KAISER

Knowledge Acquisition and Interlinking via Semantic Embeddings and Reasoning.

This project aims to harness Large Language Models, Text Transformers, Graphical Neural Networks and other models for use on the vast corpus of digital science available (Articles, Books etc).

We aim to produce datasets, finetuned models and pipelines in the open, for everyone. As such, anyone is welcome to contribute to research and the conversation, starting by joining the discord channels is probably best.

We envision that through collaborative efforts, open data, and the latest ML methods, we can produce a pipeline of many interlinked models that will be effective for synthesising existing knowledge, discovery of new knowledge and data, as well as new theories in a safe and reliable way.

Through a multi-model approach we aim to enhance the quality, reliability, safety, generalizability and scalability of the system. Meanwhile, the individual datasets and models produced will be able to stand on their own and should have uses for a large range of human & human/machine collaborative research.

A potential KAISER pipeline is shown below:

Links

We chat generally on LAOIN and on our own KAISER server Notes & Docs: https://www.notion.so/kaiserml/ Models & Data: GDrive (Ask for perms) & Huggingface (KAISERML)