This paper describes our system for the SciVQA 2025 Shared Task on Scientific Visual Question Answering. Our system employs an ensemble of two Multimodal Large Language Models and various few-shot example retrieval strategies. The model and few-shot setting are selected based on the figure and question type. We also select answers based on the models' confidence levels. On the blind test data, our system ranks third out of seven with an average F1 score of 85.12 across ROUGE-1, ROUGE-L, and BERTS. Our code is publicly available.

Software systems must comply with legal regulations, which is a resource-intensive task, particularly for small organizations and startups lacking dedicated legal expertise. Extracting metadata from regulations to elicit legal requirements for software is a critical step to ensure compliance. However, it is a cumbersome task due to the length and complex nature of legal text. Although prior work has pursued automated methods for extracting structural and semantic metadata from legal text, key limitations remain: they do not consider the interplay and interrelationships among attributes associated with these metadata types, and they rely on manual labeling or heuristic-driven machine learning, which does not generalize well to new documents. In this paper, we introduce an approach based on textual entailment and in-context learning for automatically generating a canonical representation of legal text, encodable and executable as Python code. Our representation is instantiated from a manually designed Python class structure that serves as a domain-specific metamodel, capturing both structural and semantic legal metadata and their interrelationships. This design choice reduces the need for large, manually labeled datasets and enhances applicability to unseen legislation. We evaluate our approach on 13 U.S. state data breach notification laws, demonstrating that our generated representations pass approximately 89.4% of test cases and achieve a precision and recall of 82.2 and 88.7, respectively.

Recent advancements in the reasoning capabilities of large language models (LLMs) show that employing group relative policy optimization (GRPO) algorithm for reinforcement learning (RL) training allows the models to use more thinking/reasoning tokens for generating better responses. However, LLMs can generate only a finite amount of tokens while maintaining attention to the previously generated tokens. This limit, also known as the context size of an LLM, is a bottleneck in LLM reasoning with arbitrarily large number of tokens. To think beyond the limit of context size, an LLM must employ a modular thinking strategy to reason over multiple rounds. In this work, we propose $\textbf{MOTIF: Modular Thinking via Reinforcement Finetuning}$ -- an RL training method for generating thinking tokens in multiple rounds, effectively allowing the model to think with additional context size. We trained the open-source model Qwen2.5-3B-Instruct on GSM8K dataset via parameter efficient fine-tuning and tested its accuracy on MATH500 and AIME2024 benchmarks. Our experiments show 3.8\% and 3.3\% improvements over vanilla GRPO based training in the respective benchmarks. Furthermore, this improvement was achieved with only 15\% of samples, thus demonstrating sample efficiency of MOTIF. Our code and models are available at https://github.com/purbeshmitra/MOTIF and https://huggingface.co/purbeshmitra/MOTIF, respectively.

Multiple choice benchmarks have long been the workhorse of language model evaluation because grading multiple choice is objective and easy to automate. However, we show multiple choice questions from popular benchmarks can often be answered without even seeing the question. These shortcuts arise from a fundamental limitation of discriminative evaluation not shared by evaluations of the model's free-form, generative answers. Until recently, there appeared to be no viable, scalable alternative to multiple choice--but, we show that this has changed. We consider generative evaluation via what we call answer matching: Give the candidate model the question without the options, have it generate a free-form response, then use a modern language model with the reference answer to determine if the response matches the reference. To compare the validity of different evaluation strategies, we annotate MMLU-Pro and GPQA-Diamond to obtain human grading data, and measure the agreement of each evaluation approach. We find answer matching using recent models--even small ones--achieves near-perfect agreement, in the range of inter-annotator agreement. In contrast, both multiple choice evaluation and using LLM-as-a-judge without reference answers aligns poorly with human grading. Improving evaluations via answer matching is not merely a conceptual concern: the rankings of several models change significantly when evaluating their free-form responses with answer matching. In light of these findings, we discuss how to move the evaluation ecosystem from multiple choice to answer matching.

Access to high-quality labeled data remains a limiting factor in applied supervised learning. While label variation (LV), i.e., differing labels for the same instance, is common, especially in natural language processing, annotation frameworks often still rest on the assumption of a single ground truth. This overlooks human label variation (HLV), the occurrence of plausible differences in annotations, as an informative signal. Similarly, active learning (AL), a popular approach to optimizing the use of limited annotation budgets in training ML models, often relies on at least one of several simplifying assumptions, which rarely hold in practice when acknowledging HLV. In this paper, we examine foundational assumptions about truth and label nature, highlighting the need to decompose observed LV into signal (e.g., HLV) and noise (e.g., annotation error). We survey how the AL and (H)LV communities have addressed -- or neglected -- these distinctions and propose a conceptual framework for incorporating HLV throughout the AL loop, including instance selection, annotator choice, and label representation. We further discuss the integration of large language models (LLM) as annotators. Our work aims to lay a conceptual foundation for HLV-aware active learning, better reflecting the complexities of real-world annotation.

This paper will discuss how headline data can be used to predict stock prices. The stock price in question is the SPDR S&P 500 ETF Trust, also known as SPY that tracks the performance of the largest 500 publicly traded corporations in the United States. A key focus is to use news headlines from the Wall Street Journal (WSJ) to predict the movement of stock prices on a daily timescale with OpenAI-based text embedding models used to create vector encodings of each headline with principal component analysis (PCA) to exact the key features. The challenge of this work is to capture the time-dependent and time-independent, nuanced impacts of news on stock prices while handling potential lag effects and market noise. Financial and economic data were collected to improve model performance; such sources include the U.S. Dollar Index (DXY) and Treasury Interest Yields. Over 390 machine-learning inference models were trained. The preliminary results show that headline data embeddings greatly benefit stock price prediction by at least 40% compared to training and optimizing a machine learning system without headline data embeddings.

Natural language generation (NLG) is increasingly deployed in high-stakes domains, yet common intrinsic evaluation methods, such as n-gram overlap or sentence plausibility, weakly correlate with actual decision-making efficacy. We propose a decision-oriented framework for evaluating generated text by directly measuring its influence on human and large language model (LLM) decision outcomes. Using market digest texts--including objective morning summaries and subjective closing-bell analyses--as test cases, we assess decision quality based on the financial performance of trades executed by human investors and autonomous LLM agents informed exclusively by these texts. Our findings reveal that neither humans nor LLM agents consistently surpass random performance when relying solely on summaries. However, richer analytical commentaries enable collaborative human-LLM teams to outperform individual human or agent baselines significantly. Our approach underscores the importance of evaluating generated text by its ability to facilitate synergistic decision-making between humans and LLMs, highlighting critical limitations of traditional intrinsic metrics.

Recent progress in large-scale reinforcement learning (RL) has notably enhanced the reasoning capabilities of large language models (LLMs), especially in mathematical domains. However, current multimodal LLMs (MLLMs) for mathematical reasoning often rely on one-to-one image-text pairs and single-solution supervision, overlooking the diversity of valid reasoning perspectives and internal reflections. In this work, we introduce MathV-DP, a novel dataset that captures multiple diverse solution trajectories for each image-question pair, fostering richer reasoning supervision. We further propose Qwen-VL-DP, a model built upon Qwen-VL, fine-tuned with supervised learning and enhanced via group relative policy optimization (GRPO), a rule-based RL approach that integrates correctness discrimination and diversity-aware reward functions. Our method emphasizes learning from varied reasoning perspectives and distinguishing between correct yet distinct solutions. Extensive experiments on the MathVista's minitest and Math-V benchmarks demonstrate that Qwen-VL-DP significantly outperforms prior base MLLMs in both accuracy and generative diversity, highlighting the importance of incorporating diverse perspectives and reflective reasoning in multimodal mathematical reasoning.

Reasoning Language Models (RLMs) have gained traction for their ability to perform complex, multi-step reasoning tasks through mechanisms such as Chain-of-Thought (CoT) prompting or fine-tuned reasoning traces. While these capabilities promise improved reliability, their impact on robustness to social biases remains unclear. In this work, we leverage the CLEAR-Bias benchmark, originally designed for Large Language Models (LLMs), to investigate the adversarial robustness of RLMs to bias elicitation. We systematically evaluate state-of-the-art RLMs across diverse sociocultural dimensions, using an LLM-as-a-judge approach for automated safety scoring and leveraging jailbreak techniques to assess the strength of built-in safety mechanisms. Our evaluation addresses three key questions: (i) how the introduction of reasoning capabilities affects model fairness and robustness; (ii) whether models fine-tuned for reasoning exhibit greater safety than those relying on CoT prompting at inference time; and (iii) how the success rate of jailbreak attacks targeting bias elicitation varies with the reasoning mechanisms employed. Our findings reveal a nuanced relationship between reasoning capabilities and bias safety. Surprisingly, models with explicit reasoning, whether via CoT prompting or fine-tuned reasoning traces, are generally more vulnerable to bias elicitation than base models without such mechanisms, suggesting reasoning may unintentionally open new pathways for stereotype reinforcement. Reasoning-enabled models appear somewhat safer than those relying on CoT prompting, which are particularly prone to contextual reframing attacks through storytelling prompts, fictional personas, or reward-shaped instructions. These results challenge the assumption that reasoning inherently improves robustness and underscore the need for more bias-aware approaches to reasoning design.

Recent work has shown that training loss scales as a power law with both model size and the number of tokens, and that achieving compute-optimal models requires scaling model size and token count together. However, these scaling laws assume an infinite supply of data and apply primarily in compute-bound settings. As modern large language models increasingly rely on massive internet-scale datasets, the assumption that they are compute-bound is becoming less valid. This shift highlights the need for architectures that prioritize token efficiency. In this work, we investigate the use of the 2-simplicial Transformer, an architecture that generalizes standard dot-product attention to trilinear functions through an efficient Triton kernel implementation. We demonstrate that the 2-simplicial Transformer achieves better token efficiency than standard Transformers: for a fixed token budget, similarly sized models outperform their dot-product counterparts on tasks involving mathematics, coding, reasoning, and logic. We quantify these gains by demonstrating that $2$-simplicial attention changes the exponent in the scaling laws for knowledge and reasoning tasks compared to dot product attention.