Research map / Architectures and efficiency / RNNs and normalization
Minimal Width for Universal Property of Deep RNN
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 a deep ReLU RNN for past-dependent continuous sequence-to-sequence approximation | RNN · Past-dependent continuous sequence-to-sequence functions on compact K | minimum width upper bound: d_x+d_y+3 | Helped |
| Use ReLU rather than a general non-degenerate activation for past-dependent L^p approximation | RNN · Past-dependent L^p sequence-to-sequence functions | minimum width upper bound: max{d_x+1,d_y}+1 -> max{d_x+1,d_y} | Helped |
| Add a backward recurrent direction | RNN · Unrestricted continuous sequence-to-sequence functions on compact K | universality: null | Helped |
| Use a bidirectional rather than one-way recurrent structure for unrestricted L^p sequence-to-sequence functions | RNN · Unrestricted L^p sequence-to-sequence functions | universality: null | Helped |
| Replace RNN cells with LSTM cells | Deep RNN · Past-dependent continuous sequence-to-sequence functions on compact K | minimum width upper bound: d_x+d_y+3 | No clear effect |
| Replace RNN cells with GRU cells | Deep RNN · Past-dependent continuous sequence-to-sequence functions on compact K | minimum width upper bound: d_x+d_y+3 | No clear effect |
Closest papers by meaning
- Achieve the Minimum Width of Neural Networks for Universal Approximation (ICLR 2023 · similarity 0.84)
- Minimum width for universal approximation using ReLU networks on compact domain (ICLR 2024 · similarity 0.82)
- Minimum Width for Universal Approximation using Squashable Activation Functions (ICML 2025 · similarity 0.82)
- Minimum Width for Deep, Narrow MLP: A Diffeomorphism Approach (NeurIPS 2025 · similarity 0.82)
- Depth-Bounds for Neural Networks via the Braid Arrangement (NeurIPS 2025 · similarity 0.80)
- Characterizing ResNet's Universal Approximation Capability (ICML 2024 · similarity 0.80)
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.