Scaling Closed-Loop Feature Channel Configuration with LLMs

A new arXiv preprint (2607.20516v1) reports that scaling closed-loop LLM-based channel-configuration search to 250 candidate networks per cycle yields a positive linear accuracy trend (slope 9.87e-4, p=0.043) over 2000 generated candidates from 8 cycles, with best accuracy improving from 0.3144 to 0.3676 on CIFAR-100. The scaled run also reveals improved parameter efficiency, with the best model achieving 0.3676 accuracy using 11.8M parameters versus an early model at 0.3144 with 166.5M parameters, and exposes architectural regularities including non-power-of-two channel widths in 41.8% of verified candidates.

arXiv:2607.20516v1 Announce Type: new Abstract: Promising initial results in closed-loop large-language-model-based channel-configuration search demonstrated that neural-network widths can be optimized directly through executable code generation and accuracy feedback. However, those results were obtained from a relatively sparse set of valid evaluations, leaving open whether the observed optimization behavior transfers to a denser sampling regime and whether additional architectural regularities emerge when more generated networks are evaluated. To test this, the same search setting is scaled to 250 candidate networks per fine-tuning cycle. The analysis covers 2000 generated candidates from 8 complete cycles, yielding 462 verified CIFAR-100 evaluations after task and metadata filtering. Per-cycle mean accuracy exhibits a positive linear trend with slope 9.87e-4 p=0.043 , while the high-performing frontier improves more strongly: the best observed accuracy increases from 0.3144 to 0.3676, and both the top-5 and top-10 cycle-level means exhibit positive trends. The scaled run also reveals improved parameter efficiency. The best model reaches 0.3676 with 11.8M parameters, compared with an early high-performing model at 0.3144 with 166.5M parameters. Beyond accuracy, the larger sample exposes architectural regularities that were difficult to assess from sparse observations. Non-power-of-two channel widths occur in 41.8% of verified candidates, and the strongest models share structured channel-allocation patterns characterized by moderate early widths and expanded middle or later blocks. These findings indicate that the channel-search signal observed in the initial study transfers