Research map / Architectures and efficiency / RNNs and normalization
Input-Output Equivalence of Unitary and Contractive RNNs
Melikasadat Emami, Mojtaba Sahraee Ardakan, Sundeep Rangan, Alyson K. Fletcher
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 a URNN with up to twice the hidden states | Contractive ReLU RNN · Bounded input sequences | input-output equivalence | Improved |
| Increase the URNN hidden-state count to twice the RNN count | RNN and URNN · Synthetic slowly-varying dynamical systems | test error | No clear effect |
| Impose a contractivity constraint | RNN · Pixel-Permuted MNIST | accuracy | Improved |
| Impose a unitary constraint | RNN · Pixel-Permuted MNIST | accuracy | Improved |
| Replace ReLU with sigmoid activation | Contractive RNN · Neighborhood of an input where the URNN is controllable and observable | exact input-output equivalence | Hurt |
| Limit the equivalent URNN to fewer than twice the RNN's hidden states | Contractive ReLU RNN · Worst-case exact matching | minimum hidden-state count for equivalence | Hurt |
Closest papers by meaning
- Full-Capacity Unitary Recurrent Neural Networks (NeurIPS 2016 · similarity 0.90)
- Unitary Evolution Recurrent Neural Networks (ICML 2016 · similarity 0.88)
- Tunable Efficient Unitary Neural Networks (EUNN) and their application to RNNs (ICML 2017 · similarity 0.86)
- RNNs of RNNs: Recursive Construction of Stable Assemblies of Recurrent Neural Networks (NeurIPS 2022 · similarity 0.86)
- Recurrent neural networks: vanishing and exploding gradients are not the end of the story (NeurIPS 2024 · similarity 0.86)
- On the Expressivity of Recurrent Neural Cascades (AAAI 2024 · similarity 0.85)
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.
