Late Breaking Results: Conversion of Neural Networks into Logic Flows for Edge Computing
Image generated by Gemini AI
Researchers have developed a method to enhance neural network performance on resource-constrained CPUs by converting them into decision trees and logic flows. This approach reduces latency by up to 14.9% on a simulated RISC-V CPU while maintaining accuracy. The code is available for public use at https://github.com/TUDa-HWAI/NN2Logic.
Neural Networks Transformed into Logic Flows for Enhanced Edge Computing Efficiency
Recent research reveals a novel approach to optimizing neural networks for resource-constrained edge devices, primarily utilizing central processing units (CPUs). By converting neural networks into logic flows, researchers achieved significant reductions in latency while maintaining accuracy.
The study proposes a method where neural networks are transformed into equivalent decision trees. From these decision trees, paths with constant leaves are compressed into logic flows. This allows for more effective execution on CPUs.
Experimental results indicate that this approach can lower latency by up to 14.9% on a simulated RISC-V CPU without any degradation in accuracy. The code for this transformation process is available on GitHub at https://github.com/TUDa-HWAI/NN2Logic.
Related Topics:
📰 Original Source: https://arxiv.org/abs/2601.22151v1
All rights and credit belong to the original publisher.