Research map / Architectures and efficiency / RNNs and normalization
ProtoNN: Compressed and Accurate kNN for Resource-scarce Devices
Chirag Gupta, Arun Sai Suggala, Ankit Goyal, Harsha Vardhan Simhadri, Bhargavi Paranjape, Ashish Kumar, Saurabh Goyal, Raghavendra Udupa, Manik Varma, Prateek Jain
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 |
|---|---|---|---|
| Compress to 16kB model (binary) / 64kB (multiclass) vs best uncompressed baselines | ProtoNN · binary and multiclass benchmarks | accuracy | Mixed |
| ProtoNN (16kB) vs RBF-SVM on character recognition | ProtoNN · character recognition | accuracy | Improved |
| ProtoNN (64kB) vs RBF-SVM on letter-26 | ProtoNN · letter-26 | accuracy | Hurt |
| ProtoNN vs RBF-SVM on cifar binary | ProtoNN · cifar | accuracy | Hurt |
| ProtoNN vs RBF-SVM on curet-61 | ProtoNN · curet-61 | accuracy | Hurt |
| Compare to compressed baselines (BudgetRF, Decision Jungle, LDKL, Tree Pruning, GBDT, Budget Prune, SNC, NN Pruning) at 2kB | ProtoNN · binary datasets, model size 2-16kB | accuracy | Improved |
| Use ProtoNN m=5000 instead of RBF-SVM on aloi | ProtoNN · aloi | accuracy | Mixed |
| ProtoNN m=1500 vs FastXML on aloi | ProtoNN · aloi | accuracy | No clear effect |
| ProtoNN vs Recall Tree on aloi | ProtoNN · aloi | accuracy | Improved |
| ProtoNN vs FastXML, DiSMEC, SLEEC on multilabel datasets | ProtoNN · mediamill, delicious, eurlex | P@1 | Mixed |
| Vary number of prototypes m or projection dimension d-hat against BNC and SNC | ProtoNN · mnist binary | accuracy | Improved |
| Integer-only implementation with precomputed exponential table | ProtoNN (2kB) · Arduino Uno, 4 binary datasets | energy and prediction time | Improved |
Closest papers by meaning
- Boosted Trees on a Diet: Compact Models for Resource-Constrained Devices (ICLR 2026 · similarity 0.86)
- TinyTrain: Resource-Aware Task-Adaptive Sparse Training of DNNs at the Data-Scarce Edge (ICML 2024 · similarity 0.85)
- TinyTTA: Efficient Test-time Adaptation via Early-exit Ensembles on Edge Devices (NeurIPS 2024 · similarity 0.82)
- TinyFoA: Memory Efficient Forward-Only Algorithm for On-Device Learning (AAAI 2025 · similarity 0.82)
- AtomNet: Designing Tiny Models from Operators Under Extreme MCU Constraints (AAAI 2025 · similarity 0.81)
- Benchmarking Predictive Coding Networks -- Made Simple (ICLR 2025 · similarity 0.81)
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.
