Tensor Logic: The Language of AI
summary
Proposes TensorLogic, a language where relations map to Boolean tensors and rules map to Einstein summation plus a step nonlinearity, so logical joins become tensor joins and projections. Implements perceptron, MLP, RNN, ConvNet and a transformer as concise tensor equations, learns everything by gradient descent because gradients are themselves tensor-equation programs, and supports backpropagation through structure. Claims GPU-friendly scalability and the ability to do sound reasoning in embedding space while preserving transparency and automatic inference.
AI-written summary · about
chapters (10)
- 0:00Introduction
- 3:27Ideal AI language
- 7:29Logic and tensor basics
- 11:04Unifying logic and tensors
- 13:57Core operations and syntax
- 17:44Neural networks in Tensor Logic
- 21:00Learning and inference
- 25:07ConvNets and Transformers
- 28:08Symbolic AI beyond
- 29:10Summary
related talks
18mAI Is Not Magic: The Physics, Math, and Hardware Behind It
34mThe physics of deep learning using tensor networks
45mTensorly: A Flexible Python Framework for Machine Learning
07mWTT: What the Tensor?
20mTensorRight: Automated Verification of Tensor Graph Rewrites
15mEnabling Tensor Language Model to Assist in Generating High-Performance Tensor...
58mIntroduction to the tensor-programs framework, a PL approach that helps analyse ...
33mTeIL: a type-safe imperative Tensor Intermediate Language