Research map / Architectures and efficiency / RNNs and normalization
Memory-Efficient Backpropagation Through Time
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use dynamic programming to schedule caching and recomputation under a fixed memory budget | BPTT for RNNs · Sequence length 1000 (excluding input vectors) | memory savings: 95% | Helped |
| Use dynamic programming to schedule caching and recomputation under a fixed memory budget | BPTT for RNNs · Sequence length 1000 | training-iteration time: 33% more | Hurt |
| Allow mixed hidden-state and internal-state memorization instead of restricting memorization to hidden states or internal states (ablation) | BPTT-MSM · Long sequences | computational cost | Helped |
| Use hidden-state-only memorization instead of internal-state-only memorization (ablation) | BPTT-HSM versus BPTT-ISM · Long sequences at fixed memory capacity | computational cost | Helped |
| Use internal-state-only memorization instead of hidden-state-only memorization (ablation) | BPTT-ISM versus BPTT-HSM · Relatively short sequences at fixed memory capacity | computational cost | Helped |
Closest papers by meaning
- Training Recurrent Neural Networks via Forward Propagation Through Time (ICML 2021 · similarity 0.90)
- Reducing Fine-Tuning Memory Overhead by Approximate and Memory-Sharing Backpropagation (ICML 2024 · similarity 0.88)
- Recurrent neural networks: vanishing and exploding gradients are not the end of the story (NeurIPS 2024 · similarity 0.87)
- Exploring the Promise and Limits of Real-Time Recurrent Learning (ICLR 2024 · similarity 0.87)
- Advancing Training Efficiency of Deep Spiking Neural Networks through Rate-based Backpropagation (NeurIPS 2024 · similarity 0.87)
- Exploiting Symmetric Temporally Sparse BPTT for Efficient RNN Training (AAAI 2024 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.