Research map / Architectures and efficiency / RNNs and normalization
Recurrent Orthogonal Networks and Long-Memory Tasks
Mikael Henaff, Arthur Szlam, Yann LeCun
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 |
|---|---|---|---|
| Use random orthogonal rather than identity transition initialization | LT-RNN · Fixed-length copy task | task performance | Improved |
| Use identity rather than random orthogonal transition initialization | LT-RNN · Adding task | task performance | Improved |
| Use random orthogonal rather than identity transition initialization | LT-RNN · Adding task | task performance | Hurt |
| Use identity rather than random orthogonal transition initialization | LT-RNN · Fixed-length copy task | task performance | Hurt |
| Add l2 pooling to the decoder | LT-ORNN · Adding task | task performance | Improved |
| Add l2 pooling to the decoder | LT-ORNN · Fixed-length copy task | task performance | No clear effect |
| Use an LT-RNN architecture, with identity, random orthogonal, or pooled random orthogonal initialization | LT-RNN · Variable length copy task (T=100) | task performance | Hurt |
| Use an LSTM rather than an LT-RNN | Variable length copy task (T=100) · Variable length copy task | task performance | Improved |
Closest papers by meaning
- Improved memory in recurrent neural networks with sequential non-normal dynamics (ICLR 2020 · similarity 0.89)
- Understanding and Controlling Memory in Recurrent Neural Networks (ICML 2019 · similarity 0.89)
- Approximation and Optimization Theory for Linear Continuous-Time Recurrent Neural Networks (JMLR 2022 · similarity 0.89)
- Capacity and Trainability in Recurrent Neural Networks (ICLR 2017 · similarity 0.88)
- UnICORNN: A recurrent model for learning very long time dependencies (ICML 2021 · similarity 0.88)
- Recurrent neural networks: vanishing and exploding gradients are not the end of the story (NeurIPS 2024 · similarity 0.88)
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.
