arXiv:2601.02998v1 Announce Type: cross Abstract: In many fairness and distribution robustness problems, one has access to labeled data from multiple source distributions yet the test data may come from an arbitrary member or a mixture of them. We study the problem of constructing a conformal prediction set that is uniformly valid across multiple, heterogeneous distributions, in the sense that no matter which distribution the test point is from, the coverage of the prediction set is guaranteed to exceed […]

Ver mais

arXiv:2602.04930v1 Announce Type: new Abstract: Future AI deployments will likely be monitored for malicious behaviour. The ability of these AIs to subvert monitors by adversarially selecting against them – attack selection – is particularly concerning. To study this, we let a red team create attack policies that attempt to insert attacks into code without being caught by a monitor in the concentrated BigCodeBench backdooring setting. We decompose attack selection into two problems: mapping attacks to a quality score […]

Ver mais

arXiv:2602.18582v1 Announce Type: new Abstract: When training artificial intelligence (AI) to perform tasks, humans often care not only about whether a task is completed but also how it is performed. As AI agents tackle increasingly complex tasks, aligning their behavior with human-provided specifications becomes critical for responsible AI deployment. Reward design provides a direct channel for such alignment by translating human expectations into reward functions that guide reinforcement learning (RL). However, existing methods are often too limited to […]

Ver mais

arXiv:2602.15162v1 Announce Type: new Abstract: Mobile robots operating in agroindustrial environments, such as Mediterranean greenhouses, are subject to challenging conditions, including uneven terrain, variable friction, payload changes, and terrain slopes, all of which significantly affect control performance and stability. Despite the increasing adoption of robotic platforms in agriculture, the lack of standardized, reproducible benchmarks impedes fair comparisons and systematic evaluations of control strategies under realistic operating conditions. This paper presents a comprehensive benchmarking framework for evaluating mobile robot […]

Ver mais

Hey everyone, Meta-reviews just dropped. My paper got two weak rejects and a borderline accept (got dinged for missing some VLM baselines), but the AC recommended it to the new “Findings” track after the AC triplet meeting (not sure what this is). For context, I’m a solo undergrad working entirely without a supervisor. I don’t have a PI or a lab to ask about how this stuff works, so my only source of info is whatever I can […]

Ver mais

arXiv:2601.19020v1 Announce Type: new Abstract: This paper presents an adaptive numerical framework for solving exterior “sound-soft” scattering problems governed by the Helmholtz equation. By interpreting the Method of Fundamental Solutions through the lens of rational approximation, we introduce an automated strategy for singularity placement based on the analytic continuation of boundary data. The proposed AAALS-Helmholtz algorithm leverages a “continuum” variant of the AAA algorithm to identify the singularities limiting analytic extension, and to ensure an optimal source distribution […]

Ver mais

Parallelism, kernels, and why memory matters most Being a PhD student working on AI and NLP, I’ve spent quite some time using PyTorch and other high-level frameworks that abstract away the GPU. But recent discussions about whether I should learn CUDA pushed me to step back and revisit the basics: where all of this started, and why it changed. This isn’t an official “learning guide” or a course roadmap. It’s going back to first principles and sharing the insights I’m learning […]

Ver mais

arXiv:2602.03979v1 Announce Type: new Abstract: Fine-tuning large language models (LLMs) on reasoning benchmarks via reinforcement learning requires a specific reward function, often binary, for each benchmark. This comes with two potential limitations: the need to design the reward, and the potentially sparse nature of binary rewards. Here, we systematically investigate rewards derived from the probability or log-probability of emitting the reference answer (or any other prompt continuation present in the data), which have the advantage of not relying […]

Ver mais