Research map / Architectures and efficiency / RNNs and normalization
Capacity and Trainability in Recurrent Neural Networks
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 gated architectures instead of vanilla RNNs | RNNs · Parentheses counting and arithmetic tasks | training difficulty | Helped |
| Use +RNN rather than other architectures at depth 8 | RNNs · Parentheses counting and arithmetic tasks | trainability | Helped |
| Use GRU rather than other architectures at depth 1 | RNNs · Parentheses counting and arithmetic tasks | trainability | Helped |
| Increase network depth | RNNs · Parentheses counting and arithmetic tasks | task-solving ability | Mixed |
| Use ReLU activation (ablation) | RNN capacity models · Per-parameter capacity task | bits per parameter | Hurt |
| Increase the number of unrolling steps beyond one | RNNs · Per-parameter capacity task | task capacity | Helped |
| Use a gated architecture rather than vanilla RNN | RNNs · Per-parameter capacity task | bits per parameter | Hurt |
| Use any common RNN architecture with at least as many hidden units as input dimensions | RNNs · Input memory capacity task | input reconstruction | Helped |
| Hold parameter count fixed across architectures | RNN architectures · text8 and random continuous functions | evaluation loss | No clear effect |
| Use a vanilla RNN or IRNN rather than a gated architecture | RNNs · Deep difficult tasks | trainability | Hurt |
Closest papers by meaning
- Variable Computation in Recurrent Neural Networks (ICLR 2017 · similarity 0.90)
- Efficient recurrent architectures through activity sparsity and sparse back-propagation through time (ICLR 2023 · similarity 0.90)
- Architectural Complexity Measures of Recurrent Neural Networks (NeurIPS 2016 · similarity 0.89)
- On the Convergence Rate of Training Recurrent Neural Networks (NeurIPS 2019 · similarity 0.89)
- Dilated Recurrent Neural Networks (NeurIPS 2017 · similarity 0.89)
- Recurrent neural networks: vanishing and exploding gradients are not the end of the story (NeurIPS 2024 · similarity 0.89)
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.