Differentiable Annealed Importance Sampling (DAIS)

This repository contains the code to reproduce the DAIS results from the paper Differentiable Annealed Importance Sampling and the Perils of Gradient Noise.

Bayesian Linear Regression

Please first install JAX (https://github.com/google/jax), then reproduce our results by running the following:

python blr/dais.py --gamma 0.0
python blr/dais.py --gamma 0.9
python blr.dais.py --bsize 100

Variational Autoencoder

Please first install PyTorch (https://pytorch.org/).

For training, one example command for DAIS with $K = 10$ and $S = 5$ (to adapt annealing scheme, add --adapt_beta):

python vae/mnist_train.py --lf_step 10 --n_particles 5 --lf_lrate 0.08

After training, you can find the saved checkpoint and evaluate it (e.g., using AIS):

python vae/mnist_eval.py --ais --n_particles 10 --lf_step 10000 --lf_lrate 0.05 --resume xxx

Also, you can reproduce the evaulation results of AIS/HAIS/DAIS by runing vae/mnist_eval_scaling.py:

python vae/mnist_eval_scaling.py --linear_beta --lf_lrate 0.08 --n_particles 10 --resume xxx

Citation

To cite this work, please use

@inproceedings{
	zhang2021differentiable,
	title={Differentiable Annealed Importance Sampling and the Perils of Gradient Noise},
	author={Guodong Zhang and Kyle Hsu and Jianing Li and Chelsea Finn and Roger Baker Grosse},
	booktitle={Thirty-Fifth Conference on Neural Information Processing Systems},
	year={2021},
	url={https://openreview.net/forum?id=6rqjgrL7Lq}
}

Differentiable Annealed Importance Sampling (DAIS)

Related tags

Overview

Differentiable Annealed Importance Sampling (DAIS)

Bayesian Linear Regression

Variational Autoencoder

Citation

Owner

Guodong Zhang

Solution to the first stage Quiz of Hamoye internship: Introduction to Python for Machine Learning

In the case of your data having only 1 channel while want to use timm models

Stochastic Downsampling for Cost-Adjustable Inference and Improved Regularization in Convolutional Networks

Python-based Informatics Kit for Analysing Chemical Units

Code for Talk-to-Edit (ICCV2021). Paper: Talk-to-Edit: Fine-Grained Facial Editing via Dialog.

ZSL-KG is a general-purpose zero-shot learning framework with a novel transformer graph convolutional network (TrGCN) to learn class representation from common sense knowledge graphs.

Region-aware Contrastive Learning for Semantic Segmentation, ICCV 2021

The first dataset of composite images with rationality score indicating whether the object placement in a composite image is reasonable.

Deeplab-resnet-101 in Pytorch with Jaccard loss

A minimalist implementation of score-based diffusion model

Official Pytorch Implementation of: "Semantic Diversity Learning for Zero-Shot Multi-label Classification"(2021) paper

O-CNN: Octree-based Convolutional Neural Networks for 3D Shape Analysis

Code for our SIGCOMM'21 paper "Network Planning with Deep Reinforcement Learning".

Machine Learning Platform for Kubernetes

Repo for parser tensorflow(.pb) and tflite(.tflite)

Code for technical report "An Improved Baseline for Sentence-level Relation Extraction".

Code for our paper "MG-GAN: A Multi-Generator Model Preventing Out-of-Distribution Samples in Pedestrian Trajectory Prediction" published at ICCV 2021.

Explaining in Style: Training a GAN to explain a classifier in StyleSpace

An end-to-end library for editing and rendering motion of 3D characters with deep learning [SIGGRAPH 2020]

Video Representation Learning by Recognizing Temporal Transformations. In ECCV, 2020.