Early-Exit-Models

This repository is dedicated to self-learning about early exit models, including relevant code and documentation.

Survey Papers	code	Comments
1. Split Computing and Early Exiting for Deep Learning Applications: Survey and Research Challenges		Fundamental Survey Paper Section 4
2. Adaptive Inference through Early-Exit Networks: Design, Challenges and Directions
3. Distributed Artificial Intelligence Empowered by End-Edge-Cloud Computing: A Survey		Section III-C-4
4. End-Edge-Cloud Collaborative Computing for Deep Learning: A Comprehensive Survey		Section III-B
5.Advancements in Accelerating Deep Neural Network Inference on AIoT Devices: A Survey
6.Resource Management in Mobile Edge Computing: A Comprehensive Survey

Papers	code	Comments
1. BranchyNet: Fast Inference via Early Exiting from Deep Neural Networks	Official Code code	Fundamental Paper
2.Distributed Deep Neural Networks over the Cloud, the Edge and End Devices	Offical Code	Follow up work, node-edge-cloud setting
3.Multi-Scale Dense Networks for Resource Efficient Image Classification	Official Code pytorch	MSDNet(ICLR)
4.Branchy-GNN: a Device-Edge Co-Inference Framework for Efficient Point Cloud Processing	Offical Code	GNN
5.EdgeKE: An On-Demand Deep Learning IoT System for Cognitive Big Data on Industrial Edge Devices	Offical Code	knowledge distillation, early exit to meet latency or accuracy requirements
6.SPINN: Synergistic Progressive Inference of Neural Networks over Device and Cloud		run-time scheduler(Mobicom)
7.FlexDNN: Input-Adaptive On-Device Deep Learning for Efficient Mobile Vision
8.Boomerang: On-demand cooperative deep neural network inference for edge intelligence on the industrial internet of things
9.Early-exit deep neural networks for distorted images: providing an efficient edge offloading	Offical Code	early-exit DNN with expert branches
10.FrameExit: Conditional Early Exiting for Efficient Video Recognition	Offical Code	gating module (CVPR)
11.BERxiT: Early exiting for BERT with better fine-tuning and extension to regression	Offical Code	(ACL)
12.A lightweight collaborative deep neural network for the mobile web in edge cloud		Binary neural network branch
13.A Lightweight Collaborative Recognition System with Binary Convolutional Neural Network for Mobile Web Augmented Reality		Binary neural network branch
14.DNN Inference Acceleration with Partitioning and Early Exiting in Edge Computing
15.Edge Intelligence: On-Demand Deep Learning Model Co-Inference with Device-Edge Synergy (Edgent) Edge AI: On-Demand Accelerating Deep Neural Network Inference via Edge Computing		partitions DNN computation between mobile and edge server based on the available bandwidth
16.Improved Techniques for Training Adaptive Deep Networks	Offical Code	(ICCV)
17.Learning Anytime Predictions in Neural Networks via Adaptive Loss Balancing		(AAAI)
18.Learning to Stop While Learning to Predict	Offical Code	(ICML)
19.DeepAdapter: A Collaborative Deep Learning Framework for the Mobile Web Using Context-Aware Network Pruning		Follow up work of Edgent, online inference
20.Branching in Deep Networks for Fast Inference
21.Accelerating on-device DNN inference during service outage through scheduling early exit
22.Learning Early Exit for Deep Neural Network Inference on Mobile Devices through Multi-Armed Bandits
23.Cloudedge-based lightweight temporal convolutional networks for remaining useful life prediction in IIoT		two scale prediction
24.Predictive Exit: Prediction of Fine-Grained Early Exits for Computation- and Energy-Efficient Inference		(AAAI)
25.DeeCap: Dynamic Early Exiting for Efficient Image Captioning	Offical Code
26.A Simple Hash-Based Early Exiting Approach For Language Understanding and Generation	Offical Code	(ACL)
27.It's always personal: Using Early Exits for Efficient On-Device CNN Personalisation
28.Class-specific early exit design methodology for convolutional neural networks
29.Federated Learning for Cooperative Inference Systems: The Case of Early Exit Networks		Cooperative Inference Systems settings
30.Dual Dynamic Inference: Enabling More Efficient, Adaptive, and Controllable Deep Inference		Channel with Early Exit
31.Multi-Exit Semantic Segmentation Networks		(ECCV)
32.Multi-Exit DNN Inference Acceleration Based on Multi-Dimensional Optimization for Edge Intelligence		a contextual bandit learning that learns the optimal partition point
33.Accelerating on-device DNN inference during service outage through scheduling early exit
34.Autodidactic Neurosurgeon: Collaborative Deep Inference for Mobile Edge Intelligence via Online Learning
35.Towards Edge Computing Using Early-Exit Convolutional Neural Networks		MobiletNetV2 with early exits
36.Resource-Constrained Edge AI with Early Exit Prediction
37.Temporal Decisions: Leveraging Temporal Correlation for Efficient Decisions in Early Exit Neural Networks
38.Efficient Post-Training Augmentation for Adaptive Inference in Heterogeneous and Distributed IoT Environments
39.DyCE: Dynamic Configurable Exiting for Deep Learning Compression and Scaling
40.ClassyNet: Class-Aware Early-Exit Neural Networks for Edge Devices
41.ENASFL: A Federated Neural Architecture Search Scheme for Heterogeneous Deep Models in Distributed Edge Computing Systems
42.Adaptive Early Exiting for Collaborative Inference over Noisy Wireless Channels
43.EdgeFM: Leveraging Foundation Model for Open-set Learning on the Edge
44.Channel-Adaptive Early Exiting using Reinforcement Learning for Multivariate Time Series Classification
45.SplitEE: Early Exit in Deep Neural Networks with Split Computing	Offical Code
46.Branchy Deep Learning Based Real-Time Defect Detection Under Edge-Cloud Fusion Architecture
47.Joint multi-user DNN partitioning and task offloading in mobile edge computing
48.Resource-aware Deployment of Dynamic DNNs over Multi-tiered Interconnected Systems
49.Edge Computing with Early Exiting for Adaptive Inference in Mobile Autonomous Systems

Repos on Early Exit

Papers	code	Comments
	Repo
	Repo

AI on Edge

Papers	code	Comments
Enabling AI on Edges: Techniques, Applications and Challenges	Offical Code
Green Edge AI: A Contemporary Survey

Early Exit inference model from Fig.7 in Split Computing and Early Exiting for Deep Learning Applications: Survey and Research Challenges, YOSHITOMO MATSUBARA and MARCO LEVORATO, University of California, Irvine, USA, FRANCESCO RESTUCCIA, Northeastern University, USA