mlsanityData to 14 Oct 2026

AutoMCU: Feasibility-First MCU Neural Network Customization via LLM-based Multi-Agent Systems

arXiv 2605.21560 ↗First posted 20 May 2026cs.LGcited by 0

On the map: LLM agents and tool use → Software engineering coding agents

Abstract (arXiv)

Deploying neural networks on microcontroller units (MCUs) is critical for edge intelligence but remains challenging due to tight memory, storage, and computation constraints. Existing approaches, such as model compression and hardware-aware neural architecture search (HW-NAS), often depend on proxy metrics, incur high search cost, and do not fully bridge the gap between architecture design and verified deployment. This paper presents AutoMCU, a feasibility-first large language model (LLM)-based multi-agent system for automated neural network customization under MCU constraints. Given natural-language task requirements and hardware specifications, AutoMCU iteratively generates structured architecture candidates, filters infeasible designs through vendor toolchain feedback before training, evaluates feasible models under a controlled protocol, and verifies deployability through backend-grounded deployment analysis. AutoMCU includes two key mechanisms: 1) hardware-in-the-loop architecture generation for early elimination of undeployable candidates under RAM and Flash constraints, and 2) state-isolated multi-agent scheduling for stable coordination of proposal, training, evaluation, and deployment stages. Experiments on CIFAR-10 and CIFAR-100 under strict MCU constraints show that AutoMCU achieves competitive accuracy while reducing customization time to about 1--2 hours, compared with hundreds of GPU hours for representative MCU-oriented HW-NAS baselines. Comparisons with ColabNAS and the LLM-based NAS method GENIUS on NAS-Bench-201 further demonstrate the effectiveness and stability of AutoMCU. Real-device deployments on multiple STM32 microcontrollers validate its practical applicability to MCU-scale edge intelligence.

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Placed in Software engineering coding agents, one of 16 directions in LLM agents and tool use. Builds on 43 of its 52 references that are in the map. First posted on arXiv 20 May 2026.

Builds on

43 of 52 references.

References · 43

  1. 2015Deep Residual Learning for Image Recognitioncited by 52,096
  2. 2024Qwen2.5 Technical Reportcited by 4,942
  3. 2018MobileNetV2: Inverted Residuals and Linear Bottleneckscited by 4,795
  4. 2014Neural Machine Translation by Jointly Learning to Align and Translatecited by 4,548
  5. 2017Fashion-MNIST: a Novel Image Dataset for Benchmarking Machine Learning Algorithmscited by 4,349
  6. 2015Deep Compression: Compressing Deep Neural Networks with Pruning, Trained Quantization and Huffman Codingcited by 2,329
  7. 2015Learning both Weights and Connections for Efficient Neural Networkscited by 2,031
  8. 2018DARTS: Differentiable Architecture Searchcited by 1,798
  9. 2017Quantization and Training of Neural Networks for Efficient Integer-Arithmetic-Only Inferencecited by 1,593
  10. 2016Neural Architecture Search with Reinforcement Learningcited by 1,499
  11. 2018ShuffleNet V2: Practical Guidelines for Efficient CNN Architecture Designcited by 1,136
  12. 2017Learning Transferable Architectures for Scalable Image Recognitioncited by 1,101
31 more
  1. 2016SqueezeNet: AlexNet-level accuracy with 50x fewer parameters and <0.5MB model sizecited by 1,060
  2. 2018MnasNet: Platform-Aware Neural Architecture Search for Mobilecited by 943
  3. 2018Efficient Neural Architecture Search via Parameter Sharingcited by 837
  4. 2025DeepSeek-V3.2: Pushing the Frontier of Open Large Language Modelscited by 815
  5. 2019Once-for-All: Train One Network and Specialize it for Efficient Deploymentcited by 787
  6. 2017Channel Pruning for Accelerating Very Deep Neural Networkscited by 702
  7. 2018ProxylessNAS: Direct Neural Architecture Search on Target Task and Hardwarecited by 673
  8. 2015Listen, Attend and Spellcited by 543
  9. 2018FBNet: Hardware-Aware Efficient ConvNet Design via Differentiable Neural Architecture Searchcited by 485
  10. 2018HAQ: Hardware-Aware Automated Quantization with Mixed Precisioncited by 443
  11. 2017ThiNet: A Filter Level Pruning Method for Deep Neural Network Compressioncited by 442
  12. 2020NAS-Bench-201: Extending the Scope of Reproducible Neural Architecture Searchcited by 416
  13. 2019Single Path One-Shot Neural Architecture Search with Uniform Samplingcited by 414
  14. 2015Attention-Based Models for Speech Recognitioncited by 401
  15. 2019AutoML: A Survey of the State-of-the-Artcited by 390
  16. 2017Large-Scale Evolution of Image Classifierscited by 348
  17. 2020MCUNet: Tiny Deep Learning on IoT Devicescited by 273
  18. 2019PC-DARTS: Partial Channel Connections for Memory-Efficient Architecture Searchcited by 265
  19. 2026MiMo-V2-Flash Technical Reportcited by 191
  20. 2020TensorFlow Lite Micro: Embedded Machine Learning on TinyML Systemscited by 165
  21. 2019FairNAS: Rethinking Evaluation Fairness of Weight Sharing Neural Architecture Searchcited by 130
  22. 2018ChamNet: Towards Efficient Network Design through Platform-Aware Model Adaptationcited by 76
  23. 2023Can GPT-4 Perform Neural Architecture Search?cited by 74
  24. 2021Zen-NAS: A Zero-Shot NAS for High-Performance Deep Image Recognitioncited by 73
  25. 2020μNAS: Constrained Neural Architecture Search for Microcontrollerscited by 55
  26. 2021EPE-NAS: Efficient Performance Estimation Without Training for Neural Architecture Searchcited by 24
  27. 2023Zero-Shot Neural Architecture Search: Challenges, Solutions, and Opportunitiescited by 21
  28. 2020AdaDeep: A Usage-Driven, Automated Deep Model Compression Framework for Enabling Ubiquitous Intelligent Mobilescited by 10
  29. 2021LSPnet: A 2D Localization-oriented Spacecraft Pose Estimation Neural Networkcited by 10
  30. 2020Evolving Neural Architecture Using One Shot Modelcited by 8
  31. 2025NestQuant: Post-Training Integer-Nesting Quantization for On-Device DNNcited by 4

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2026From Human Guidance to Autonomy: Agent Skill System for End-to-End LLM Deployment on Spatial NPUs
  2. 2025MARCO: Hardware-Aware Neural Architecture Search for Edge Devices with Multi-Agent Reinforcement Learning and Conformal Prediction Filtering
  3. 2026ChatNeuroSim: An LLM Agent Framework for Automated Compute-in-Memory Accelerator Deployment and Optimization
  4. 2026Embedded Arena: Iterative Optimization via Hardware Feedback
  5. 2026Hardware Co-Design Scaling Laws via Roofline Modelling for On-Device LLMs
  6. 2026PrototypeNAS: Rapid Design of Deep Neural Networks for Microcontroller Units
  7. 2026Retrieve, Schedule, Reflect: LLM Agents for Chip QoR Optimization
  8. 2026LLM as a Tool, Not an Agent: Code-Mined Tree Transformations for Neural Architecture Search