cs.LGNov 17, 2026

What Does Weight-Space Programming Add to Diffusion Editing? Program Construction versus Score-Guided Refinement

Authors: Zitian Gao, Jingyi Xu, Ziqi Han, Chengkai Lei, Ye Liu, Liangliang Shi

Organizations: Fudan University · Northwest University · Tencent · Shanghai Institute for Mathematics and Interdisciplinary Sciences

Abstract

Weight-space programming (WSP) lets a language model construct executable weight updates for another network. We study diffusion concept editing through an interface in which programs define training contexts, preservation constraints and update directions, while a shared numerical fitter selects their coefficients. We compare generated programs with published editors, a shared-panel hand-written baseline and budget-matched search over conventional edits on 24 TIMED test edits for Stable Diffusion 1.4. The preregistered rewrite pipeline shows no advantage over the strong baseline. In an exploratory comparison, one-shot Astra programs improve fresh-scene VLM joint success, which requires subject presence together with whole-image target matching, by +10.0 points over the shared-panel baseline, while CLIP relative match shows no detectable change. These programs organise adjustable directions around protected text-state subspaces. Probing and rewriting the Astra programs then raises CLIP by +6.2 points while joint success changes by -9.7. On a fixed 144-image sample, five additional evaluators give point estimates in the same directions under whole-image target matching; checks of subject-local change give less decisive differences. Together, these results identify useful behaviour in complete generated programs and show that score-guided revision need not preserve it: program construction and revision should be evaluated as distinct stages of a WSP pipeline.

Figures & tables

Appendix figures & tables20 assets

Supplementary material from the paper’s appendix.

Appendix

Explore similar work

Nov 3, 2026

Weight-Space Programming for Improving Image-Restoration Diffusion Models: Technical Conjectures and a Recursive Improvement Loop

Can an LLM directly edit the weights of a diffusion-based image-restoration model to improve restoration quality---without gradients, paired data, or fine-tuning? Recent experiments show frontier LLMs can fix structural bugs by modifying as little as 0.016%0.016\% of parameters, but cannot instill new capabilities. We ask whether weight-space programming (WSP) can move from debugging to improvement in the precise-regression setting of image restoration (deblurring, super-resolution, denoising, inpainting). Restoration diffusion UNets present a middle ground: low-level signal-processing layers are concentrated enough for targeted weight-direction edits, while high-level semantic layers remain distributed and resistant to sparse modification. We propose three technical conjectures: (C1) an LLM can localize the layer behind a quality gap (over-smoothing, color shift, hallucinated texture) within a 200 forward-pass budget; (C2) editing a small set of identified weight directions (e.g., scaling high-frequency filter norms, correcting output-bias channels) can measurably improve PSNR and perceptual quality; (C3) a recursive loop---diagnose →\rightarrow edit →\rightarrow evaluate →\rightarrow re-diagnose---can compound gains across rounds, forming a minimal recursive self-improvement (RSI) loop for vision. We analyze the required AI capability (arguing targeted edits need not superintelligence, while general capability generation does) and detail the challenges of precise pixel-level regression. We provide a PyTorch-style deployment pipeline and outline a benchmark. This manuscript is conjectural, intended to open a new subtopic for image restoration.
Nov 6, 2026

Exploring Weight-Space Programming: Capabilities, Generalization, and Limitations

Weight-Space Programming (WSP) enables reasoning agents to directly inspect and modify neural network parameters, offering an alternative to gradient-based model adaptation. However, its capabilities, cross-task generalization, and practical limitations remain insufficiently understood. We systematically investigate WSP through 16 editing trajectories across seven benchmarks and two model scales (Qwen3.5-0.8B and Qwen3.5-9B). Our findings reveal three key insights. First, WSP can achieve substantial performance gains through highly localized interventions. On full held-out evaluations, accuracy improves by up to 29.45, 4.42, 36.72, and 17.11 percentage points on MATH-500, GPQA Diamond, IFEval, and SciBench, respectively. Notably, modifying only 1,023 parameters yields a 29.45-point improvement. Second, WSP exhibits promising but uneven cross-task generalization. A Math-edited checkpoint improves GPQA Diamond, IFEval, and SciBench by 3.03, 12.18, and 20.10 points, respectively, without further editing, while leaving SciCode performance unchanged. Third, effective editing remains challenging and inconsistent. Outcomes vary substantially across trajectories, and six of the 16 trajectories retain the original weights when candidate improvements cannot be reliably validated. Together, these results highlight WSP's potential for highly localized model adaptation while revealing important challenges in intervention discovery, reliability, and cross-task transfer. Homepage: https://wsp-scope.github.io/ GitHub: https://github.com/WSP-Scope/WSP-Scope-Project
Nov 16, 2026

Weight-Space Programming Changes How an LLM Uses Python for Computational Biology

Weight-space programming allows a language model to edit another model's weights without gradient-based training. We used Claude Opus 5.5 to edit Qwen3.5-0.8B and investigated how the edits changed its use of Python on computational biology problems. We compared four initial edited checkpoints with the original model on 50 test problems across three test seeds, yielding 600 paired evaluations. Strictly correct answers increased from 44 to 77, and technically correct answers, which included correct final values rejected for formatting, increased from 64 to 100. Repeated Python calls decreased from 1,834 to 1,044. Among the 64 pairs in which only the edited response was technically correct, the original model had already computed the correct value in 36 but did not submit a technically correct final answer. The edited model corrected the computation in 13 of the 64 pairs. With six additional checkpoints on the same 50 problems, technically correct answers rose from 96 to 154 across 900 pairs. We explored three selected problems with ten test seeds for each of ten checkpoints. On the melting-temperature problem, correct Python outputs decreased from 60 to 22 while technically correct final answers increased from 0 to 18. Weight editing changed both the calculations Qwen ran and whether it submitted a correct answer.