutic tool.[216] Joseph Weizenbaum's critique of AI: * Weizenbaum 1976 * Crevier 1993, pp. 132–144 * McCorduck 2004, pp. 356–373 * Russell & Norvig 2003, p. 961 Weizenbaum (the AI researcher who developed the first chatterbot program, ELIZA) argued in 1976 that the misuse of artificial intelligence has the potential to devalue human life. Wendell Wallach (2010). Moral Machines, Oxford University Press. Wallach, pp 37–54. Wallach, pp 55–73. Wallach, Introduction chapter. Michael Anderson and Susan Leigh Anderson (2011), Machine Ethics, Cambridge University Press. "Machine Ethics". aaai.org. Archived from the original on 29 November 2014. Rubin, Charles (Spring 2003). "Artificial Intelligence and Human Nature". The New Atlantis. 1: 88–100. Archived from the original on 11 June 2012. Brooks, Rodney (10 November 2014). "artificial intelligence is a tool, not a threat". Archived from the original on 12 November 2014. "Stephen Hawking, Elon Musk, and Bill Gates Warn About Artificial Intelligence". Observer. 19 August 2015. Archived from the original on 30 October 2015. Retrieved 30 October 2015. Chalmers, David (1995). "Facing up to the problem of consciousness". Journal of Consciousness Studies. 2 (3): 200–219. Archived from the original on 8 March 2005. Retrieved 11 October 2018. See also this link Archived 8 April 2011 at the Wayback Machine Horst, Steven, (2005) "The Computational Theory of Mind" Archived 11 September 2018 at the Wayback Machine in The Stanford Encyclopedia of Philosophy Searle 1980, p. 1. This version is from Searle (1999), and is also quoted in Dennett 1991, p. 435. Searle's original formulation was "The appropriately programmed computer really is a mind, in the sense that computers given the right programs can be literally said to understand and have other cognitive states." [230] Strong AI is defined similarly by Russell & Norvig (2003, p. 947): "The assertion that machines could possibly act intelligently
adamlilith / EnmSdmFaster, better, smarter ecological niche modeling and species distribution modeling
tud-amr / Biased MppiPyTorch examples for the IEEE Robotics and Automation Letters paper titled "Biased-MPPI: Informing Sampling-Based Model Predictive Control by Fusing Ancillary Controllers"
indraneelpatil / Bottleneck Guided RrtThis project consists of C++ implementations of a 3D Rapidly Exploring Random Tree and three other extensions called RRT*, Execution Extended RRT and Synchronised Greedy Biased RRT. It also includes a heuristically guided RRT* with biased sampling towards relevant bottleneck points predicted by a 3D CNN(modified VoxNet in Tensorflow).
reddyprasade / Machine Learning Interview PreparationPrepare to Technical Skills Here are the essential skills that a Machine Learning Engineer needs, as mentioned Read me files. Within each group are topics that you should be familiar with. Study Tip: Copy and paste this list into a document and save to your computer for easy referral. Computer Science Fundamentals and Programming Topics Data structures: Lists, stacks, queues, strings, hash maps, vectors, matrices, classes & objects, trees, graphs, etc. Algorithms: Recursion, searching, sorting, optimization, dynamic programming, etc. Computability and complexity: P vs. NP, NP-complete problems, big-O notation, approximate algorithms, etc. Computer architecture: Memory, cache, bandwidth, threads & processes, deadlocks, etc. Probability and Statistics Topics Basic probability: Conditional probability, Bayes rule, likelihood, independence, etc. Probabilistic models: Bayes Nets, Markov Decision Processes, Hidden Markov Models, etc. Statistical measures: Mean, median, mode, variance, population parameters vs. sample statistics etc. Proximity and error metrics: Cosine similarity, mean-squared error, Manhattan and Euclidean distance, log-loss, etc. Distributions and random sampling: Uniform, normal, binomial, Poisson, etc. Analysis methods: ANOVA, hypothesis testing, factor analysis, etc. Data Modeling and Evaluation Topics Data preprocessing: Munging/wrangling, transforming, aggregating, etc. Pattern recognition: Correlations, clusters, trends, outliers & anomalies, etc. Dimensionality reduction: Eigenvectors, Principal Component Analysis, etc. Prediction: Classification, regression, sequence prediction, etc.; suitable error/accuracy metrics. Evaluation: Training-testing split, sequential vs. randomized cross-validation, etc. Applying Machine Learning Algorithms and Libraries Topics Models: Parametric vs. nonparametric, decision tree, nearest neighbor, neural net, support vector machine, ensemble of multiple models, etc. Learning procedure: Linear regression, gradient descent, genetic algorithms, bagging, boosting, and other model-specific methods; regularization, hyperparameter tuning, etc. Tradeoffs and gotchas: Relative advantages and disadvantages, bias and variance, overfitting and underfitting, vanishing/exploding gradients, missing data, data leakage, etc. Software Engineering and System Design Topics Software interface: Library calls, REST APIs, data collection endpoints, database queries, etc. User interface: Capturing user inputs & application events, displaying results & visualization, etc. Scalability: Map-reduce, distributed processing, etc. Deployment: Cloud hosting, containers & instances, microservices, etc. Move on to the final lesson of this course to find lots of sample practice questions for each topic!
drimpossible / Sampling Bias Active LearningCode for the paper "Sampling Bias in Deep Active Classification: An Empirical Study"
PhoebusSi / MMBSCode for our EMNLP-2022 paper: "Towards Robust Visual Question Answering: Making the Most of Biased Samples via Contrastive Learning"
UNHSAILLab / SentimentalLIAROur Sentimental LIAR dataset is a modified and further extended version of the LIAR extension introduced by Kirilin et al. In our dataset, the multi-class labeling of LIAR is converted to a binary annotation by changing half-true, false, barely-true and pants-fire labels to False, and the remaining labels to True. Furthermore, we convert the speaker names to numerical IDs in order to avoid bias with regards to the textual representation of names. The binary-label dataset is then extended by adding sentiments derived using the Google NLP API . Sentiment analysis determines the overall attitude of the text (i.e., whether it is positive or negative), and is quantified by a numerical score. If the sentiment score is positive, then we assign Positive for the sentiment attribute, otherwise Negative is assigned. We also introduced a further extension by adding emotion scores extracted using the IBM NLP API for each claim, which determine the detected level of 6 emotional states namely anger, sadness, disgust, fear and joy. The score for each emotion is between the range of 0 and 1. Table I demonstrates a sample record in Sentimental LIAR for a short claim in the LIAR dataset This repository contains the dataset for this paper: https://arxiv.org/abs/2009.01047
KikimorMay / FSL TCBRAlleviating the Sample Selection Bias in Few-shot Learning by Removing Projection to the Centroid
mqraitem / Bias MimickingBias Mimicking: A simple sampling approach for Bias Mitigation (CVPR 23)
