Research map / Architectures and efficiency / RNNs and normalization
Resource-efficient Machine Learning in 2 KB RAM for the Internet of Things
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 re-training (fixed-support fine-tuning) after termination (ablation) | Bonsai · Chars4K-2 | accuracy (%): 75.19 -> 76.67 | Helped |
| add proper initialization (unconstrained gradient steps then hard thresholding) and re-training (ablation) | Bonsai · Chars4K-2 | accuracy (%): 74.12 -> 76.67 | Helped |
| learn the projection matrix jointly instead of via sparse PCA first (ablation) | Bonsai · Chars4K-2 | accuracy (%): 58.32 -> 76.67 | Helped |
| train Tree Pruning in sparse PCA space (ablation) | Tree Pruning · Chars4K-2 | accuracy (%): 63.57 | No clear effect |
| train RBF-SVM in sparse PCA space (ablation) | RBF-SVM · Chars4K-2 | accuracy (%): 71.10 | Hurt |
| use 1-byte fixed point and tanh approximation (BonsaiOpt) instead of unoptimized Bonsai | Bonsai · Eye-2 Arduino Uno | accuracy (%): 88.26 -> 88.78 | Helped |
| increase model size from 2KB to 16KB | Bonsai · Chars4K-2 | accuracy (%): 74.28 -> 76.67 | Helped |
| increase model size from 2KB to 16KB | Bonsai · MNIST-2 | accuracy (%): 94.38 -> 96.47 | Helped |
| replace uncompressed neural net with Bonsai on MNIST-2 | Bonsai · MNIST-2 | accuracy (%): 98.08 -> 94.38 | Hurt |
| replace Bing FastRank L3 ranker with Bonsai at 300 bytes | Bonsai · L3 Ranking | nDCG: 8.3% | Helped |
| replace linear classifier with BonsaiOpt on Arduino Uno | Bonsai · Eye-2 | accuracy (%): 80.10 -> 88.78 | Helped |
| use Bonsai instead of other resource-efficient algorithms at equal model size | Bonsai · Chars4K-62 | accuracy (%): 30.7% | Helped |
Closest papers by meaning
- TinyTTA: Efficient Test-time Adaptation via Early-exit Ensembles on Edge Devices (NeurIPS 2024 · similarity 0.82)
- TinyTrain: Resource-Aware Task-Adaptive Sparse Training of DNNs at the Data-Scarce Edge (ICML 2024 · similarity 0.81)
- TinyFoA: Memory Efficient Forward-Only Algorithm for On-Device Learning (AAAI 2025 · similarity 0.79)
- QUTE: Quantifying Uncertainty in TinyML models with Early-exit-assisted ensembles for model-monitoring (ICML 2025 · similarity 0.79)
- AtomNet: Designing Tiny Models from Operators Under Extreme MCU Constraints (AAAI 2025 · similarity 0.79)
- DEX: Data Channel Extension for Efficient CNN Inference on Tiny AI Accelerators (NeurIPS 2024 · similarity 0.78)
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.