SNAC-Pack 2.0: Scaled-Out Surrogate Neural Architecture Codesign

2026/05/31 by Jason Weitz, Dmitri Demler, Benjamin Hawks +3
Computer Science · Physics and Astronomy · #cs.LG #cs.AI #hep-ex

paper · pdf

16 pages, 3 figures, Camera-ready version for International Conference on Automated Machine Learning (AutoML) 2026

arxiv created 2026/07/29 · arxiv updated 2026/07/31

Abstract

Neural architecture search (NAS) is a powerful approach for automating model design, but existing methods often optimize for accuracy alone or rely on proxy metrics such as bit operations (BOPs) that correlate poorly with hardware cost. This gap is particularly large for FPGA deployment, where cost is dominated by a multi-dimensional budget of lookup tables, DSPs, flip-flops, BRAM, and latency. We present the Surrogate Neural Architecture Codesign Package (SNAC-Pack), an open-source AutoML framework for hardware-aware neural architecture codesign and end-to-end FPGA deployment. SNAC-Pack runs a multi-objective global search with Optuna and NSGA-II, loading trials to a shared SQLite store that enables parallel workers across compute nodes. A hardware surrogate model outputs per-trial resource and latency estimates, avoiding the synthesis cost that would otherwise dominate the search loop. A local search stage then applies quantization-aware training (QAT) together with iterative magnitude pruning in a combined compression loop, after which the final model is synthesized to FPGA firmware via the hls4ml Python library. A YAML configuration and an optional agentic frontend let users run the pipeline on new datasets without modifying the framework. We demonstrate SNAC-Pack on jet classification at the Large Hadron Collider and superconducting qubit readout, discovering compact Pareto-optimal architectures that either match baseline task performance while reducing FPGA resource utilization or substantially reduce hardware utilization with only modest decreases in task performance. In the qubit readout case, SNAC-Pack also reduces the design space exploration process from months of manual fine-tuning to hours of automated search.

Citations

• hls4ml: A Flexible, Open-Source Platform for Deep Learning Acceleration on Reconfigurable Hardware
• wa-hls4ml: A Benchmark and Surrogate Models for hls4ml Resource and Latency Estimation
• End-to-end workflow for machine learning-based qubit readout with QICK and hls4ml
• Neural Architecture Codesign for Fast Physics Applications
• rule4ml: An Open-Source Tool for Resource Utilization and Latency Estimation for ML Models on FPGA
• AutoGluon-Multimodal (AutoMM): Supercharging Multimodal AutoML with Foundation Models
• Multi-Objective Hardware Aware Neural Architecture Search using Hardware Cost Diversity
• On Latency Predictors for Neural Architecture Search
• AutoMMLab: Automatically Generating Deployable Models from Language Instructions for Computer Vision Tasks
• Neural Network Quantization for Efficient Inference: A Survey
• HW-NAS-Bench:Hardware-Aware Neural Architecture Search Benchmark
• BRP-NAS: Prediction-based NAS using GCNs
• APQ: Joint Search for Network Architecture, Pruning and Quantization Policy
• Linear Mode Connectivity and the Lottery Ticket Hypothesis
• FLAML: A Fast and Lightweight AutoML Library
• Once-for-All: Train One Network and Specialize it for Efficient Deployment
• Optuna: A Next-generation Hyperparameter Optimization Framework
• Accuracy vs. Efficiency: Achieving Both through FPGA-Implementation Aware Neural Architecture Search
• FBNet: Hardware-Aware Efficient ConvNet Design via Differentiable Neural Architecture Search
• ProxylessNAS: Direct Neural Architecture Search on Target Task and Hardware
• DARTS: Differentiable Architecture Search
• Efficient Multi-objective Neural Architecture Search via Lamarckian Evolution
• Fast inference of deep neural networks in FPGAs for particle physics
• The Lottery Ticket Hypothesis: Finding Sparse, Trainable Neural Networks
• Efficient Neural Architecture Search via Parameter Sharing
• Deep Compression: Compressing Deep Neural Networks with Pruning, Trained Quantization and Huffman Coding
• Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift

Related

• Surrogate Neural Architecture Codesign Package (SNAC-Pack)
• Neural Architecture Codesign for Fast Physics Applications
• LLM-NAS: LLM-driven Hardware-Aware Neural Architecture Search
• Best of Both Worlds: AutoML Codesign of a CNN and its Hardware\n Accelerator
• On Neural Architecture Search for Resource-Constrained Hardware Platforms