zhaoyl18

AlignProp

AlignProp uses direct reward backpropogation for the alignment of large-scale text-to-image diffusion models. Our method is 25x more sample and compute efficient than reinforcement learning methods (PPO) for finetuning Stable Diffusion

bandit_sim

cos598d_pruning

Assignments for COS598D: System and Machine Learning

ddpo-jax

Code for the paper "Training Diffusion Models with Reinforcement Learning"

ddpo-pytorch

DDPO for finetuning diffusion models, implemented in PyTorch with LoRA support

Deep-PCA

DiGress

code for the paper "DiGress: Discrete Denoising diffusion for graph generation"

ratio_game

policy gradient methods for von Neumann's ratio game

SEIKO

SEIKO is a novel reinforcement learning method to efficiently fine-tune diffusion models in an online setting. Our methods outperform all baselines (PPO, classifier-based guidance, direct reward backpropagation) for fine-tuning Stable Diffusion.

zhaoyl18.github.io

zhaoyl18/SEIKO

SEIKO is a novel reinforcement learning method to efficiently fine-tune diffusion models in an online setting. Our methods outperform all baselines (PPO, classifier-based guidance, direct reward backpropagation) for fine-tuning Stable Diffusion.

zhaoyl18/ratio_game

policy gradient methods for von Neumann's ratio game

zhaoyl18/zhaoyl18.github.io

zhaoyl18/Deep-PCA

zhaoyl18/AlignProp

AlignProp uses direct reward backpropogation for the alignment of large-scale text-to-image diffusion models. Our method is 25x more sample and compute efficient than reinforcement learning methods (PPO) for finetuning Stable Diffusion

zhaoyl18/bandit_sim

zhaoyl18/cos598d_pruning

Assignments for COS598D: System and Machine Learning

zhaoyl18/ddpo-jax

Code for the paper "Training Diffusion Models with Reinforcement Learning"

zhaoyl18/ddpo-pytorch

DDPO for finetuning diffusion models, implemented in PyTorch with LoRA support

zhaoyl18/DiGress

code for the paper "DiGress: Discrete Denoising diffusion for graph generation"

zhaoyl18/GDPO

Graph Diffusion Policy Optimization

zhaoyl18/online_CDM

zhaoyl18/gReLU

zhaoyl18/mol_prop

zhaoyl18/MOOD

Official code repository for the paper Exploring Chemical Space with Score-based Out-of-distribution Generation (ICML 2023)

zhaoyl18/RCGDM

zhaoyl18/SVDD

Derivative-Free Guidance in Diffusion Models with Soft Value-Based Decoding. For controlled generation in DNA, RNA, proteins, molecules (+ images)

zhaoyl18/SVDD-image

Derivative-Free, Training-Free, Guidance in Diffusion Models