# Stochastic Backpropagation: A Memory Efficient Strategy for Training Video Models

arXiv: https://arxiv.org/abs/2203.16755 · HTML: https://mlsanity.com/paper/2203.16755/

## Key facts

- Cited by 14 papers in the map (data through 14 Oct 2026).
- Builds on 49 of its 54 references that are in the map.
- First posted on arXiv 31 Mar 2022.

## Builds on (most cited first, 30 of 49)

- [Deep Residual Learning for Image Recognition](https://mlsanity.com/paper/1512.03385/) (2015)
- [An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale](https://mlsanity.com/paper/2010.11929/) (2020)
- [Quo Vadis, Action Recognition? A New Model and the Kinetics Dataset](https://mlsanity.com/paper/1705.07750/) (2017)
- [The Kinetics Human Action Video Dataset](https://mlsanity.com/paper/1705.06950/) (2017)
- [Transformer-XL: Attentive Language Models Beyond a Fixed-Length Context](https://mlsanity.com/paper/1901.02860/) (2019)
- [Similarity of Neural Network Representations Revisited](https://mlsanity.com/paper/1905.00414/) (2019)
- [Is Space-Time Attention All You Need for Video Understanding?](https://mlsanity.com/paper/2102.05095/) (2021)
- [ViViT: A Video Vision Transformer](https://mlsanity.com/paper/2103.15691/) (2021)
- [A Closer Look at Spatiotemporal Convolutions for Action Recognition](https://mlsanity.com/paper/1711.11248/) (2017)
- [Video Swin Transformer](https://mlsanity.com/paper/2106.13230/) (2021)
- [Two-Stream Convolutional Networks for Action Recognition in Videos](https://mlsanity.com/paper/1406.2199/) (2014)
- [Temporal Segment Networks: Towards Good Practices for Deep Action Recognition](https://mlsanity.com/paper/1608.00859/) (2016)
- [Training Deep Nets with Sublinear Memory Cost](https://mlsanity.com/paper/1604.06174/) (2016)
- [Scaling Egocentric Vision: The EPIC-KITCHENS Dataset](https://mlsanity.com/paper/1804.02748/) (2018)
- [Real-world Anomaly Detection in Surveillance Videos](https://mlsanity.com/paper/1801.04264/) (2018)
- [VATT: Transformers for Multimodal Self-Supervised Learning from Raw Video, Audio and Text](https://mlsanity.com/paper/2104.11178/) (2021)
- [BMN: Boundary-Matching Network for Temporal Action Proposal Generation](https://mlsanity.com/paper/1907.09702/) (2019)
- [BSN: Boundary Sensitive Network for Temporal Action Proposal Generation](https://mlsanity.com/paper/1806.02964/) (2018)
- [Temporal Action Detection with Structured Segment Networks](https://mlsanity.com/paper/1704.06228/) (2017)
- [G-TAD: Sub-Graph Localization for Temporal Action Detection](https://mlsanity.com/paper/1911.11462/) (2019)
- [Long-Term Feature Banks for Detailed Video Understanding](https://mlsanity.com/paper/1812.05038/) (2018)
- [The THUMOS Challenge on Action Recognition for Videos "in the Wild"](https://mlsanity.com/paper/1604.06182/) (2016)
- [Sparse Networks from Scratch: Faster Training without Losing Performance](https://mlsanity.com/paper/1907.04840/) (2019)
- [Rethinking the Faster R-CNN Architecture for Temporal Action Localization](https://mlsanity.com/paper/1804.07667/) (2018)
- [Learning Salient Boundary Feature for Anchor-free Temporal Action Localization](https://mlsanity.com/paper/2103.13137/) (2021)
- [SCSampler: Sampling Salient Clips from Video for Efficient Action Recognition](https://mlsanity.com/paper/1904.04289/) (2019)
- [Relaxed Transformer Decoders for Direct Action Proposal Generation](https://mlsanity.com/paper/2102.01894/) (2021)
- [TURN TAP: Temporal Unit Regression Network for Temporal Action Proposals](https://mlsanity.com/paper/1703.06189/) (2017)
- [CDC: Convolutional-De-Convolutional Networks for Precise Temporal Action Localization in Untrimmed Videos](https://mlsanity.com/paper/1703.01515/) (2017)
- [VidTr: Video Transformer Without Convolutions](https://mlsanity.com/paper/2104.11746/) (2021)

## Built on by (most cited first, 14 of 14)

- [TALLFormer: Temporal Action Localization with a Long-memory Transformer](https://mlsanity.com/paper/2204.01680/) (2022)
- [End-to-End Temporal Action Detection with 1B Parameters Across 1000 Frames](https://mlsanity.com/paper/2311.17241/) (2023)
- [SnAG: Scalable and Accurate Video Grounding](https://mlsanity.com/paper/2404.02257/) (2024)
- [SlowFast-LLaVA-1.5: A Family of Token-Efficient Video Large Language Models for Long-Form Video Understanding](https://mlsanity.com/paper/2503.18943/) (2025)
- [Training a Large Video Model on a Single Machine in a Day](https://mlsanity.com/paper/2309.16669/) (2023)
- [To Adapt or Not to Adapt? Real-Time Adaptation for Semantic Segmentation](https://mlsanity.com/paper/2307.15063/) (2023)
- [ETAD: Training Action Detection End to End on a Laptop](https://mlsanity.com/paper/2205.07134/) (2022)
- [E2E-LOAD: End-to-End Long-form Online Action Detection](https://mlsanity.com/paper/2306.07703/) (2023)
- [MS-RNN: A Flexible Multi-Scale Framework for Spatiotemporal Predictive Learning](https://mlsanity.com/paper/2206.03010/) (2022)
- [LoSA: Long-Short-range Adapter for Scaling End-to-End Temporal Action Localization](https://mlsanity.com/paper/2404.01282/) (2024)
- [An In-depth Study of Stochastic Backpropagation](https://mlsanity.com/paper/2210.00129/) (2022)
- [Uncovering the Unseen: Discover Hidden Intentions by Micro-Behavior Graph Reasoning](https://mlsanity.com/paper/2308.15169/) (2023)
- [Beyond the Horizon: Decoupling Multi-View UAV Action Recognition via Partial Order Transfer](https://mlsanity.com/paper/2504.20530/) (2025)
- [SUS backprop: linear backpropagation algorithm for long inputs in transformers](https://mlsanity.com/paper/2505.15080/) (2025)

## Abstract (arXiv)

We propose a memory efficient method, named Stochastic Backpropagation (SBP), for training deep neural networks on videos. It is based on the finding that gradients from incomplete execution for backpropagation can still effectively train the models with minimal accuracy loss, which attributes to the high redundancy of video. SBP keeps all forward paths but randomly and independently removes the backward paths for each network layer in each training step. It reduces the GPU memory cost by eliminating the need to cache activation values corresponding to the dropped backward paths, whose amount can be controlled by an adjustable keep-ratio. Experiments show that SBP can be applied to a wide range of models for video tasks, leading to up to 80.0% GPU memory saving and 10% training speedup with less than 1% accuracy drop on action recognition and temporal action detection.