Research map / Architectures and efficiency / RNNs and normalization
FastGRNN: A Fast, Accurate, Stable and Tiny Kilobyte Sized Gated Recurrent Neural Network
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 |
|---|---|---|---|
| Add a weighted residual connection to a standard RNN | FastRNN · Google-12 | Accuracy (%): 73.25 -> 92.21+ | Helped |
| Add a weighted residual connection to a standard RNN | FastRNN · Pixel-MNIST-10 | Accuracy (%): 94.10 -> 96.44 | Helped |
| Extend the residual connection to a gate that reuses the RNN matrices | FastGRNN · Google-12 | Accuracy (%): 92.21+ -> 93.18 | Helped |
| Extend the residual connection to a gate that reuses the RNN matrices | FastGRNN · DSA-19 | Accuracy (%): 84.14 -> 85.00 | Helped |
| Use low-rank matrices (ablation) | FastGRNN · multiple datasets | prediction accuracy | Helped |
| Add sparsity and quantization to the matrices (ablation) | FastGRNN · multiple datasets | prediction accuracy | Hurt |
| Add sparsity and quantization to the matrices (ablation) | FastGRNN · multiple datasets | model size | Helped |
| Quantize weights and use integer arithmetic (ablation) | FastGRNN-Q · Google-12 | prediction time (ms): 537 -> 2282 | Hurt |
| Enforce β = 1 − α on short sequences (ablation) | FastRNN · short sequences | accuracy: up to 1.5% | Hurt |
| Use a 1 KB FastGRNN model | FastGRNN · Wakeword-2 | F1 score | No clear effect |
Closest papers by meaning
- Sparse Persistent RNNs: Squeezing Large Recurrent Networks On-Chip (ICLR 2018 · similarity 0.88)
- FlashRNN: I/O-Aware Optimization of Traditional RNNs on modern hardware (ICLR 2025 · similarity 0.88)
- Recurrent neural networks: vanishing and exploding gradients are not the end of the story (NeurIPS 2024 · similarity 0.87)
- M$^2$RNN: Non-Linear RNNs with Matrix-Valued States for Scalable Language Modeling (NeurIPS 2026 · similarity 0.87)
- Exploiting Symmetric Temporally Sparse BPTT for Efficient RNN Training (AAAI 2024 · similarity 0.87)
- Accelerating Linear Recurrent Neural Networks for the Edge with Unstructured Sparsity (ICML 2025 · similarity 0.87)
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.