arXiv subjects

Tony Yu

Publications and source records attributed to Tony Yu.

6 recordsLinked to original sources

arXiv2026

RUBRIC-ARROW: Alternating Pointwise Rubric Reward Modeling for LLM Post-training in Non-verifiable Domains

Pointwise reward modeling offers critical signals for LLM post-training, yet struggles with absolute scoring in subjective, non-verifiable settings. Rubric-based methods address this by decomposing evaluation into explicit criteria, but existing approaches typically depend on frontier LLMs and suffer from ties caused by hard Boolean aggregation. We present RUBRIC-ARROW, an alternating framework that jointly trains a rubric generator and a rubric-conditioned judge, with its RL stage using only pairwise preference data. Our method couples a probability-based scoring rule that reduces ties with phase-specific preference-based rewards and an alternating GRPO scheme that together train the pointwise evaluator. Extensive experiments show that RUBRIC-ARROW achieves competitive reward-modeling accuracy and yields consistent gains for downstream policy post-training.

cs.LG↗

arXiv2026

Alternating Reinforcement Learning for Rubric-Based Reward Modeling in Non-Verifiable LLM Post-Training

Standard reward models typically predict scalar scores that fail to capture the multifaceted nature of response quality in non-verifiable domains, such as creative writing or open-ended instruction following. To address this limitation, we propose Rubric-ARM, a framework that jointly optimizes a rubric generator and a judge using reinforcement learning from preference feedback. Unlike existing methods that rely on static rubrics or disjoint training pipelines, our approach treats rubric generation as a latent action learned to maximize judgment accuracy. We introduce an alternating optimization strategy to mitigate the non-stationarity of simultaneous updates, providing theoretical analysis that demonstrates how this schedule reduces gradient variance during training. Extensive experiments show that Rubric-ARM achieves state-of-the-art performance among baselines on multiple benchmarks and significantly improves downstream policy alignment in both offline and online reinforcement learning settings.

cs.CL↗

arXiv2025

OpenRubrics: Towards Scalable Synthetic Rubric Generation for Reward Modeling and LLM Alignment

Reward modeling lies at the core of reinforcement learning from human feedback (RLHF), yet most existing reward models rely on scalar or pairwise judgments that fail to capture the multifaceted nature of human preferences. Recent studies have explored rubrics-as-rewards (RaR) that uses structured criteria to capture multiple dimensions of response quality. However, producing rubrics that are both reliable and scalable remains a key challenge. In this work, we introduce OpenRubrics, a diverse, large-scale collection of (prompt, rubric) pairs for training rubric-generation and rubric-based reward models. To elicit discriminative and comprehensive evaluation signals, we introduce Contrastive Rubric Generation (CRG), which derives both hard rules (explicit constraints) and principles (implicit qualities) by contrasting preferred and rejected responses. We further remove noisy rubrics via preserving preference-label consistency. Across multiple reward-modeling benchmarks, our rubric-based reward model, Rubric-RM, surpasses strong size-matched baselines by 8.4%. These gains transfer to policy models on instruction-following and biomedical benchmarks.

cs.CL↗

arXiv2014

scikit-image: Image processing in Python

scikit-image is an image processing library that implements algorithms and utilities for use in research, education and industry applications. It is released under the liberal "Modified BSD" open source license, provides a well-documented API in the Python programming language, and is developed by an active, international team of collaborators. In this paper we highlight the advantages of open source to achieve the goals of the scikit-image library, and we showcase several real-world image processing applications that use scikit-image.

cs.MS↗

arXiv2009

Ionic high-pressure form of elemental boron

Boron is an element of fascinating chemical complexity. Controversies have shrouded this element since its discovery was announced in 1808: the new 'element' turned out to be a compound containing less than 60-70 percent of boron, and it was not until 1909 that 99-percent pure boron was obtained. And although we now know of at least 16 polymorphs, the stable phase of boron is not yet experimentally established even at ambient conditions. Boron's complexities arise from frustration: situated between metals and insulators in the periodic table, boron has only three valence electrons, which would favour metallicity, but they are sufficiently localized that insulating states emerge. However, this subtle balance between metallic and insulating states is easily shifted by pressure, temperature and impurities. Here we report the results of high-pressure experiments and ab initio evolutionary crystal structure predictions that explore the structural stability of boron under pressure and, strikingly, reveal a partially ionic high-pressure boron phase. This new phase is stable between 19 and 89 GPa, can be quenched to ambient conditions, and has a hitherto unknown structure (space group Pnnm, 28 atoms in the unit cell) consisting of icosahedral B12 clusters and B2 pairs in a NaCl-type arrangement. We find that the ionicity of the phase affects its electronic bandgap, infrared adsorption and dielectric constants, and that it arises from the different electronic properties of the B2 pairs and B12 clusters and the resultant charge transfer between them.

cond-mat.mtrl-sci↗

arXiv2009

New high-pressure form of boron is significantly ionic

The comment of Dubrovinskaia et al. is scientifically flawed. The high-pressure form of boron, discovered by Oganov et al., is indeed new and its bonding has a significant ionic character, as demonstrated in Ref. 1.

cond-mat.mtrl-sci↗