neural-architecture-search

This survey reviews neural architecture search (NAS) methods applied to traffic prediction, organizing them by search strategy (gradient-based, evolutionary, one-shot weight-sharing) and discussing challenges such as computational scalability, cross-city generalization, and future directions.

The paper proposes OrchNAS, an energy-aware personalized federated edge intelligence framework that uses a Neural Architecture Search service to automatically design service-adaptive models for heterogeneous edge environments, addressing energy constraints and statistical heterogeneity.

This paper scales a closed-loop LLM-based channel configuration search to 250 candidates per cycle, showing positive accuracy trends and improved parameter efficiency on CIFAR-100, and revealing architectural regularities in LLM-generated channel priors.

The paper presents BearingNAS, a hardware-aware neural architecture search framework that designs intelligent fault diagnosis systems for bearings, targeting microcontrollers and sensor processing units with extremely limited memory (4-8 KiB RAM, 16-32 KiB Flash) while running entirely on a laptop CPU and achieving 99.50% diagnostic accuracy.

This paper presents an LLM-driven pipeline using GPT-5, GPT-4o, and Claude Sonnet 4 to automatically design neural network architectures for cross-lingual handwritten OCR, achieving over 93% accuracy across Arabic, English, and Persian scripts without human intervention.

NVIDIA introduces Llama-Nemotron, an open family of reasoning models (Nano 8B, Super 49B, Ultra 253B) that rival DeepSeek-R1 with superior inference efficiency, dynamic reasoning toggle, and open post-training datasets.

Introduces AgentNAS, a mechanism that uses an LLM to generate a seed architecture and decompose it into a slotted architecture, defining a task-specific search space for conventional NAS to explore, achieving state-of-the-art on 11 of 17 tasks.

LEMUR 2 introduces a large-scale dataset of over 14,000 neural network architectures and 750,000 training records across multimodal tasks, supporting NAS, AutoML, and deployment analysis.

This paper proposes a source-guided protocol where an LLM generates candidate modifications for a weak target model using a stronger same-family source model, showing substantial accuracy improvements on CIFAR-10 and SVHN benchmarks while disentangling transfer from adaptation effects.

Introduces CamoNAS, a frequency-aware multi-resolution Neural Architecture Search framework for camouflaged object detection, achieving state-of-the-art results on four benchmarks.

Introduces an evolutionary neural architecture search framework (EvoTS) for discovering task-adaptive Transformer-like models for multivariate time-series forecasting. The approach uses a modular genome representation and achieves competitive performance on ETT benchmark datasets.

Introduces EVOM, an agentic meta-evolution framework using an LLM-based design agent to automatically discover high-performance actor-critic architectures for reinforcement learning, outperforming manual baselines and prior methods on continuous control tasks.

This paper provides a comprehensive review of Neural Architecture Search (NAS) methods applied to Generative Adversarial Networks (GANs), categorizing approaches and highlighting benefits and limitations.

Proposes a lightweight neural architecture search performed directly on the deployment device for near-sensor computing, validated on sEMG sign language and fault diagnosis datasets, achieving improved accuracy and reduced RAM occupancy.

This paper presents an automated pipeline for searching heterogeneous 4-expert Mixture-of-Experts architectures, exploring 4.8% of the theoretical combination space and identifying high- and low-yield expert families. The work releases analysis artifacts and a corrected generator as part of the open-source NNGPT project.

Researchers from UiT and University of Oslo propose a differentiable NAS framework that jointly optimizes architectural configurations and mixed-precision quantization for LLM compression, achieving up to 1.4× faster inference or 6% higher accuracy across seven reasoning tasks compared to sequential NAS-then-quantization baselines.

Meta's new paper presents an agentic system that autonomously discovers neural architectures outperforming Llama 3.2 at 350M, 1B, and 3B scales within a 24-hour compute budget.

This paper introduces AIRA-Compose and AIRA-Design, dual frameworks using AI agents to autonomously discover neural architectures that outperform standard Transformers and scale efficiently.

This paper introduces LSAMD, a method for extracting 'learngenes' across multiple datasets to initialize variable-sized Vision Transformer models, significantly reducing training costs and storage while maintaining performance comparable to pretrain-finetune methods.

RF-DETR introduces a lightweight detection transformer that uses weight-sharing neural architecture search to achieve state-of-the-art real-time object detection, outperforming prior methods on COCO and Roboflow100-VL while running up to 20x faster.