Awesome Transformer Architecture Search:

To keep track of the large number of recent papers that look at the intersection of Transformers and Neural Architecture Search (NAS), we have created this awesome list of curated papers and resources, inspired by awesome-autodl, awesome-architecture-search, and awesome-computer-vision. Papers are divided into the following categories:

General Transformer search
Domain Specific, applied Transformer search (divided into NLP, Vision, ASR)
Insights on Transformer components or searchable parameters
Transformer Surveys

This repository is maintained by the AutoML Group Freiburg. Please feel free to pull requests or open an issue to add papers.

General Transformer Search

Title	Venue	Group
UniNet: Unified Architecture Search with Convolutions, Transformer and MLP	arxiv [Oct'21]	SenseTime
Analyzing and Mitigating Interference in Neural Architecture Search	arxiv [Aug'21]	Tsinghua, MSR
BossNAS: Exploring Hybrid CNN-transformers with Block-wisely Self-supervised Neural Architecture Search	ICCV'21	Sun Yat-sen University
Memory-Efficient Differentiable Transformer Architecture Search	ACL-IJCNLP'21	MSR, Peking University
Finding Fast Transformers: One-Shot Neural Architecture Search by Component Composition	arxiv [Aug'20]	Google Research
AutoTrans: Automating Transformer Design via Reinforced Architecture Search	arxiv [Sep'20]	Fudan University
NAT: Neural Architecture Transformer for Accurate and Compact Architectures	NeurIPS'19	Tencent AI
The Evolved Transformer	ICML'19	Google Brain

Domain Specific Transformer Search

Vision

Title	Venue	Group
AutoFormer: Searching Transformers for Visual Recognition	ICCV'21	MSR
GLiT: Neural Architecture Search for Global and Local Image Transformer	ICCV'21	University of Sydney
Searching for Efficient Multi-Stage Vision Transformers	ICCV'21 workshop	MIT
HR-NAS: Searching Efficient High-Resolution Neural Architectures with Lightweight Transformers	CVPR'21	Bytedance Inc.
Vision Transformer Architecture Search	arxiv [June'21]	SenseTime, Tsingua University

Natural Language Processing

Title	Venue	Group
AutoTinyBERT: Automatic Hyper-parameter Optimization for Efficient Pre-trained Language Models	ACL'21	MIT
NAS-BERT: Task-Agnostic and Adaptive-Size BERT Compression with Neural Architecture Search	KDD'21	MSR, Tsinghua University
AutoBERT-Zero: Evolving the BERT backbone from scratch	arxiv [July'21]	Huawei Noah’s Ark Lab
HAT: Hardware-Aware Transformers for Efficient Natural Language Processing	ACL'20	MIT

Automatic Speech Recognition

Title	Venue	Group
LightSpeech: Lightweight and Fast Text to Speech with Neural Architecture Search	ICASSP'21	MSR
Darts-Conformer: Towards Efficient Gradient-Based Neural Architecture Search For End-to-End ASR	arxiv [Aug'21]	NPU, Xi'an
Improved Conformer-based End-to-End Speech Recognition Using Neural Architecture Search	arxiv [April'21]	Chinese Academy of Sciences
Evolved Speech-Transformer: Applying Neural Architecture Search to End-to-End Automatic Speech Recognition	INTERSPEECH'20	VUNO Inc.

Insights on Transformer components and interesting papers

Title	Venue	Group
Patches are All You Need ?	ICLR'22 under review	-
Swin Transformer: Hierarchical Vision Transformer using Shifted Windows	ICCV'21 best paper	MSR
Rethinking Spatial Dimensions of Vision Transformers	ICCV'21	NAVER AI
What makes for hierarchical vision transformers	arxiv [Sept'21]	HUST
AutoAttend: Automated Attention Representation Search	ICML'21	Tsinghua University
Rethinking Attention with Performers	ICLR'21 Oral	Google
LambdaNetworks: Modeling long-range Interactions without Attention	ICLR'21	Google Research
HyperGrid Transformers	ICLR'21	Google Research
LocalViT: Bringing Locality to Vision Transformers	arxiv [April'21]	ETH Zurich
NASABN: A Neural Architecture Search Framework for Attention-Based Networks	IJCNN'20	Chinese Academy of Sciences
Analyzing Multi-Head Self-Attention: Specialized Heads Do the Heavy Lifting, the Rest Can Be Pruned	ACL'19	Yandex

Transformer Surveys

Title	Venue	Group
Transformers in Vision: A Survey	arxiv [Oct'21]	MBZ University of AI
Efficient Transformers: A Survey	arxiv [Sept'21]	Google Research

Misc resources

Awesome Visual Transformer

A curated list of awesome resources combining Transformers with Neural Architecture Search

Related tags

Overview

Awesome Transformer Architecture Search:

General Transformer Search

Domain Specific Transformer Search

Vision

Natural Language Processing

Automatic Speech Recognition

Insights on Transformer components and interesting papers

Transformer Surveys

Misc resources

Owner

Yash Mehta

The official repo of the CVPR2021 oral paper: Representative Batch Normalization with Feature Calibration

[NeurIPS 2021] Deceive D: Adaptive Pseudo Augmentation for GAN Training with Limited Data

Python project to take sound as input and output as RGB + Brightness values suitable for DMX

Traditional deepdream with VQGAN+CLIP and optical flow. Ready to use in Google Colab

PyTorch code for the ICCV'21 paper: "Always Be Dreaming: A New Approach for Class-Incremental Learning"

Build tensorflow keras model pipelines in a single line of code. Created by Ram Seshadri. Collaborators welcome. Permission granted upon request.

An efficient and effective learning to rank algorithm by mining information across ranking candidates. This repository contains the tensorflow implementation of SERank model. The code is developed based on TF-Ranking.

An introduction to satellite image analysis using Python + OpenCV and JavaScript + Google Earth Engine

本步态识别系统主要基于GaitSet模型进行实现

SenseNet is a sensorimotor and touch simulator for deep reinforcement learning research

Towards Fine-Grained Reasoning for Fake News Detection

Investigating automatic navigation towards standard US views integrating MARL with the virtual US environment developed in CT2US simulation

Efficient 3D human pose estimation in video using 2D keypoint trajectories

Code for the Image similarity challenge.

Implementation of "RaScaNet: Learning Tiny Models by Raster-Scanning Image" from CVPR 2021.

Action Segmentation Evaluation

The repository for the paper "When Do You Need Billions of Words of Pretraining Data?"

QuakeLabeler is a Python package to create and manage your seismic training data, processes, and visualization in a single place — so you can focus on building the next big thing.

Learning Dynamic Network Using a Reuse Gate Function in Semi-supervised Video Object Segmentation.

Embracing Single Stride 3D Object Detector with Sparse Transformer