Implementations of paper Controlling Directions Orthogonal to a Classifier

Overview

Classifier Orthogonalization

Implementations of paper Controlling Directions Orthogonal to a Classifier , ICLR 2022,  Yilun Xu, Hao He, Tianxiao Shen, Tommi Jaakkola

Let's construct orthogonal classifiers for controlled style transfer, domain adaptation with label shifts and fairness problems 🤠 !

Outline

Controlled Style Transfer

Prepare CelebA-GH dataset:

python style_transfer/celeba_dataset.py --data_dir {path}

path: path to the CelebA dataset

bash example: python style_transfer/celeba_dataset.py --data_dir ./data

One can modify the domain_fn dictionary in the style_transfer/celeba_dataset.py file to create new groups 💡

Step 1: Train principal, full and oracle orthogonal classifiers

sh style_transfer/train_classifiers.sh {gpu} {path} {dataset} {alg}

gpu: the number of gpu
path: path to the dataset (Celeba or MNIST)
dataset: dataset (Celeba | CMNIST)
alg: ERM, Fish, TRM or MLDG

CMNIST bash example: sh style_transfer/train_classifiers.sh 0 ./data CMNIST ERM

Step 2: Train controlled CycleGAN

python style_transfer/train_cyclegan.py --data_dir {path} --dataset {dataset} \
  --obj {obj} --name {name}

path: path to the dataset (Celeba or MNIST)
dataset: dataset (Celeba | CMNIST)
obj: training objective (vanilla | orthogonal)
name: name of the model

CMNIST bash example: python style_transfer/train_cyclegan.py --data_dir ./data --dataset CMNIST --obj orthogonal --name cmnist

To view training results and loss plots, run python -m visdom.server and click the URL http://localhost:8097

Evaluation and Generation

python style_transfer/generate.py --data_dir {path} --dataset {dataset} --name {name} \
 --obj {obj} --out_path {out_path} --resume_epoch {epoch} (--save)

path: path to the dataset (Celeba or MNIST)
dataset: dataset (Celeba | CMNIST)
name: name of the model
obj: training objective (vanilla | orthogonal)
out_path: output path
epoch: resuming epoch of checkpoint

Images will be save to style_transfer/generated_images/out_path

CMNIST bash example: python style_transfer/generate.py --data_dir ./data --dataset CMNIST --name cmnist --obj orthogonal --out_path cmnist_out --resume_epoch 5


Domain Adaptation (DA) with label shifts

Prepare src/tgt pairs with label shifts

Please cd /da/data and run

python {dataset}.py --r {r0} {r1}

r0: subsample ratio for the first half classes (default=0.7)
r1: subsample ratio for the first half classes (default=0.3)
dataset: mnist | mnistm | svhn | cifar | stl | signs | digits

For SynthDigits / SynthSignsdataset, please download them at link_digits / link_signs. All the other datasets will be automatically downloaded 😉

Training

python da/vada_train.py --r {r0} {r1} --src {source} --tgt {target}  --seed {seed} \
 (--iw) (--orthogonal) (--source_only)

r0: subsample ratio for the first half classes (default=0.7)
r1: subsample ratio for the first half classes (default=0.3)
source: source domain (mnist | mnistm | svhn | cifar | stl | signs | digits)
target: target domain (mnist | mnistm | svhn | cifar | stl | signs | digits)
seed: random seed
--source_only: vanilla ERM on the source domain
--iw: use importance-weighted domain adaptation algorithm [1]
--orthogonal: use orthogonal classifier
--vada: vanilla VADA [2]

Fairness

python fairness/methods/train.py --data {data} --gamma {gamma} --sigma {sigma} \
 (--orthogonal) (--laftr) (--mifr) (--hsic)

data: dataset (adult | german)
gamma: hyper-parameter for MIFR, HSIC, LAFTR
sigma: hyper-parameter for HSIC (kernel width)
--orthogonal: use orthogonal classifier
--MIFR: use L-MIFR algorithm [3]
--HSIC: use ReBias algorithm [4]
--LAFTR: use LAFTR algorithm [5]



Reference

[1] Remi Tachet des Combes, Han Zhao, Yu-Xiang Wang, and Geoffrey J. Gordon. Domain adaptation with conditional distribution matching and generalized label shift. ArXiv, abs/2003.04475, 2020.

[2] Rui Shu, H. Bui, H. Narui, and S. Ermon. A dirt-t approach to unsupervised domain adaptation. ArXiv, abs/1802.08735, 2018.

[3] Jiaming Song, Pratyusha Kalluri, Aditya Grover, Shengjia Zhao, and S. Ermon. Learning controllable fair representations. In AISTATS, 2019.

[4] Hyojin Bahng, Sanghyuk Chun, Sangdoo Yun, Jaegul Choo, and Seong Joon Oh. Learning de-biased representations with biased representations. In ICML, 2020.

[5] David Madras, Elliot Creager, T. Pitassi, and R. Zemel. Learning adversarially fair and transferable representations. In ICML, 2018.


The implementation of this repo is based on / inspired by:

Owner
Yilun Xu
Hello!
Yilun Xu
Additional environments compatible with OpenAI gym

Decentralized Control of Quadrotor Swarms with End-to-end Deep Reinforcement Learning A codebase for training reinforcement learning policies for quad

Zhehui Huang 40 Dec 06, 2022
LV-BERT: Exploiting Layer Variety for BERT (Findings of ACL 2021)

LV-BERT Introduction In this repo, we introduce LV-BERT by exploiting layer variety for BERT. For detailed description and experimental results, pleas

Weihao Yu 14 Aug 24, 2022
Transfer style api - An API to use with Tranfer Style App, where you can use two image and transfer the style

Transfer Style API It's an API to use with Tranfer Style App, where you can use

Brian Alejandro 1 Feb 13, 2022
Deep Q Learning with OpenAI Gym and Pokemon Showdown

pokemon-deep-learning An openAI gym project for pokemon involving deep q learning. Made by myself, Sam Little, and Layton Webber. This code captures g

2 Dec 22, 2021
Model that predicts the probability of a Twitter user being anti-vaccination.

stylebody {text-align: justify}/style AVAXTAR: Anti-VAXx Tweet AnalyzeR AVAXTAR is a python package to identify anti-vaccine users on twitter. The

10 Sep 27, 2022
DRLib:A concise deep reinforcement learning library, integrating HER and PER for almost off policy RL algos.

DRLib:A concise deep reinforcement learning library, integrating HER and PER for almost off policy RL algos A concise deep reinforcement learning libr

329 Jan 03, 2023
Transfer Reinforcement Learning for Differing Action Spaces via Q-Network Representations

Transfer-Learning-in-Reinforcement-Learning Transfer Reinforcement Learning for Differing Action Spaces via Q-Network Representations Final Report Tra

Trung Hieu Tran 4 Oct 17, 2022
Implementing SYNTHESIZER: Rethinking Self-Attention in Transformer Models using Pytorch

Implementing SYNTHESIZER: Rethinking Self-Attention in Transformer Models using Pytorch Reference Paper URL Author: Yi Tay, Dara Bahri, Donald Metzler

Myeongjun Kim 66 Nov 30, 2022
Look Who’s Talking: Active Speaker Detection in the Wild

Look Who's Talking: Active Speaker Detection in the Wild Dependencies pip install -r requirements.txt In addition to the Python dependencies, ffmpeg

Clova AI Research 60 Dec 08, 2022
Image-retrieval-baseline - MUGE Multimodal Retrieval Baseline

MUGE Multimodal Retrieval Baseline This repo is implemented based on the open_cl

47 Dec 16, 2022
A Simple LSTM-Based Solution for "Heartbeat Signal Classification and Prediction" in Tianchi

LSTM-Time-Series-Prediction A Simple LSTM-Based Solution for "Heartbeat Signal Classification and Prediction" in Tianchi Contest. The Link of the Cont

KevinCHEN 1 Jun 13, 2022
Identifying a Training-Set Attack’s Target Using Renormalized Influence Estimation

Identifying a Training-Set Attack’s Target Using Renormalized Influence Estimation By: Zayd Hammoudeh and Daniel Lowd Paper: Arxiv Preprint Coming soo

Zayd Hammoudeh 2 Oct 08, 2022
A mini-course offered to Undergrad chemistry students

The best way to use this material is by forking it by click the Fork button at the top, right corner. Then you will get your own copy to play with! Th

Raghu 19 Dec 19, 2022
No Code AI/ML platform

NoCodeAIML No Code AI/ML platform - Community Edition Video credits: Uday Kiran Typical No Code AI/ML Platform will have features like drag and drop,

Bhagvan Kommadi 5 Jan 28, 2022
Cockpit is a visual and statistical debugger specifically designed for deep learning.

Cockpit: A Practical Debugging Tool for Training Deep Neural Networks

Felix Dangel 421 Dec 29, 2022
Convolutional 2D Knowledge Graph Embeddings resources

ConvE Convolutional 2D Knowledge Graph Embeddings resources. Paper: Convolutional 2D Knowledge Graph Embeddings Used in the paper, but do not use thes

Tim Dettmers 586 Dec 24, 2022
Pytorch implementation of the paper Progressive Growing of Points with Tree-structured Generators (BMVC 2021)

PGpoints Pytorch implementation of the paper Progressive Growing of Points with Tree-structured Generators (BMVC 2021) Hyeontae Son, Young Min Kim Pre

Hyeontae Son 9 Jun 06, 2022
Code for the tech report Toward Training at ImageNet Scale with Differential Privacy

Differentially private Imagenet training Code for the tech report Toward Training at ImageNet Scale with Differential Privacy by Alexey Kurakin, Steve

Google Research 29 Nov 03, 2022
Differentiable molecular simulation of proteins with a coarse-grained potential

Differentiable molecular simulation of proteins with a coarse-grained potential This repository contains the learned potential, simulation scripts and

UCL Bioinformatics Group 44 Dec 10, 2022
Compute FID scores with PyTorch.

FID score for PyTorch This is a port of the official implementation of Fréchet Inception Distance to PyTorch. See https://github.com/bioinf-jku/TTUR f

2.1k Jan 06, 2023