(T) I attended this week, the third Stanford Graph Learning Workshop. The workshop is organized by Professor Jure Leskovec, and his students. Some of the paper presentations are very impressive. Following are my notes from the workshop:
Technology:
- Latest update on PyTorch Geometric (PyG)
- Positional encodings for graphs in order to use transformers for graph learning
- Retrieving from knowledge bases (e.g graph-based) to assist LLMs predictions
- Kumo‘s declarative language that takes SQL tables, learns the labels from past data, and builds the model based on graph learning to make predictions
Applications:
- Cell engineering/drug discovery: Predicting the outcome of a genetic cell perturbation
- Zero-shot casual learning to predict the outcome of a personalized medical treatment leveraging a meta-learning framework
- 3D molecule generation using geometric latent diffusion models
- Temporal graphs for modeling supply chains networks (suppliers, buyers, products)
- Replacing researchers with AI research agents, based on LLMs, and benchmarking AI research agents actions (such as reading/writing files, executing code, and inspecting outputs)
Video the workshop:
Note: The picture above is “still life and blossoming almond trees” by Diego Rivera.
Copyright © 2005-2023 by Serge-Paul Carrasco. All rights reserved.
Contact Us: asvinsider at gmail dot com.
Categories: Algorithms, Deep Learning