AlphaTensor: Using Reinforcement Learning for Efficient Matrix Multiplication

Channel:
Singapore John Tan Chong Min
Subscribers:
5,330
Published on ● Video Link: https://www.youtube.com/watch?v=6nYqyTzOQjw


Category:
Let's Play
Duration: 1:13:55
611 views
16

DeepMind has recently released AlphaTensor, using an AlphaZero-like approach to treat Matrix Multiplication as a single player game, in order to find the minimum number of multiplication steps needed.

There are a lot of problem-specific modifications to the baseline AlphaZero architecture, most notably axial attention and the 3D tensor grid modelling, which likely helps AlphaTensor learn more effectively. Data augmentation also plays a huge role in giving success cases for AlphaTensor to learn from. In the future, I believe a combination of human guided heuristic and AI-led smart "brute force" search will be the key to solving complicated problems.

Slides can be downloaded at: https://github.com/tanchongmin/TensorFlow-Implementations/tree/main/Paper_Reviews

AlphaTensor Paper: https://www.nature.com/articles/s41586-022-05172-4

AlphaGo Zero Paper: https://discovery.ucl.ac.uk/id/eprint/10045895/1/agz_unformatted_nature.pdf

FBHHRBNRSSSHK-Algorithm Paper: https://arxiv.org/abs/2210.04045

~~~~~~~~~

Timestamps:
0:00 Introduction
7:28 How AlphaGo Zero / AlphaZero works
13:33 AlphaTensor: Applying AlphaZero to Matrix Multiplication
15:13 The Simplified Game (Lightbulbs)
17:45 Formulating Matrix Multiplication as a 3D Tensor
24:21 Strassen's Algorithm
31:08 Main Algorithm using 3D Tensors
33:30 How to play the game
35:15 How to reward the agent
39:21 Acting and Learning Process
41:57 Additional Data Augmentation
49:28 Optimising for Hardware
54:15 Training Time
55:04 Network Architecture
1:00:22 Thoughts on why AlphaTensor works
1:06:13 Questions to Ponder
1:09:10 FBHHRBNRSSSHK-Algorithm
1:11:32 Discussion

~~~~~~~~~~

AI and ML enthusiast. Likes to think about the essences behind breakthroughs of AI and explain it in a simple and relatable way. Also, I am an avid game creator.

Discord: https://discord.gg/fXCZCPYs
LinkedIn: https://www.linkedin.com/in/chong-min-tan-94652288/
Online AI blog: https://delvingintotech.wordpress.com/.
Twitter: https://twitter.com/johntanchongmin
Try out my games here: https://simmer.io/@chongmin



Other Videos By John Tan Chong Min


2023-02-07Using Logic Gates as Neurons - Deep Differentiable Logic Gate Networks!
2023-01-31Learn from External Memory, not just Weights: Large-Scale Retrieval for Reinforcement Learning
2023-01-17How ChatGPT works - From Transformers to Reinforcement Learning with Human Feedback (RLHF)
2023-01-09HyperTree Proof Search - Automated Theorem Proving with AlphaZero and Transformers!
2022-12-23CodinGame Fall Challenge 2022: A First Look (managed to get to Silver!)
2022-12-21Can ChatGPT solve CodinGame/Google Kickstart problems?
2022-12-19Reinforcement Learning Fast and Slow: Goal-Directed and Memory Retrieval Mechanism!
2022-12-12A New Framework of Memory for Learning (Part 1)
2022-11-14Hippocampal Replay for Learning (Full Length with Questions)
2022-11-14Hippocampal Replay for Learning (3 min summary)
2022-11-07AlphaTensor: Using Reinforcement Learning for Efficient Matrix Multiplication
2022-10-27Playing Go on TyGem and learning from AI (~ 3 kyu)
2022-10-13Heroes of Might and Magic III - Armageddon's Blade Campaign (First Playthrough) - Final!!!
2022-10-13Heroes of Might and Magic III - Armageddon's Blade Campaign (First Playthrough) - Part 6
2022-10-11Playing Go on Tygem + AI Analysis (~4 kyu)
2022-10-11Heroes of Might and Magic III - Armageddon's Blade Campaign (First Playthrough) - Part 5
2022-10-11Heroes of Might and Magic III - Armageddon's Blade Campaign (First Playthrough) - Part 4
2022-10-10Playing Go on Tygem + AI Analysis (~4 kyu)
2022-10-10Heroes of Might and Magic III - Armageddon's Blade Campaign (First Playthrough) - Part 3
2022-10-10Heroes of Might and Magic III - Armageddon's Blade Campaign (First Playthrough) - Part 2
2022-10-10Heroes of Might and Magic III - Armageddon's Blade Campaign (First Playthrough) - Part 1