AI
BeFit: Shifting Complexity from Weights to Bernstein Activations for Resource-Constrained Sensing
Publication: cs.AI updates on arXiv.org
Author: Malak Gamal El-Din, Yasser Shoukry, Salma Elmalaki
Source published: 19/11/2026, 05:00:00 UTC
Briefing created: 19/11/2026, 05:04:44 UTC
Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 19/11/2026, 05:03:25 UTC.
Briefings
BeFit is a framework for resource-constrained sensing that shifts model complexity from weights to learnable Bernstein polynomial activations. On TI CC3220SF microcontrollers, it achieves 2.75% higher accuracy than ReLU CNNs while reducing instruction counts by up to 147x on RISC-V cores. The method projects weights to multiplication-free codes, freezes them, and recovers accuracy via learnable activations, with final models staying within 1.56 percentage points of float32 teachers across all tested tasks.
Key points
- BeFit shifts model complexity from weights to learnable Bernstein polynomial activations to enable resource-constrained sensing.
- On TI CC3220SF microcontrollers, BeFit achieves 2.75% higher accuracy than ReLU CNNs while reducing instruction counts by up to 147x on RISC-V cores.
Context
Time-series sensing increasingly relies on neural networks, yet deployment on resource-constrained microcontrollers remains challenging. Aggressive model compression can reduce accuracy, while restricting weights to multiplication-free representations introduces further losses. BeFit addresses this by shifting complexity to learnable Bernstein activations.
Limitations
Evidence is limited to human activity recognition and transformer/MLP tasks on TI CC3220SF and RISC-V cores. Performance metrics are based on emulation and hardware-aware NAS; specific deployment scenarios without hardware validation are not covered.
Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 19/11/2026, 05:04:44 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.
View archived briefing on IPFS
bafybeihlvkou6tjqtoxtynvmxr6hy24zd24o7ll33suaheak3iswkqkzcu