AI Models Blog

 

Robot Learning Part 7: Neural Manifolds — The Geometry of Skill Representation

robotics
neuroscience
manifolds
dimensionality-reduction
motor-control
robot-learning
In Part 6, we explored how eigenvalues control the dynamics of neural circuits and RNNs. We saw that eigenvalues near the unit circle with imaginary components produce…
Jan 25, 2026
Hujie Wang

Robot Learning Part 6: The Eigenvalue Sweet Spot — How Brain Dynamics and RNN Training Converged on the Same Answer

robotics
neuroscience
RNN
state-space-models
eigenvalues
motor-control
robot-learning
In Part 4, we discussed how motor cortex exhibits rotational dynamics — lawful oscillatory patterns during reaching, even though reaching itself is non-periodic. Churchland…
Jan 24, 2026
Hujie Wang

Robot Learning Part 5: Six Architectural Principles from Motor Neuroscience

robotics
neuroscience
motor-control
architecture
robot-learning
In Part 4, we surveyed how the brain plans and executes movements. We saw that the motor system employs hierarchical processing, forward models, reinforcement learning, and…
Jan 24, 2026
Hujie Wang

Robot Learning Part 1.5: Action Chunking with Transformers (ACT)

robotics
transformers
imitation-learning
VAE
robot-learning
Imagine teaching a robot to pick up a mug. In traditional behavioral cloning (recall from Part 1), the robot predicts one tiny action at a time — 50 micro-adjustments per…
Jan 23, 2026
Hujie Wang

Robot Learning Part 3: Vision-Language-Action (VLA) Models

robotics
VLA
transformers
foundation-models
robot-learning
Imagine you’re teaching someone to cook. You don’t just show them what ingredients look like, or explain recipes in words, or demonstrate knife techniques in isolation. You…
Jan 23, 2026
Hujie Wang

Robot Learning Part 0: Introduction & Roadmap

robotics
diffusion-policy
VLA
robot-learning
After spending time understanding diffusion models and flow matching, I realized these same techniques are revolutionizing robotics. The same principles that generate…
Jan 20, 2026
Hujie Wang

Robot Learning Part 1: Background & Current State of the Field

robotics
reinforcement-learning
imitation-learning
robot-learning
In Part 0, we surveyed the robot learning landscape and saw how diffusion models are revolutionizing robotic manipulation. But to truly understand why Diffusion Policy works…
Jan 20, 2026
Hujie Wang

Diffusion & Flow Matching Part 11: Building an Image Generator

diffusion
flow-matching
generative-models
tutorial
cifar-10
In the previous posts, we built up the mathematical foundations of flow matching:
Jan 20, 2026
Hujie Wang

Diffusion & Flow Matching Part 12: Why Diffusion?

The multimodal problem — why direct regression fails
diffusion
generative-models
deep-learning
flow-matching
Want to see the mode averaging problem in action? Check out the companion notebook where you can:
Jan 20, 2026
Hujie Wang

Diffusion & Flow Matching Part 10: Diffusion Transformers

diffusion
flow-matching
generative-models
transformer
vit
dit
architecture

From Vision Transformers to Diffusion Transformers: how DiT adapts ViT for image generation

Jan 17, 2026
Hujie Wang

Diffusion & Flow Matching Part 9: U-Net Architecture

diffusion
flow-matching
generative-models
u-net
architecture

U-Net architecture for learning vector fields and score functions in diffusion models

Jan 15, 2026
Hujie Wang
 

Diffusion & Flow Matching Part 8: Guided Generative Models

diffusion
flow-matching
generative-models
guidance
classifier-free-guidance

Conditional generation, classifier guidance, and classifier-free guidance for controllable image synthesis

Dec 20, 2025
Hujie Wang
 

Diffusion & Flow Matching Part 7: Training Targets and Algorithms

diffusion
flow-matching
generative-models
score-matching
training

The complete training algorithms for flow matching and diffusion models, including the DDPM loss and conversion formulas

Dec 14, 2025
Hujie Wang

Diffusion & Flow Matching Part 6: Conditional and Marginal Score Functions

diffusion
flow-matching
generative-models
score-function
score-matching
fokker-planck

How score functions connect ODEs to SDEs, and the elegant training trick that avoids intractable integrals

Dec 2, 2025
Hujie Wang

Diffusion & Flow Matching Part 5: Diffusion Models and SDEs

diffusion
flow-matching
generative-models
sde
In Parts 2-4 (flows, probability paths, training), we developed flow matching — a framework where a neural network learns a vector field \(u_t^\theta(x)\) that transports…
Dec 1, 2025
Hujie Wang

Diffusion & Flow Matching Part 4: The Flow Matching Loss

diffusion
flow-matching
generative-models
In the previous post, we constructed conditional probability paths and conditional vector fields. But there’s a fundamental problem: these are conditional on knowing the…
Nov 30, 2025
Hujie Wang

Diffusion & Flow Matching Part 3: Probability Paths

diffusion
flow-matching
generative-models
In the previous post, we learned that vector fields generate flows. Now we need to design a flow that transforms probability distributions — specifically, one that turns…
Nov 29, 2025
Hujie Wang

Diffusion & Flow Matching Part 2: Understanding Flows

diffusion
flow-matching
generative-models
In flow matching, we learn a vector field that transports samples from noise to data. But what exactly is a vector field? And what does it mean to “follow” one? This post…
Nov 28, 2025
Hujie Wang

Diffusion & Flow Matching Part 1: Introduction

diffusion
flow-matching
generative-models
This multi-part series explores the theory behind modern generative models. Each post builds on the previous one:
Nov 27, 2025
Timothy Wang
 

Welcome to My AI Blog

news
introduction
Welcome to my blog! This is where I’ll be sharing my thoughts and explorations on AI models.
Nov 27, 2025
Timothy Wang
No matching items