Code of the paper "Multi-Task Meta-Learning Modification with Stochastic Approximation".

Overview

Multi-Task Meta-Learning Modification with Stochastic Approximation

This repository contains the code for the paper
"Multi-Task Meta-Learning Modification with Stochastic Approximation".

Method pipeline

Dependencies

This code has been tested on Ubuntu 16.04 with Python 3.8 and PyTorch 1.8.

To install the required dependencies:

pip install -r requirements.txt

Usage

To reproduce the results on benchmarks described in our article, use the following scripts. To vary types of the experiments, change the parameters of the scripts responsible for benchmark dataset, shot and way (e.g. miniImageNet 1-shot 5-way or CIFAR-FS 5-shot 2-way).

MAML

Multi-task modification (MTM) for Model-Agnostic Meta-Learning (MAML) (Finn et al., 2017).

Multi-task modifications for MAML are trained on top of baseline MAML model which has to be trained beforehand.

To train MAML (reproduced) on miniImageNet 1-shot 2-way benchmark, run:

python maml/train.py ./datasets/ \
    --run-name reproduced-miniimagenet \
    --dataset miniimagenet \
    --num-ways 2 \
    --num-shots 1 \
    --num-steps 5 \
    --num-epochs 300 \
    --use-cuda \
    --output-folder ./results

To train MAML MTM SPSA-Track on miniImageNet 1-shot 2-way benchmark, run:

python maml/train.py ./datasets/ \
    --run-name mini-imagenet-mtm-spsa-track \
    --load "./results/reproduced-miniimagenet/model.th" \
    --dataset miniimagenet \
    --num-ways 2 \
    --num-shots 1 \
    --num-steps 5 \
    --task-weighting spsa-track \
    --normalize-spsa-weights-after 100 \
    --num-epochs 40 \
    --use-cuda \
    --output-folder ./results

To train MAML (reproduced) on tieredImageNet 1-shot 2-way benchmark, run:

python maml/train.py ./datasets/ \
    --run-name reproduced-tieredimagenet \
    --dataset tieredimagenet \
    --num-ways 2 \
    --num-shots 1 \
    --num-steps 5 \
    --num-epochs 300 \
    --use-cuda \
    --output-folder ./results

To train MAML MTM SPSA on tieredImageNet 1-shot 2-way benchmark, run:

python maml/train.py ./datasets/ \
    --run-name tiered-imagenet-mtm-spsa \
    --load "./results/reproduced-tieredimagenet/model.th" \
    --dataset tieredimagenet \
    --num-ways 2 \
    --num-shots 1 \
    --num-steps 5 \
    --task-weighting spsa-delta \
    --normalize-spsa-weights-after 100 \
    --num-epochs 40 \
    --use-cuda \
    --output-folder ./results

To train MAML (reproduced) on FC100 5-shot 5-way benchmark, run:

python maml/train.py ./datasets/ \
    --run-name reproduced-fc100 \
    --dataset fc100 \
    --num-ways 5 \
    --num-shots 5 \
    --num-steps 5 \
    --num-epochs 300 \
    --use-cuda \
    --output-folder ./results

To train MAML MTM SPSA-Coarse on FC100 5-shot 5-way benchmark, run:

python maml/train.py ./datasets/ \
    --run-name fc100-mtm-spsa-coarse \
    --load "./results/reproduced-fc100/model.th" \
    --dataset fc100 \
    --num-ways 5 \
    --num-shots 5 \
    --num-steps 5 \
    --task-weighting spsa-per-coarse-class \
    --num-epochs 40 \
    --use-cuda \
    --output-folder ./results

To train MAML (reproduced) on CIFAR-FS 1-shot 5-way benchmark, run:

python maml/train.py ./datasets/ \
    --run-name reproduced-cifar \
    --dataset cifarfs \
    --num-ways 5 \
    --num-shots 1 \
    --num-steps 5 \
    --num-epochs 600 \
    --use-cuda \
    --output-folder ./results

To train MAML MTM Inner First-Order on CIFAR-FS 1-shot 5-way benchmark, run:

python maml/train.py ./datasets/ \
    --run-name cifar-mtm-inner-first-order \
    --load "./results/reproduced-cifar/model.th" \
    --dataset cifarfs \
    --num-ways 5 \
    --num-shots 1 \
    --num-steps 5 \
    --task-weighting gradient-novel-loss \
    --use-inner-optimizer \
    --num-epochs 40 \
    --use-cuda \
    --output-folder ./results

To train MAML MTM Backprop on CIFAR-FS 1-shot 5-way benchmark, run:

python maml/train.py ./datasets/ \
    --run-name cifar-mtm-backprop \
    --load "./results/reproduced-cifar-5shot-5way/model.th" \
    --dataset cifarfs \
    --num-ways 5 \
    --num-shots 1 \
    --num-steps 5 \
    --task-weighting gradient-novel-loss \
    --num-epochs 40 \
    --use-cuda \
    --output-folder ./results

To test any of the above-described benchmarks, run:

python maml/test.py ./results/path-to-config/config.json --num-steps 10 --use-cuda

For instance, to test MAML MTM SPSA-Track on miniImageNet 1-shot 2-way benchmark, run:

python maml/test.py ./results/mini-imagenet-mtm-spsa-track/config.json --num-steps 10 --use-cuda

Prototypical Networks

Multi-task modification (MTM) for Prototypical Networks (ProtoNet) (Snell et al., 2017).

To train ProtoNet MTM SPSA-Track with ResNet-12 backbone on miniImageNet 1-shot 5-way benchmark, run:

python protonet/train.py \
    --dataset miniImageNet \
    --network ResNet12 \
    --tracking \
    --train-shot 1 \
    --train-way 5 \
    --val-shot 1 \
    --val-way 5

To test ProtoNet MTM SPSA-Track with ResNet-12 backbone on miniImageNet 1-shot 5-way benchmark, run:

python protonet/test.py --dataset miniImageNet --network ResNet12 --shot 1 --way 5

To train ProtoNet MTM Backprop with 64-64-64-64 backbone on CIFAR-FS 1-shot 2-way benchmark, run:

python protonet/train.py \
    --dataset CIFAR_FS \
    --train-weights \
    --train-weights-layer \
    --train-shot 1 \
    --train-way 2 \
    --val-shot 1 \
    --val-way 2

To test ProtoNet MTM Backprop with 64-64-64-64 backbone on CIFAR-FS 1-shot 5-way benchmark, run:

python protonet/test.py --dataset CIFAR_FS --shot 1 --way 2

To train ProtoNet MTM Inner First-Order with 64-64-64-64 backbone on FC100 10-shot 5-way benchmark, run:

python protonet/train.py \
    --dataset FC100 \
    --train-weights \
    --train-weights-opt \
    --train-shot 10 \
    --train-way 5 \
    --val-shot 10 \
    --val-way 5

To test ProtoNet MTM Inner First-Order with 64-64-64-64 backbone on FC100 10-shot 5-way benchmark, run:

python protonet/test.py --dataset FC100 --shot 10 --way 5

To train ProtoNet MTM SPSA with 64-64-64-64 backbone on tieredImageNet 5-shot 2-way benchmark, run:

python protonet/train.py \
    --dataset tieredImageNet \
    --train-shot 5 \
    --train-way 2 \
    --val-shot 5 \
    --val-way 2

To test ProtoNet MTM SPSA with 64-64-64-64 backbone on tieredImageNet 5-shot 2-way benchmark, run:

python protonet/test.py --dataset tieredImageNet --shot 5 --way 2

Acknowledgments

Our code uses some dataloaders from Torchmeta.

Code in maml folder is based on the extended implementation from Torchmeta and pytorch-maml. The code has been updated so that baseline scores more closely follow those of the original MAML paper.

Code in protonet folder is based on the implementation from MetaOptNet. All .py files in this folder except for dataloaders.py and optimize.py were adopted from this implementation and modified afterwards. A copy of Apache License, Version 2.0 is available in protonet folder.

Owner
Andrew
Andrew
Code for HLA-Face: Joint High-Low Adaptation for Low Light Face Detection (CVPR21)

HLA-Face: Joint High-Low Adaptation for Low Light Face Detection The official PyTorch implementation for HLA-Face: Joint High-Low Adaptation for Low L

Wenjing Wang 77 Dec 08, 2022
TLDR; Train custom adaptive filter optimizers without hand tuning or extra labels.

AutoDSP TLDR; Train custom adaptive filter optimizers without hand tuning or extra labels. About Adaptive filtering algorithms are commonplace in sign

Jonah Casebeer 48 Sep 19, 2022
Powerful and efficient Computer Vision Annotation Tool (CVAT)

Computer Vision Annotation Tool (CVAT) CVAT is free, online, interactive video and image annotation tool for computer vision. It is being used by our

OpenVINO Toolkit 8.6k Jan 01, 2023
NeRF Meta-Learning with PyTorch

NeRF Meta Learning With PyTorch nerf-meta is a PyTorch re-implementation of NeRF experiments from the paper "Learned Initializations for Optimizing Co

Sanowar Raihan 78 Dec 18, 2022
[ICCV 2021] FaPN: Feature-aligned Pyramid Network for Dense Image Prediction

FaPN: Feature-aligned Pyramid Network for Dense Image Prediction [arXiv] [Project Page] @inproceedings{ huang2021fapn, title={{FaPN}: Feature-alig

Shihua Huang 23 Jul 22, 2022
PIGLeT: Language Grounding Through Neuro-Symbolic Interaction in a 3D World [ACL 2021]

piglet PIGLeT: Language Grounding Through Neuro-Symbolic Interaction in a 3D World [ACL 2021] This repo contains code and data for PIGLeT. If you like

Rowan Zellers 51 Oct 08, 2022
This project is based on RIFE and aims to make RIFE more practical for users by adding various features and design new models

CPM 项目描述 CPM(Chinese Pretrained Models)模型是北京智源人工智能研究院和清华大学发布的中文大规模预训练模型。官方发布了三种规模的模型,参数量分别为109M、334M、2.6B,用户需申请与通过审核,方可下载。 由于原项目需要考虑大模型的训练和使用,需要安装较为复杂

hzwer 190 Jan 08, 2023
내가 보려고 정리한 <프로그래밍 기초 Ⅰ> / organized for me

Programming-Basics 프로그래밍 기초 Ⅰ 아카이브 Do it! 점프 투 파이썬 주차 강의주제 비고 1주차 Syllabus 2주차 자료형 - 숫자형 3주차 자료형 - 문자열형 4주차 입력과 출력 5주차 제어문 - 조건문 if 6주차 제어문 - 반복문 whil

KIMMINSEO 1 Mar 07, 2022
Pytorch Implementation of Auto-Compressing Subset Pruning for Semantic Image Segmentation

Pytorch Implementation of Auto-Compressing Subset Pruning for Semantic Image Segmentation Introduction ACoSP is an online pruning algorithm that compr

Merantix 8 Dec 07, 2022
CR-FIQA: Face Image Quality Assessment by Learning Sample Relative Classifiability

This is the official repository of the paper: CR-FIQA: Face Image Quality Assessment by Learning Sample Relative Classifiability A private copy of the

Fadi Boutros 33 Dec 31, 2022
Weakly Supervised Learning of Instance Segmentation with Inter-pixel Relations, CVPR 2019 (Oral)

Weakly Supervised Learning of Instance Segmentation with Inter-pixel Relations The code of: Weakly Supervised Learning of Instance Segmentation with I

Jiwoon Ahn 472 Dec 29, 2022
Code and Data for the paper: Molecular Contrastive Learning with Chemical Element Knowledge Graph [AAAI 2022]

Knowledge-enhanced Contrastive Learning (KCL) Molecular Contrastive Learning with Chemical Element Knowledge Graph [ AAAI 2022 ]. We construct a Chemi

Fangyin 58 Dec 26, 2022
SuperSDR: multiplatform KiwiSDR + CAT transceiver integrator

SuperSDR SuperSDR integrates a realtime spectrum waterfall and audio receive from any KiwiSDR around the world, together with a local (or remote) cont

Marco Cogoni 30 Nov 29, 2022
A wrapper around SageMaker ML Lineage Tracking extending ML Lineage to end-to-end ML lifecycles, including additional capabilities around Feature Store groups, queries, and other relevant artifacts.

ML Lineage Helper This library is a wrapper around the SageMaker SDK to support ease of lineage tracking across the ML lifecycle. Lineage artifacts in

AWS Samples 12 Nov 01, 2022
ETMO: Evolutionary Transfer Multiobjective Optimization

ETMO: Evolutionary Transfer Multiobjective Optimization To promote the research on ETMO, benchmark problems are of great importance to ETMO algorithm

Songbai Liu 0 Mar 16, 2021
Geometric Vector Perceptron --- a rotation-equivariant GNN for learning from biomolecular structure

Geometric Vector Perceptron Code to accompany Learning from Protein Structure with Geometric Vector Perceptrons by B Jing, S Eismann, P Suriana, RJL T

Dror Lab 85 Dec 29, 2022
DM-ACME compatible implementation of the Arm26 environment from Mujoco

ACME-compatible implementation of Arm26 from Mujoco This repository contains a customized implementation of Mujoco's Arm26 model, that can be used wit

1 Dec 24, 2021
Datasets and pretrained Models for StyleGAN3 ...

Datasets and pretrained Models for StyleGAN3 ... Dear arfiticial friend, this is a collection of artistic datasets and models that we have put togethe

lucid layers 34 Oct 06, 2022
PaddleBoBo是基于PaddlePaddle和PaddleSpeech、PaddleGAN等开发套件的虚拟主播快速生成项目

PaddleBoBo - 元宇宙时代,你也可以动手做一个虚拟主播。 PaddleBoBo是基于飞桨PaddlePaddle深度学习框架和PaddleSpeech、PaddleGAN等开发套件的虚拟主播快速生成项目。PaddleBoBo致力于简单高效、可复用性强,只需要一张带人像的图片和一段文字,就能

502 Jan 08, 2023
[ICCV2021] IICNet: A Generic Framework for Reversible Image Conversion

IICNet - Invertible Image Conversion Net Official PyTorch Implementation for IICNet: A Generic Framework for Reversible Image Conversion (ICCV2021). D

felixcheng97 55 Dec 06, 2022