Leaderboards, eval harnesses and ablations — the contested business of deciding which model is actually better.
arXiv:2610.29294v1 Announce Type: cross Abstract: Large language models (LLMs) are increasingly used to support consequential decisions, yet their reliability under equivalent representations of unce...
arXiv:2610.29308v1 Announce Type: cross Abstract: Dynamic scheduling requires dispatching online arrivals while anticipating future jobs, as poor decisions can compound downstream costs under nonline...
arXiv:2610.29812v1 Announce Type: cross Abstract: Literature-searching language agents are becoming a standard component of scientific assistants for the life sciences, yet there is little controlled...
The paper introduces AnoMix, a method that distills the complementary anomaly scoring principles of multiple unsupervised time‑series detectors into a single lightweight model. By normalizing each teacher’s score sequence and applying a multi‑teacher ranking‑distillation objective called CoRank, AnoMix learns a student detector that uses pairwise supervision derived from teacher consensus. Experiments on seven benchmark datasets show that the compact student (under 80k parameters) outperforms individual detectors and static ensembles, and a zero‑shot variant can even surpass large pretrained time‑series foundation models on unseen data.
arXiv:2610.29090v1 Announce Type: new Abstract: Abdominal organ segmentation is essential in modern medical applications. While existing deep learning approaches demonstrate promising performance, th...
arXiv:2610.29098v1 Announce Type: new Abstract: One-shot landmark localization in volume scans with varying fields of view (FOVs) is challenging. While existing self-supervised learning methods have...
arXiv:2610.29075v1 Announce Type: new Abstract: We aim to address the problem of Out-of-Distribution (OOD) Mode Averaging, a critical problem that emerges when distilling Flow Matching (FM) based beh...
arXiv:2610.29188v1 Announce Type: new Abstract: Tables in real-world documents, including scientific papers, financial reports, and statistical summaries, frequently use merged cells, multi-level hea...
arXiv:2610.29249v1 Announce Type: new Abstract: Omni-modal assistants must remember not only what users say, but also what they show and convey. Existing long-term memory benchmarks are mostly text-a...
arXiv:2610.29413v1 Announce Type: new Abstract: Recently, we see a rise in conversational agents as a form of social and emotional support, where disclosure will unfold over time. Yet, current evalua...
arXiv:2610.29507v1 Announce Type: new Abstract: Temporal reasoning is widely assumed to decompose into two modes: temporal arithmetic (date and time calculations), and temporal semantics (understandi...
arXiv:2610.29518v1 Announce Type: cross Abstract: While Chain-of-Thought (CoT) prompting improves the reasoning performance of large language models (LLMs), recent studies have shown that final answe...
arXiv:2610.29538v1 Announce Type: new Abstract: Direct preference optimization (DPO) incurs training costs in output scoring, sequence execution, and backpropagation, so reducing supervision alone do...
arXiv:2610.29722v1 Announce Type: new Abstract: Retrieval Augmented Generation (RAG) grounds large language models in external knowledge. Modern pipelines combine embedding, chunking, retrieval, rera...
The paper argues that a benchmark is more than a set of test questions; it defines an empirical distribution over difficulty. It identifies two failure modes—measure mismatch, where the benchmark’s difficulty distribution diverges from the intended one, and coverage gaps, where parts of the difficulty spectrum are missing. The authors propose diagnostic metrics with theoretical guarantees, analyze 25 benchmarks across seven domains, and introduce BenchFill, an algorithm that rewrites items to fill coverage gaps.
The paper introduces a differentially private framework for sparse reward estimation in reinforcement learning with human feedback (RLHF). It assumes that only a small subset of features influences human preferences and provides theoretical error bounds for both Sample-DP and Label-DP settings under the Bradley-Terry-Luce model. Empirical results show that the proposed methods outperform dense baselines on sentiment generation and safety alignment tasks, achieving a better privacy‑utility trade‑off.
BatchBench is a new leaderboard designed to assess AI robustness to batch effects and domain shift in biomedical images. It evaluates performance on unseen experimental batches and data sources using the large JUMP-CP dataset, which is generated via the Cell Painting assay and exhibits pronounced batch effects. The benchmark aims to highlight remaining performance gaps despite recent domain adaptation advances.
arXiv:2610.24899v2 Announce Type: replace-cross Abstract: While LLM agents are increasingly capable of gathering and synthesizing evidence from a single homogeneous source, such as structured tables...
arXiv:2503.00128v2 Announce Type: replace Abstract: Language models offer a way to scale costly, labour-intensive expert annotation, but their ability to reproduce expert judgements remains uncertain...
arXiv:2508.10954v3 Announce Type: replace Abstract: Medical AI systems must remain reliable as data distributions shift across hospitals, imaging devices, and patient populations, yet privacy constra...