Quantization Effects of Artificial Neural Networks for Embedded Edge-Computing Applications

2025/11/07 by Alperen Aksoy, Ilja Bekman, Aksoy, Alperen +17
Computer Science · Engineering · Physics and Astronomy · #Advanced Memory and Neural Computing #Advanced Neural Network Applications #FOS: Computer and information sciences #FOS: Physical sciences #Instrumentation and Detectors (physics.ins-det) #Neural and Evolutionary Computing (cs.NE) #Wireless Signal Modulation Classification #cs.NE #physics.ins-det

paper · pdf · doi:10.48550/arxiv.2511.05479

deRSE26 proceedings pre-print

openalex publication_date 2025/11/07 · openalex created_date 2025/12/05 · openalex updated_date 2026/07/28 · arxiv created 2026/08/03 · arxiv updated 2026/08/05

Abstract

This paper examines the use of Quantized Neural Networks (QNNs) for two resource-constrained scientific applications: automated calibration of semi-conductor quantum bits (qubits) and scientific particle detectors. We evaluate the trade-offs between Post-Training Quantization (PTQ), Quantization-Aware Training (QAT), and ultra-low-bit Binary Neural Networks (BNNs) with respect to latency and resource usage. Our results demonstrate that PTQ achieves a four-fold reduction in memory usage for U-shaped CNN (U-Net) architectures. For the training of non-differentiable custom BNNs , we propose a novel, hardware-constrained learning approach using Genetic Algorithms (GAs). We showcase a LUT-based BNN architecture suitable for direct conversion to VHDL via the HCL4BNN framework. This method achieves nanosecond-scale inference latencies (10-15 ns) without requiring specialized DSP or BRAM resources.

Citations

• Automated Charge Transition Detection in Quantum Dot Charge Stability Diagrams
• AI-Enhanced Self-Triggering for Extensive Air Showers: Performance and FPGA Feasibility
• Design Space Exploration for Particle Detector Read-out Implementations in Matlab and Simulink on the Example of the SHiP SBT
• MLaaS4HEP: Machine Learning as a Service for HEP
• LogicNets: Co-Designed Neural Networks and Circuits for\n Extreme-Throughput Applications
• LUTNet: Rethinking Inference in FPGA Soft Logic
• Quantization and Training of Neural Networks for Efficient\n Integer-Arithmetic-Only Inference
• Quantized Neural Networks: Training Neural Networks with Low Precision Weights and Activations
• U-Net: Convolutional Networks for Biomedical Image Segmentation

Related

• Learning Before Filtering: Real-Time Hardware Learning at the Detector Level
• FPGA implementation of a deep learning algorithm for real-time signal reconstruction in radiation detectors under high pile-up conditions
• Real-time Inference with 2D Convolutional Neural Networks on Field Programmable Gate Arrays for High-rate Particle Imaging Detectors
• Analysis of Hardware Synthesis Strategies for Machine Learning in Collider Trigger and Data Acquisition
• Neural Network Acceleration on MPSoC board: Integrating SLAC's SNL, Rogue Software and Auto-SNL