ProCerDCL is a query‑efficient framework for black‑box probabilistic verification of deep learning compiler optimizations. It uses a single sequential test to decide whether the violation probability of a binary property oracle is below a threshold, above a higher threshold plus a margin, or inconclusive, with bounded error probabilities. Applied to TorchInductor and TensorRT, it achieves higher sampling efficiency and provides diagnostic verdicts on model semantic consistency across 24 tasks.
By Zexin Li, Weibo Zhang, Linyi Li, Arvind Easwaran, Simin Chen
arXiv:2512. 02080v3 Announce Type: replace-cross Abstract: The integration of Formal Verification tools with Large Language Models (LLMs) offers a path to scale software verification beyond manual workflows.
By Pierre Dantas, Lucas Cordeiro, Youcheng Sun, Waldir Junior
arXiv:2608. 01575v1 Announce Type: new Abstract: Whether large language models perform genuine algorithmic reasoning or mere pattern completion is hard to test, because most benchmarks lack a ground truth for correct inductive inference.
By Hector Zenil, Luan Ozelim
Whether large language models perform genuine algorithmic reasoning or mere pattern completion is hard to test, because most benchmarks lack a ground truth for correct inductive inference. We introduce F-ICL, an in-context-learning benchmark that supplies one exactly.
arXiv:2607. 29431v1 Announce Type: new Abstract: Large language models increasingly generate optimization models from natural language, but existing evaluation often reduces a generated model and its ground truth to a single equivalent/not-equivalent verdict or an execution-success rate--labels that are neither independently checkable nor faithful to the multiple distinct senses in which two formulations can agree.
By Penglin Zhu, Jungang Xu
arXiv:2607. 04562v1 Announce Type: new Abstract: Large language models (LLMs) generate fluent outputs that can be wrong.
By MY Pitsane, Hope Mogale
arXiv:2606. 01682v1 Announce Type: cross Abstract: Selecting the best response from multiple small-model samples using a stronger scorer is a simple inference-time strategy, but fails when the small model has already committed to incorrect reasoning paths.
By Atoosa Chegini, Soheil Feizi
The paper introduces Adaptive Reference Guidance (ARG), a method that balances using a verified reference solution with free generation in trajectory rollouts for reasoning reinforcement learning. ARG selects a prefix of the reference to guide the model, continuing from it only if the resulting trajectory passes verification, otherwise falling back to the reference. Experiments on Qwen3-4B and Qwen3-8B across five mathematical reasoning benchmarks demonstrate that ARG achieves the highest aggregate pass@12 while maintaining competitive average sampled accuracy.
By Hanyu Wang, Nakul Agarwal, Hossein Nourkhiz Mahjoub, Ehsan Moradi Pari, Makoto Fukushima, Jinghui Chen, Vaishnav Tadiparthi
arXiv:2606. 17529v1 Announce Type: cross Abstract: Scientific machine-learning (SciML) surrogates approximate expensive simulations, but exact expected outputs for arbitrary inputs are unavailable (the oracle problem).
By Meng Li, Xiaohua Yang, Jie Liu, Shiyu Yan
arXiv:2608. 12426v1 Announce Type: new Abstract: Large language models are increasingly deployed in settings that require simultaneous adherence to multiple explicit constraints - reasoning structure, safety boundaries, output schemas.
By Mariya I. Vasileva
Selecting the best response from multiple small-model samples using a stronger scorer is a simple inference-time strategy, but fails when the small model has already committed to incorrect reasoning paths. PRM guided search avoids this by scoring candidate continuations during generation, but requires a reward model trained with step-level labels.
PRISM is a new inference algorithm for DEEPTHINK systems that uses a Process Reward Model (PRM) to guide step‑level verification, resampling, and aggregation of candidate solutions. By concentrating computation on higher‑quality reasoning while preserving diversity, PRISM improves performance on complex mathematical and scientific tasks, achieving 90.0%, 75.4%, and 71.4% on AIME25, HMMT25, and GPQA Diamond with gpt‑oss‑20b and outperforming larger models. The method also provides consistent net‑directional correction during refinement and remains accurate even when the initial population has few correct candidates.
By Rituraj Sharma, Weiyuan Chen, Noah Provenzano, Tu Vu