analiza aiqbench
This research introduces LocalTrain, an FPGA-based system utilizing integer-only arithmetic to enable energy-efficient CNN training on edge devices. By sharing matrix engines and folding gradients tile-by-tile, the KV260 prototype achieves significant power reductions compared to GPU counterparts while maintaining functional agreement with software references. The work addresses critical bottlenecks in cross-layer training dependencies, offering a viable path for deploying deep learning models in resource-constrained environments.
LocalTrain: Integer-Only Local Learning for Energy-Efficient CNN Training on Edge FPGAs
arXiv:2610.17099v1 Announce Type: cross Abstract: Local learning reduces cross-layer training dependencies, but an efficient accelerator must still retain operands, accumulate gradients, and select quantization scales. We present an FPGA system that trains a CNN and its regional auxiliary classifiers using a shared integer matrix engine. Compact errors are expanded on chip, gradients are folded tile by tile, and historical scales permit direct output conversion. At 200 MHz, the KV260 prototype achieves 76.57% accuracy on CIFAR-10 and 91.85% on SVHN at approximately 4.20 W mean module power. Effective matrix throughput is 31.22 and 31.57 GMAC/s, respectively. Against the same integer method on Jetson Orin Nano, measured training energy is 40.4% and 42.1% lower, with $1.86\times$ and $1.82\times$ longer runtime; measurements use the respective platform power domains. Selected state checks and matching complete-run prediction counts support functional agreement with the software reference.
Abstract:Local learning reduces cross-layer training dependencies, but an efficient accelerator must still retain operands, accumulate gradients, and select quantization scales. We present an FPGA system that trains a CNN and its regional auxiliary classifiers using a shared integer matrix engine. Compact errors are expanded on chip, gradients are folded tile by tile, and historical scales permit direct output conversion. At 200 MHz, the KV260 prototype achieves 76.57% accuracy on CIFAR-10 and 91.85% on SVHN at approximately 4.20 W mean module power. Effective matrix throughput is 31.22 and 31.57 GMAC/s, respectively. Against the same integer method on Jetson Orin Nano, measured training energy is 40.4% and 42.1% lower, with $1.86\times$ and $1.82\times$ longer runtime; measurements use the respective platform power domains. Selected state checks and matching complete-run prediction counts support functional agreement with the software reference.
Hardware Architecture (cs.AR); Machine Learning (cs.LG)
arXiv:2610.17099 [cs.AR]
(or arXiv:2610.17099v1 [cs.AR] for this version)
https://doi.org/10.48550/arXiv.2610.17099 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 232 cuvinte
Subiecte conexe
On-Device Inference versus Wireless Streaming: Energy-Efficient Multi-Modal Deep Learning for Wearable Cardiovascular Patches
arXiv · 28 iul. 2026
News
Quantum Computing Report · 25 iul. 2026
Synthesis Superintelligence: from Semiconductors to Superconductors — Periodic Labs’ Liam Fedus and Ekin Dogus Cubuk
Latent Space · 8 oct. 2026
NVIDIA GTC Taipei at COMPUTEX: Live Updates on What’s Next in AI
NVIDIA · 21 mai 2026