How Residual Connections Are Getting an Upgrade [mHC]

Jia-Bin Huang · Beginner ·🧠 Large Language Models ·2mo ago
Residual connections are adopted in virtually every deep learning model. BUT, can we further improve the residual connections? Hyper-connections are an exciting recent exploration to generalize residual connections. In this video, we will cover the following: 00:00 Training deep models 00:46 Residual connections 02:45 Hyper-connections - Intuition 04:48 Hyper-connections - Math 05:54 Dynamic hyper-connections 06:47 Training instability 07:52 Example of instability 08:44 mHC: Stabilizing training 11:02 mHC: Improving parametrization 12:30 mHC: Efficient infrastructure designs References: …
Watch on YouTube ↗ (saves to browser)

Chapters (10)

Training deep models
0:46 Residual connections
2:45 Hyper-connections - Intuition
4:48 Hyper-connections - Math
5:54 Dynamic hyper-connections
6:47 Training instability
7:52 Example of instability
8:44 mHC: Stabilizing training
11:02 mHC: Improving parametrization
12:30 mHC: Efficient infrastructure designs
5 Levels of AI Agents - From Simple LLM Calls to Multi-Agent Systems
Next Up
5 Levels of AI Agents - From Simple LLM Calls to Multi-Agent Systems
Dave Ebbelaar (LLM Eng)