33 skills found · Page 1 of 2
hkociemba / RubiksCube TwophaseSolverSolve Rubik's Cube in less than 19 moves on average with Python.
cs0x7f / Min2phaseRubik's Cube Solver. An optimized implementation of Kociemba's two-phase algorithm.
hkociemba / CubeExplorerA program about Rubik's Cube with a lot of features.
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.
abhir98 / RansomwareProject Summary This project was developed for the Computer Security course at my academic degree. Basically, it will encrypt your files in background using AES-256-CTR, a strong encryption algorithm, using RSA-4096 to secure the exchange with the server, optionally using the Tor SOCKS5 Proxy. The base functionality is what you see in the famous ransomware Cryptolocker. The project is composed by three parts, the server, the malware and the unlocker. The server store the victim's identification key along with the encryption key used by the malware. The malware encrypt with a RSA-4096 (RSA-OAEP-4096 + SHA256) public key any payload before send then to the server. This approach with the optional Tor Proxy and a .onion domain allow you to hide almost completely your server. Features Run in Background (or not) Encrypt files using AES-256-CTR(Counter Mode) with random IV for each file. Multithreaded. RSA-4096 to secure the client/server communication. Includes an Unlocker. Optional TOR Proxy support. Use an AES CTR Cypher with stream encryption to avoid load an entire file into memory. Walk all drives by default. Docker image for compilation. Building the binaries DON'T RUN ransomware.exe IN YOUR PERSONAL MACHINE, EXECUTE ONLY IN A TEST ENVIRONMENT! I'm not resposible if you acidentally encrypt all of your disks! First of all download the project outside your $GOPATH: git clone github.com/mauri870/ransomware cd ransomware If you have Docker skip to the next section. You need Go at least 1.11.2 with the $GOPATH/bin in your $PATH and $GOROOT pointing to your Go installation folder. For me: export GOPATH=~/gopath export PATH=$PATH:$GOPATH/bin export GOROOT=/usr/local/go Build the project require a lot of steps, like the RSA key generation, build three binaries, embed manifest files, so, let's leave make do your job: make deps make You can build the server for windows with make -e GOOS=windows. Docker ./build-docker.sh make Config Parameters You can change some of the configs during compilation. Instead of run only make, you can use the following variables: HIDDEN='-H windowsgui' # optional. If present the malware will run in background USE_TOR=true # optional. If present the malware will download the Tor proxy and use it to contact the server SERVER_HOST=mydomain.com # the domain used to connect to your server. localhost, 0.0.0.0, 127.0.0.1 works too if you run the server on the same machine as the malware SERVER_PORT=8080 # the server port, if using a domain you can set this to 80 GOOS=linux # the target os to compile the server. Eg: darwin, linux, windows Example: make -e USE_TOR=true SERVER_HOST=mydomain.com SERVER_PORT=80 GOOS=darwin The SERVER_ variables above only apply to the malware. The server has a flag --port that you can use to change the port that it will listen on. DON'T RUN ransomware.exe IN YOUR PERSONAL MACHINE, EXECUTE ONLY IN A TEST ENVIRONMENT! I'm not resposible if you acidentally encrypt all of your disks! Step by Step Demo and How it Works For this demo I'll use two machines, my personal linux machine and a windows 10 VM. For the sake of simplicity, I have a folder mapped to the VM, so I can compile from my linux and copy to the vm. In this demo we will use the Ngrok tool, this will allow us to expose our server using a domain, but you can use your own domain or ip address if you want. We are also going to enable the Tor transport, so .onion domains will work without problems. First of all lets start our external domain: ngrok http 8080 This command will give us a url like http://2af7161c.ngrok.io. Keep this command running otherwise the malware won't reach our server. Let's compile the binaries (remember to replace the domain): make -e SERVER_HOST=2af7161c.ngrok.io SERVER_PORT=80 USE_TOR=true The SERVER_PORT needs to be 80 in this case, since ngrok redirects 2af7161c.ngrok.io:80 to your local server port 8080. After build, a binary called ransomware.exe, and unlocker.exe along with a folder called server will be generated in the bin folder. The execution of ransomware.exe and unlocker.exe (even if you use a diferent GOOS variable during compilation) is locked to windows machines only. Enter the server directory from another terminal and start it: cd bin/server && ./server --port 8080 To make sure that all is working correctly, make a http request to http://2af7161c.ngrok.io: curl http://2af7161c.ngrok.io If you see a OK and some logs in the server output you are ready to go. Now move the ransomware.exe and unlocker.exe to the VM along with some dummy files to test the malware. You can take a look at cmd/common.go to see some configuration options like file extensions to match, directories to scan, skipped folders, max size to match a file among others. Then simply run the ransomware.exe and see the magic happens 😄. The window that you see can be hidden using the HIDDEN option described in the compilation section. After download, extract and start the Tor proxy, the malware waits until the tor bootstrapping is done and then proceed with the key exchange with the server. The client/server handshake takes place and the client payload, encrypted with an RSA-4096 public key must be correctly decrypted on the server. The victim identification and encryption keys are stored in a Golang embedded database called BoltDB (it also persists on disk). When completed we get into the find, match and encrypt phase, up to N-cores workers start to encrypt files matched by the patterns defined. This proccess is really quick and in seconds all of your files will be gone. The encryption key exchanged with the server was used to encrypt all of your files. Each file has a random primitive called IV, generated individually and saved as the first 16 bytes of the encrypted content. The algorithm used is AES-256-CTR, a good AES cypher with streaming mode of operation such that the file size is left intact. The only two sources of information available about what just happen are the READ_TO_DECRYPT.html and FILES_ENCRYPTED.html in the Desktop. In theory, to decrypt your files you need to send an amount of BTC to the attacker's wallet, followed by a contact sending your ID(located on the file created on desktop). If the attacker can confirm your payment it will possibly(or maybe not) return your encryption key and the unlocker.exe and you can use then to recover your files. This exchange can be accomplished in several ways and WILL NOT be implemented in this project for obvious reasons. Let's suppose you get your encryption key back. To recover the correct key point to the following url: curl -k http://2af7161c.ngrok.io/api/keys/:id Where :id is your identification stored in the file on desktop. After, run the unlocker.exe by double click and follow the instructions. That's it, got your files back 😄 The server has only two endpoints: POST api/keys/add - Used by the malware to persist new keys. Some verifications are made, like the verification of the RSA autenticity. Returns 204 (empty content) in case of success or a json error. GET api/keys/:id - Id is a 32 characters parameter, representing an Id already persisted. Returns a json containing the encryption key or a json error The end As you can see, building a functional ransomware, with some of the best existing algorithms is not difficult, anyone with some programming skills can build that in any programming language.
Renovamen / Just A CubeA rubik's cube solver | 魔方还原（层先法 + Two-phase）
manthanthakker / SpeakerIdentificationNeuralNetworks⇨ The Speaker Recognition System consists of two phases, Feature Extraction and Recognition. ⇨ In the Extraction phase, the Speaker's voice is recorded and typical number of features are extracted to form a model. ⇨ During the Recognition phase, a speech sample is compared against a previously created voice print stored in the database. ⇨ The highlight of the system is that it can identify the Speaker's voice in a Multi-Speaker Environment too. Multi-layer Perceptron (MLP) Neural Network based on error back propagation training algorithm was used to train and test the system. ⇨ The system response time was 74 µs with an average efficiency of 95%.
Kunal30 / Non Intrusive Attendance Marking System Using AIThe project that we worked on this summer internship falls in the domain of research in IoT (Internet of Things). Initially, the mentor asked us to find real-life problems, which we would attempt to solve by using the tools of Information Technology. We were allowed to discuss and work in a group of three. We picked the problem of devising an attendance monitoring system, which would mark the presence of the students in a big room, in a non-intrusive manner using image recognition, for e.g. an auditorium or our college’s lecture theatre. Our project was divided into two phases, which would be illustrated in the subsequent passages. The first phase involved doing a literature survey on the tools and technologies through various authentic research papers and the existing libraries, which would enable us to devise a backend structure for our project. We, then developed a flowchart, which comprised of two modules of processes, through which the procedure would pass through. The first module involves the initial training of a machine learning based classifier by training it with the various images of a specific person. The second module involves the testing part in the real environment, which involves face detection and face recognition. A camera would take the frames/image of a live audience. Then, these frames would be pre-processed (involves grey-scaling and image resizing) for achieving better performance in the subsequent face detection module. The face-detection algorithm would detect all the faces present in the frame, and would crop the detected faces, and would pass them to the face recognition classifier for testing. The classifier would classify the cropped images and would mark the attendance accordingly. The libraries used for face-detection were that of OpenCV, and a convolutional neural network was trained for the image recognition part. The libraries which were used for training the convolutional neural network was Keras. The second phase involved the implementation part, where we had to gather the data for training the neural network, and find out the parameters of the image, for which we are getting better accuracy performance. We trained the neural network with the images of about 64 students, with about 20 images per student, covering different angles and brightness levels. We trained the network with 70 percent of the image corpus, and used the remaining 30 percent for testing. We got an accuracy of 93 percent. For testing the face detection part, we took a video of a classroom of about 40 students. Then, we generated frames from the video and passed it to the face detection algorithm. We extrapolated that the accuracy of an individual frame was not that high, but if we consider all the detected members in all the frames, we are covering almost every student. Hence, considering multiple frames for testing is crucial to get a high detection accuracy. We are currently trying to figure out the camera and its mounting position, which would be conducive for the algorithm, to give us accurate results.
abdallahkhairy / GP Data Analysis And MLHuman locomotion affects our daily living activities. Losing limbs or having neurological disorders with motor deficits could affect the quality of life. Gait analysis is a systematic study of human locomotion, which is defined as body movements through aerial, aquatic, or terrestrial space. This analysis has been used to study people ambulation, registration, and reconstruction of physical location and orientation of individual limbs used to quantify and characterize human locomotion using different gait parameters including gait activities such as walking, stairs ascending/descending, … etc., phases, and spatiotemporal parameters of human gait. Additionally, gait analysis parameters can be used to evaluate the functionality of patients and wearable system users. The evaluation is based on patient's stability, energy consumption, gait symmetry, ability to recover from perturbations, and ability to perform activities of daily living. Many companies develop assistive, wearable, and rehabilitation devices for patients with lower limb neurological disorders. These devices are tested and evaluated inside controlled lab environments. However, they don’t have enough data on the patient's performance in real world and harsh environments. Collecting large datasets of device users and their gait performance data in real environment are notoriously difficult. Additionally, collecting data on less prevalent or on gait activities other than level walking, stair ascending/descending, sitting, standing, …etc. on hard surfaces is rarely attempted. However, the scope for collecting gait data from alternative sources other than traditional gait labs could be attained with the help of IoT data collection embedded on the wearable and assistive devices and well-established cloud platforms equipped with big-data analytics and data visualization capabilities. This project aims to develop a cloud platform capable of collect data from wearable and assistive devices such as prostheses, exoskeleton, gait analysis wearable sensors, …etc. using IoT technologies. This platform is capable of automatically use data mining and visualization tools. Additionally, it uses statistical and machine learning techniques to estimate gait events, gait symmetry, gait speed, gait activities, stability, energy consumption, …etc. Also, it is capable of predicting patient's progress over time. The project will be composed of two major components, hardware component and software component. In hardware component, the students will design and implement the IoT that collects the different readings for gait analysis and send them to the cloud. Meanwhile, in the software component, the students will design and implement a set of algorithms to visualize the collected data, then design and implement data analytics to automatically analyze the collected data, so that we can estimate gait events, gait symmetry, gait speed, classify gait activities, stability, energy consumption, …etc. and predicting patient's progress over time. By analyzing the collected data, the patient's progress can be predicted over time. Additionally, these data can be used through manufacturers of prostheses legs to improve their products, as well as through health-care centers to assess the patient's performance. The following figures describe the main modules of our graduation project.
Megalomatt / KociembaC# Port of Kociemba's Two Phase Algorithm for Solving Rubik's cubes
tcbegley / Cube SolverA pure Python implementation of Herbert Kociemba's two-phase algorithm for solving the Rubik's Cube
daniel-j / Node Two Phase AlgorithmA port of Herbert Kociemba's Two-Phase-Algorithm for solving Rubik's cube.
crisbour / PhaseRetrievalCGH using Phase Retrieval Algorithm for parallel two photon polymerization.
tiagohm / CuberDart package for modeling and solving the 3x3x3 Rubik's Cube using Herbert Kociemba's two-phase algorithm.
rcombs / Cube.jsRubik's Cube solver, implementing Kociemba's Two-Phase Algorithm in JS
songyuncen / UnwrapSource code from the book "Two-Dimensional Phase Unwrapping: Theory, Algorithms, and Software"
aliasgharheidaricom / RUN Beyond The Metaphor An Efficient Optimization Algorithm Based On Runge Kutta MethodThe optimization field suffers from the metaphor-based “pseudo-novel” or “fancy” optimizers. Most of these cliché methods mimic animals' searching trends and possess a small contribution to the optimization process itself. Most of these cliché methods suffer from the locally efficient performance, biased verification methods on easy problems, and high similarity between their components' interactions. This study attempts to go beyond the traps of metaphors and introduce a novel metaphor-free population-based optimization based on the mathematical foundations and ideas of the Runge Kutta (RK) method widely well-known in mathematics. The proposed RUNge Kutta optimizer (RUN) was developed to deal with various types of optimization problems in the future. The RUN utilizes the logic of slope variations computed by the RK method as a promising and logical searching mechanism for global optimization. This search mechanism benefits from two active exploration and exploitation phases for exploring the promising regions in the feature space and constructive movement toward the global best solution. Furthermore, an enhanced solution quality (ESQ) mechanism is employed to avoid the local optimal solutions and increase convergence speed. The RUN algorithm's efficiency was evaluated by comparing with other metaheuristic algorithms in 50 mathematical test functions and four real-world engineering problems. The RUN provided very promising and competitive results, showing superior exploration and exploitation tendencies, fast convergence rate, and local optima avoidance. In optimizing the constrained engineering problems, the metaphor-free RUN demonstrated its suitable performance as well. The authors invite the community for extensive evaluations of this deep-rooted optimizer as a promising tool for real-world optimization. The source codes, supplementary materials, and guidance for the developed method will be publicly available at different hubs at http://aliasgharheidari.com/RUN.html.
maruthi1986 / Lung Cancer ClassificationThe primary goal of developing a smart city is to enhance the quality of lives of its citizens by providing infrastructure and offering smart healthcare. Smart Healthcare plays a significant role in achieving this objective of developing smart cities. Here, in this proposal our objective is to develop a smart health care system which uses artificial intelligence for the development of a decision support system in the medical field for the detection and segmentation of lung cancer. The proposed system is consisting of two phases, First phase will be consisting of various stages like Pre-processing, feature extraction, feature selection, classification and finally segmentation of the tumor. Input CT image is sent through the pre-processing phase where noise removal will be taken care and then texture features are extracted from the pre-processed image, and in the next stage features will be selected by making use of Crow Search Optimization Algorithm, later Artificial Neural Network is used for the classification of the normal lung images from abnormal images. Finally, abnormal images will be processed through the Fuzzy K-Means algorithm for segmenting the tumors separately. In the second phase, SVM classifier is used for the reduction of false positives. This methodology delivers a 96% of accuracy, 100% specificity and sensitivity of 99%. The accuracy of the decision taken by the Smart Health Care System exceed when compared to the accuracy of the decision taken by the doctors.
ZongSingHuang / Grey Wolf Optimizer Algorithm With A Two Phase MutationAbdel-Basset, M., El-Shahat, D., El-henawy, I., de Albuquerque, V. H. C., & Mirjalili, S. (2019). A new fusion of grey wolf optimizer algorithm with a two-phase mutation for feature selection. Expert Systems with Applications, 112824. doi:10.1016/j.eswa.2019.112824
cs0x7f / Min2phase RustAn optimized implementation of two-phase algorithm for solving Rubik's cube
