Research map / Architectures and efficiency / RNNs and normalization
Target Propagation in Recurrent Neural Networks
Nikolay Manchev, Michael Spratling
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 |
|---|---|---|---|
| replace BPTT with target propagation (TPTT) | SRN with orthogonal init · Random Permutation | max sequence length: 70 -> 300 | Improved |
| replace BPTT with TPTT | SRN · Temporal Order | max sequence length: 120 -> 150 | Improved |
| replace BPTT with TPTT | SRN · 3-bit Temporal Order | max sequence length: 70 -> 150 | Improved |
| replace BPTT with TPTT | SRN · Adding | max sequence length: 90 -> 60 | Hurt |
| replace BPTT with TPTT (best params) | SRN · Temporal Order tmax=120 | iterations to convergence: 46,500 -> 1,600 | Improved |
| replace BPTT with TPTT common model | SRN · Random Permutation tmax=70 | iterations to convergence: 12,200 -> 12,700 | No clear effect |
| add Gaussian noise injection when learning the inverse G (ablation) | TPTT · Adding tmax=70 | accuracy (%): 82.67 -> 94.99 | Improved |
| add Gaussian noise injection (ablation) | TPTT · Adding tmax=90 | accuracy (%): 92.74 -> 96.52 | Improved |
| replace BPTT with TPTT, test on 2x longer sequences | SRN · Random Permutation (train 100, test 200) | accuracy (%): 51.52 -> 100.00 | Improved |
| replace BPTT with TPTT, test on 2x longer sequences | SRN · Adding (train 100, test 200) | accuracy (%): 65.63 -> 66.23 | No clear effect |
| replace TPTT with LSTM | TPTT · Adding (train 100, test 200) | accuracy (%): 66.23 -> 91.90 | Improved |
| increase hidden size Whh from 64 to 1024 (ablation) | TPTT · MNIST sequence of pixels | maximal accuracy (%): 49.49 -> 74.02 | Improved |
Closest papers by meaning
- Training Recurrent Neural Networks via Forward Propagation Through Time (ICML 2021 · similarity 0.91)
- Memory-Efficient Backpropagation Through Time (NeurIPS 2016 · similarity 0.88)
- Training Recurrent Neural Networks Online by Learning Explicit State Variables (ICLR 2020 · similarity 0.87)
- Reviving and Improving Recurrent Back-Propagation (ICML 2018 · similarity 0.87)
- Efficient recurrent architectures through activity sparsity and sparse back-propagation through time (ICLR 2023 · similarity 0.86)
- Exploiting Symmetric Temporally Sparse BPTT for Efficient RNN Training (AAAI 2024 · similarity 0.83)
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.
