Deep Residual Learning for Image Recognition
Problem
Very deep neural networks were difficult to train, blocking accuracy gains from increased depth in visual recognition.
Why it matters
Depth of representations is of central importance to many visual recognition tasks.
What the AI did
Layers are reformulated to learn residual functions with reference to layer inputs, making substantially deeper networks easier to optimize and more accurate.
Results
Ensemble achieved 3.57% error on the ImageNet test set, winning 1st place in ILSVRC 2015 classification; 152-layer nets are 8× deeper than VGG with lower complexity.
Impact
Won ILSVRC 2015; extremely deep representations shown central to visual recognition.
Record
| Venue | IEEE Computer Society Conference on Computer Vision and Pattern Recognition |
|---|---|
| Published | 2016 |
| Work type | conference-paper |
| Authors | Kaiming He, Xiangyu Zhang, Shaoqing Ren, Jian Sun |
| Citations | 221,134 (OpenAlex, at import time) |
| References | 80 |
| Model or method | residual networks (up to 152 layers) |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1109/cvpr.2016.90 |
| Open access | green |
| Keywords | residual networks, deep learning, ImageNet, image recognition, ILSVRC |
Importance rating
major — millions of lives, an industry or a field
Abstract states depth of representations is of central importance to many visual recognition tasks; won ILSVRC 2015 and 28% relative COCO improvement, impacting the foundational computer vision field.
More in Software & Coding
| # | Paper | Year | Citations | Importance |
|---|---|---|---|---|
| 1 | ImageNet classification with deep convolutional neural networks | 2017 | 106,345 |
8
|
| 2 | Faster sorting algorithms discovered using deep reinforcement learning | 2023 | 175 |
8
|
| 3 | A deep learning‐based approach for software vulnerability detection using code metrics | 2022 | 42 |
7
|
| 4 | Deep Learning-Based Attack Detection and Classification in Android Devices | 2023 | 41 |
7
|
| 5 | Dota 2 with Large Scale Deep Reinforcement Learning | 2019 | 969 |
6
|