LLM-Driven Neural Network Generation with Same-Family Architecture Guidance: Disentangling Transfer and Adaptation

2026-07-06

AI Summary

This work addresses the challenges of large language models (LLMs) generating invalid or detrimental modifications when designing neural architectures and their limited ability to leverage knowledge from related model families. To overcome these issues, the authors propose a source-guided candidate generation mechanism that employs a strong source model to guide structural refinements of a weaker target model, thereby enhancing performance while ensuring architectural validity. The approach explicitly separates knowledge transfer from adaptation and, for the first time, systematically demonstrates under equal evaluation budgets that LLMs can effectively adapt network structures rather than merely replicating them. Experiments using DeepSeek-Coder-6.7B, a database of related architectures, and controlled designs show significant improvements over non-source-guided methods, achieving an AlexNet accuracy of 0.8069 on SVHN—over 0.55 higher than the baseline—with consistent gains across most architecture families on both CIFAR-10 and SVHN.

Abstract

Large language models (LLMs) can generate neural-network modifications, but unrestricted generation is often invalid or harmful. This paper studies a narrower setting: improving a weak target model using a stronger same-family source model from a neural-network database. We propose a source-guided candidate-generation protocol with non-source controls, source-conditioned candidates, and a no-LLM hp_copy ablation under equal evaluation budgets. The protocol reports validity separately from accuracy and selects the best valid candidate only when it improves the target. On CIFAR-10, the strongest source-guided candidate reaches 0.5049 accuracy versus 0.2398 for the best non-source candidate, a +0.2651 advantage, while improving a weak target originally at 0.1254; a five-epoch check preserves the gain at 0.7686 versus 0.4839. On SVHN AlexNet with DeepSeek-Coder-6.7B, source-guided transfer reaches 0.7880 versus 0.2254, a +0.5626 advantage; a fresh repeat reaches 0.8069 versus 0.2509, a +0.5560 advantage. Direct source-recipe copy produces 0.1959 on SVHN AlexNet, matching the original target, while hp_transfer reaches 0.7880, showing that the LLM adapts rather than copies the source recipe. Family-level analysis shows the clearest positive signals for AlexNet, with 6/8 wins across SVHN, Imagenette, and CelebA-Gender, and alt_nn1, with 8/10 wins on CIFAR-10.

Problem

• LLM-driven neural network generation
• same-family architecture guidance
• transfer and adaptation disentanglement
• validity of generated models
• neural network improvement

Innovation

• LLM-driven generation
• same-family architecture guidance
• neural network transfer
• adaptation vs. copying
• validity-aware candidate selection