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

A new arXiv preprint (2608.13889v1) presents a consensus-gated multi-agent neural architecture search (NAS) system that uses three large language models (Claude, GPT-5.1, and Gemini 2.5 Pro) to debate and author candidate architectures, discovering a 425K-parameter encoder-decoder for seismic fault segmentation that achieves the highest F1 (0.578) and IoU among tested models, outperforming a 31M-parameter U-Net (F1 0.484), DeepLabV3-ResNet50 (39.6M, F1 0.516), and Attention U-Net (1.83M, F1 0.502) on the Thebe fault dataset. The search cost 101 LLM calls (~1.15M input / 0.39M output tokens) and roughly one GPU-day, demonstrating a low-cost route to domain-specific architecture discovery.

arXiv:2608.13889v1 Announce Type: new 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 Gemini~2.5~Pro 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 $\sim$1.15M 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.