cs.CVAug 14, 2026

Consensus-gated Multi-Agent Neural Architecture Search for Seismic Fault Segmentation

Authors: Shehram Baig, Ahmad Mustafa

Organizations: Information Technology University (ITU), Lahore, Pakistan · King Fahd University of Petroleum and Minerals, Dhahran, Saudi Arabia

Read PDF
Ask AI
Cite

Copy citation

DOINot available for this paper

Abstract

Neural networks for seismic fault segmentation are often borrowed from computer vision and medical imaging domains where they train under relatively much larger labeled data resources. Optimizing their architecture under tight labeled data budgets as are common in geophysical applications is not a trivial problem. Manually designing data-optimal architectures is time-consuming while classical neural architecture search (NAS) is restricted to hand-crafted search spaces and large compute budgets. We present an agentic NAS system in which a panel of three large language models (Claude, GPT-5.1, and Gemini2.5Pro) debates each candidate architecture to unanimous consensus, authors the complete PyTorch implementation, cross-reviews it, and submits it to an automated validate-train-score loop with a hard 450K parameter budget, keep-or-revert lineage, and a memory of failed mechanisms. Operating on source code rather than a predefined operation menu, the search ran on a single consumer GPU and trained only eight candidates. It discovered \ours{}: a 425K-parameter encoder-decoder with a strip-pooling bottleneck, squeeze-and-excitation gating, an asymmetric one-conv decoder, and a feature-pyramid fusion neck. Trained under a protocol identical to all baselines on sections derived from the Thebe fault dataset, it attains the highest F1 (0.578) and IoU of all models tested while being the smallest, outperforming a published-capacity U-Net (31M parameters, F1 0.484), DeepLabV3-ResNet50 (39.6M, 0.516), an Attention U-Net(1.83M, 0.502). The search cost 101 LLM calls (∼\sim1.15M input / 0.39M output tokens) and roughly one GPU-day, making consensus-gated LLM panels a practical, low-cost route to domain-specific architecture discovery.

Figures & tables

Fig. 1: One search iteration. Training, the expensive step, is spent only on designs that survived unanimous three-model consensus, cross-review, and the pre-training validator.
Fig. 2: The discovered StripPool-UNet+FPN architecture (425,177 parameters): a narrow SE-gated encoder (16/32/48/64/80 channels), a strip-pooling bottleneck at H/16H/16 , an asymmetric single-conv decoder, and an FPN neck that fuses the four decoder scales into the map read by both output heads (Sec. IV). The output panel overlays predicted faults (red, threshold 0.5) on a test section.
Fig. 3: Round-4 search trajectory. Top: test F1 under the 5-section proxy; each marker is the mean over 3 trials, the whisker rises to the best single trial; a long whisker flags a seed-sensitive design (e.g. run 8). Filled markers were kept, open markers reverted; the gate compares means against the current parent (the running-best line), not the baseline, which is why runs 7–9 clear the dashed baseline yet were rejected. Run 6 aborted during the debate on a provider transport failure. Bottom: parameter count against the 450K budget; the 1.49M reference baseline is exempt.
runmutationparamsmean F1best F1verdict
0baseline (reference)1,492,7630.49230.5033ref.
1depthwise-sep. + ASPP-lite430,1190.47400.4941revert
2strip-pooling U-Net420,2170.50170.5085keep
3+ FPN fusion neck425,1770.51760.5294keep
4+ per-scale strip gates434,8730.49590.5073revert
5+ coordinate priors425,7530.43900.4652revert
TABLE I: Final search round: one debated mutation per run (proxy protocol, mean/best F1 over 3 trials).
modelparamsF1IoUPrec.Rec.PR-AUC
StripPool-UNet+FPN (ours)0.43M0.5780.4060.4850.7150.863
DeepLabV3-R50 [ 6 ]39.63M0.5160.3480.4500.6070.824
Attention U-Net [ 21 ]1.83M0.5020.3360.3710.7770.852
U-Net (1.8M) [ 5 ]1.81M0.4940.3280.3720.7370.829
U-Net (published) [ 5 ]31.04M0.4840.3190.3810.6630.813
TABLE II: Full-data benchmark: every model trained under the identical protocol on Thebe-derived sections; test-volume metrics, F1-sorted.
Fig. 4: Test-set fault predictions (threshold 0.5, identical protocol) on the test section where the discovered network’s advantage over the published baselines is largest (per-slice F1 in each panel). Left to right: ground truth; published-capacity U-Net (31.0M, F1 0.466); DeepLabV3-ResNet50 (39.6M, 0.488); discovered StripPool-UNet+FPN (0.43M, 0.609). The large models draw thick, blobby responses or fragmented strands; the compact network draws thin, continuous, well-separated strands tracking the ground-truth geometry, and attains the best per-slice F1 on 96 of the 100 test sections.
More in this topic

Neural Architecture Search

Neural Architecture Search automates neural network design for target tasks, resource budgets, and hardware constraints.

See all 132 papers

Explore similar work

CardsList
Jul 8, 2026

Agentic Neural Architecture Search

Neural architecture search (NAS) methods have grown increasingly efficient, yet they remain bounded by manually engineered search spaces that require substantial domain expertise and must be rebuilt for every new task. Large language models (LLMs) can generate architectures in an open-ended space, but how to optimally divide the labor between LLM-driven design and NAS-driven search remains unexplored. We propose a mechanism that bridges these two paradigms: an LLM produces a high-quality seed architecture, then decomposes it into a "slotted architecture", a scaffold with named, interchangeable module slots that automatically defines a bounded, task-specific search space for conventional NAS to explore, without manual engineering. We instantiate this mechanism in AgentNAS, a modular three-phase pipeline in which each component's contribution can be measured independently. On 17 tasks spanning classification, dense regression, segmentation, and multi-label tagging across diverse modalities (NAS-Bench-360 and Unseen NAS), AgentNAS establishes a new state of the art on 11 tasks, outperforming published baselines including task-specific expert designs. Ablation studies show that the two search mechanisms are broadly complementary: the LLM-generated seed already surpasses published baselines on the majority of tasks, and NAS delivers additional gains in most cases through combinatorial recombination across slots, a mode of search that independent LLM samples cannot replicate. These patterns hold across three LLMs of different capability levels, confirming that the division of labor is robust. Our code is available at https://github.com/alroimfebruary/AgentNAS.

Seokhoon Jeong, Mijung Kim, Taehwan Kim

Artificial Intelligence Graduate School, Ulsan National Institute of Science and Engineering, Ulsan, Republic of Korea · Department of Computer Science and Engineering, Ulsan National Institute of Science and Engineering, Ulsan, Republic of Korea

Large Language Model-Guided NASLarge Language Model-Guided OptimizationNeural Architecture Search
AskPDF
Mar 12, 2026

Resource-Efficient Iterative LLM-Based NAS with Feedback Memory

Neural Architecture Search (NAS) automates network design, but conventional methods demand substantial computational resources. We propose a closed-loop pipeline leveraging large language models (LLMs) to iteratively generate, evaluate, and refine convolutional neural network architectures for image classification on a single consumer-grade GPU without LLM fine-tuning. Central to our approach is a historical feedback memory inspired by Markov chains: a sliding window of K=5K{=}5 recent improvement attempts keeps context size constant while providing sufficient signal for iterative learning. Unlike prior LLM optimizers that discard failure trajectories, each history entry is a structured diagnostic triple -- recording the identified problem, suggested modification, and resulting outcome -- treating code execution failures as first-class learning signals. A dual-LLM specialization reduces per-call cognitive load: a Code Generator produces executable PyTorch architectures while a Prompt Improver handles diagnostic reasoning. Since both the LLM and architecture training share limited VRAM, the search implicitly favors compact, hardware-efficient models suited to edge deployment. We evaluate three frozen instruction-tuned LLMs (≤7{\leq}7B parameters) across up to 2000 iterations in an unconstrained open code space, using one-epoch proxy accuracy on CIFAR-10, CIFAR-100, and ImageNette as a fast ranking signal. On CIFAR-10, DeepSeek-Coder-6.7B improves from 28.2% to 69.2%, Qwen2.5-7B from 50.0% to 71.5%, and GLM-5 from 43.2% to 62.0%. A full 2000-iteration search completes in ≈18{\approx}18 GPU hours on a single RTX~4090, establishing a low-budget, reproducible, and hardware-aware paradigm for LLM-driven NAS without cloud infrastructure.

Xiaojie Gu, Dmitry Ignatov, Radu Timofte

Computer Vision Lab, CAIDAS & IFI, University of Würzburg, Germany

Large Language Model-Guided NASNeural Architecture SearchLarge Language Model-Guided OptimizationHardware-Aware NAS
AskPDF
Apr 17, 2026

LLM as a Tool, Not an Agent: Code-Mined Tree Transformations for Neural Architecture Search

Neural Architecture Search (NAS) aims to automatically discover high-performing deep neural network (DNN) architectures. However, conventional algorithm-driven NAS relies on carefully hand-crafted search spaces to ensure executability, which restricts open-ended exploration. Recent coding-based agentic approaches using large language models (LLMs) reduce manual design, but current LLMs struggle to reliably generate complex, valid architectures, and their proposals are often biased toward a narrow set of patterns observed in their training data. To bridge reliable algorithmic search with powerful LLM assistance, we propose LLMasTool, a hierarchical tree-based NAS framework for stable and open-ended model evolution. Our method automatically extracts reusable modules from arbitrary source code and represents full architectures as hierarchical trees, enabling evolution through reliable tree transformations rather than code generation. At each evolution step, coarse-level planning is governed by a diversity-guided algorithm that leverages Bayesian modeling to improve exploration efficiency, while the LLM resolves the remaining degrees of freedom to ensure a meaningful evolutionary trajectory and an executable generated architecture. With this formulation, instead of fully agentic LLM approaches, our method explores diverse directions beyond the inherent biases in the LLM. Our method improves over existing NAS methods by 0.69, 1.83, and 2.68 points on CIFAR-10, CIFAR-100, and ImageNet16-120, demonstrating its effectiveness.

Masakazu Yoshimura, Zitang Sun, Yuiko Sakuma +3

Sony Group Corporation, Tokyo, Japan

Large Language Model-Guided NASLarge Language Model-Guided OptimizationNeural Architecture Search
AskPDF