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.
From: Ilja Bekman [view email]
[v1]
Fri, 7 Nov 2025 18:44:50 UTC (458 KB)
[v2]
Fri, 27 Feb 2026 11:08:13 UTC (490 KB)
[v3]
Thu, 21 May 2026 16:57:05 UTC (2,602 KB)
[v4]
Mon, 3 Aug 2026 23:22:25 UTC (155 KB)
此内容由惯性聚合(RSS阅读器)自动聚合整理，仅供阅读参考。 原文来自 — 版权归原作者所有。