Neural Network Renormalization Group

Explore the intersection of renormalization group techniques and deep learning in this 27-minute conference talk by Lei Wang from the Chinese Academy of Sciences. Delve into the concept of Neural Network Renormalization Group, starting with an introduction to RG and deep learning. Examine the Multi-Scale Entanglement Renormalization Ansatz (MERA) and its representation as a quantum circuit. Understand probability transformation through visual aids and study a toy problem using harmonic oscillators. Learn about neural bijectors, their group properties, and training methods like Probability Density Distillation. Gain insights into the WaveNet story and variational loss. Discover how to interpret latent variables and their practical applications. Explore the latent space through wandering and hybrid Monte Carlo methods. Investigate the connection between machine learning and holographic renormalization group. Conclude with a timeline of generative models and a perspective on deep learning as a fluid control problem.

Intro
RG and Deep Learning
Motivation
Multi-Scale Entanglement Renormalization Ansatz
MERA as a quantum circuit
Neural Network Renormalization Group
Probability transformation in picture
Toy problem: Harmonic oscillator
Neural Bijectors
Bijectors form a group
Training: Probability Density Distillation
Interlude: The WaveNet Story
Variational Loss
What is the network learning?
How to interpret the latent variables ?
How is this useful?
Wander in the latent space
Latent space Hybrid MC
MI and holographic RG
Timeline on Generative Models
DL as a fluid control problem