Ontology-based question generation is an important application of semantic-aware systems that enables the creation of large question banks for diverse learning environments. The effectiveness of these systems, both in terms of the calibre and cognitive difficulty of the resulting questions, depends heavily on the quality and modelling approach of the underlying ontologies, making it crucial to assess their fitness for this task. To date, there has been no comprehensive investigation into the specific ontology aspects or characteristics that affect the question generation process. Therefore, this paper proposes a set of requirements and task-specific metrics for evaluating the fitness of ontologies for question generation tasks in pedagogical settings. Using the ROMEO methodology (a structured framework used for identifying task-specific metrics), a set of evaluation metrics have been derived from an expert assessment of questions generated by a question generation model. To validate the proposed metrics, we apply them to a set of ontologies previously used in question generation to illustrate how the metric scores align with and complement findings reported in earlier studies. The analysis confirms that ontology characteristics significantly impact the effectiveness of question generation, with different ontologies exhibiting varying performance levels. This highlights the importance of assessing ontology quality with respect to Automatic Question Generation (AQG) tasks.

Generative AI is transforming computing education by enabling the automatic generation of personalized content and feedback. We investigate its capabilities in providing high-quality programming tasks to students. Despite promising advancements in task generation, a quality gap remains between AI-generated and expert-created tasks. The AI-generated tasks may not align with target programming concepts, could be incomprehensible to students, or may contain critical issues such as incorrect tests. Existing works often require interventions from human teachers for validation. We address these challenges by introducing PyTaskSyn, a novel synthesis technique that first generates a programming task and then decides whether it meets certain quality criteria to be given to students. The key idea is to break this process into multiple stages performed by expert and student agents simulated using both strong and weaker generative models. Through extensive evaluation, we show that PyTaskSyn significantly improves task quality compared to baseline techniques and showcases the importance of each specialized agent type in our validation pipeline. Additionally, we conducted user studies using our publicly available web application and show that PyTaskSyn can deliver high-quality programming tasks comparable to expert-designed ones while reducing workload and costs, and being more engaging than programming tasks that are available in online resources.

The nodes' interconnections on a social network often reflect their dependencies and information-sharing behaviors. Nevertheless, abnormal nodes, which significantly deviate from most of the network concerning patterns or behaviors, can lead to grave consequences. Therefore, it is imperative to design efficient online learning algorithms that robustly learn users' preferences while simultaneously detecting anomalies. We introduce a novel bandit algorithm to address this problem. Through network knowledge, the method characterizes the users' preferences and residuals of feature information. By learning and analyzing these preferences and residuals, it develops a personalized recommendation strategy for each user and simultaneously detects anomalies. We rigorously prove an upper bound on the regret of the proposed algorithm and experimentally compare it with several state-of-the-art collaborative contextual bandit algorithms on both synthetic and real-world datasets.

For human cognitive process, spatial reasoning and perception are closely entangled, yet the nature of this interplay remains underexplored in the evaluation of multimodal large language models (MLLMs). While recent MLLM advancements show impressive performance on reasoning, their capacity for human-like spatial cognition remains an open question. In this work, we introduce a systematic evaluation framework to assess the spatial reasoning abilities of state-of-the-art MLLMs relative to human performance. Central to our work is 11Plus-Bench, a high-quality benchmark derived from realistic standardized spatial aptitude tests. 11Plus-Bench also features fine-grained expert annotations of both perceptual complexity and reasoning process, enabling detailed instance-level analysis of model behavior. Through extensive experiments across 14 MLLMs and human evaluation, we find that current MLLMs exhibit early signs of spatial cognition. Despite a large performance gap compared to humans, MLLMs' cognitive profiles resemble those of humans in that cognitive effort correlates strongly with reasoning-related complexity. However, instance-level performance in MLLMs remains largely random, whereas human correctness is highly predictable and shaped by abstract pattern complexity. These findings highlight both emerging capabilities and limitations in current MLLMs' spatial reasoning capabilities and provide actionable insights for advancing model design.

The rapid advancement of large vision language models (L VLMs) and agent systems has heightened interest in mobile GUI agents that can reliably translate natural language into interface operations. Existing single-agent approaches, however, remain limited by structural constraints. Although multi-agent systems naturally decouple different competencies, recent progress in multi-agent reinforcement learning (MARL) has often been hindered by inefficiency and remains incompatible with current L VLM architectures. To address these challenges, we introduce SWIRL, a staged workflow for interleaved reinforcement learning designed for multi-agent systems. SWIRL reformulates MARL into a sequence of single-agent reinforcement learning tasks, updating one agent at a time while keeping the others fixed. This formulation enables stable training and promotes efficient coordination across agents. Theoretically, we provide a stepwise safety bound, a cross-round monotonic improvement theorem, and convergence guarantees on return, ensuring robust and principled optimization. In application to mobile GUI control, SWIRL instantiates a Navigator that converts language and screen context into structured plans, and an Interactor that grounds these plans into executable atomic actions. Extensive experiments demonstrate superior performance on both high-level and low-level GUI benchmarks. Beyond GUI tasks, SWIRL also demonstrates strong capability in multi-agent mathematical reasoning, underscoring its potential as a general framework for developing efficient and robust multi-agent systems.

Fine-tuning LLMs on narrowly harmful datasets can lead to behavior that is broadly misaligned with respect to human values. To understand when and how this emergent misalignment occurs, we develop a comprehensive framework for detecting and characterizing rapid transitions during fine-tuning using both distributional change detection methods as well as order parameters that are formulated in plain English and evaluated by an LLM judge. Using an objective statistical dissimilarity measure, we quantify how the phase transition that occurs during fine-tuning affects multiple aspects of the model. In particular, we assess what percentage of the total distributional change in model outputs is captured by different aspects, such as alignment or verbosity, providing a decomposition of the overall transition. We also find that the actual behavioral transition occurs later in training than indicated by the peak in the gradient norm alone. Our framework enables the automated discovery and quantification of language-based order parameters, which we demonstrate on examples ranging from knowledge questions to politics and ethics.

Large Language Models (LLMs) have achieved high accuracy on complex commonsense and mathematical problems that involve the composition of multiple reasoning steps. However, current compositional benchmarks testing these skills tend to focus on either commonsense or math reasoning, whereas LLM agents solving real-world tasks would require a combination of both. In this work, we introduce an Agentic Commonsense and Math benchmark (AgentCoMa), where each compositional task requires a commonsense reasoning step and a math reasoning step. We test it on 61 LLMs of different sizes, model families, and training strategies. We find that LLMs can usually solve both steps in isolation, yet their accuracy drops by ~30% on average when the two are combined. This is a substantially greater performance gap than the one we observe in prior compositional benchmarks that combine multiple steps of the same reasoning type. In contrast, non-expert human annotators can solve the compositional questions and the individual steps in AgentCoMa with similarly high accuracy. Furthermore, we conduct a series of interpretability studies to better understand the performance gap, examining neuron patterns, attention maps and membership inference. Our work underscores a substantial degree of model brittleness in the context of mixed-type compositional reasoning and offers a test bed for future improvement.

The use of inappropriate language--such as outdated, exclu-sionary, or non-patient-centered terms--in medical instructional materials can significantly influence clinical training, patient interactions, and health outcomes. Despite their reputability, many materials developed over past decades contain examples now considered inappropriate by current medical standards. Given the volume of curricular content, manually identifying instances of inappropriate use of language (IUL) and its subcategories for systematic review is prohibitively costly and impractical. To address this challenge, we conduct a first-in-class evaluation of small language models (SLMs) fine-tuned on labeled data and pre-trained LLMs with in-context learning on a dataset containing approximately 500 documents and over 12,000 pages. For SLMs, we consider: (1) a general IUL classifier, (2) subcategory-specific binary classifiers, (3) a multilabel classifier, and (4) a two-stage hierarchical pipeline for general IUL detection followed by mul-tilabel classification. For LLMs, we consider variations of prompts that include subcategory definitions and/or shots. We found that both LLama-3 8B and 70B, even with carefully curated shots, are largely outperformed by SLMs. While the multilabel classifier performs best on annotated data, supplementing training with unflagged excerpts as negative examples boosts the specific classifiers' AUC by up to 25%, making them most effective models for mitigating harmful language in medical curricula.

Qualitative analysis of open-ended survey responses is a commonly-used research method in the social sciences, but traditional coding approaches are often time-consuming and prone to inconsistency. Existing solutions from Natural Language Processing such as supervised classifiers, topic modeling techniques, and generative large language models have limited applicability in qualitative analysis, since they demand extensive labeled data, disrupt established qualitative workflows, and/or yield variable results. In this paper, we introduce a text embedding-based classification framework that requires only a handful of examples per category and fits well with standard qualitative workflows. When benchmarked against human analysis of a conceptual physics survey consisting of 2899 open-ended responses, our framework achieves a Cohen's Kappa ranging from 0.74 to 0.83 as compared to expert human coders in an exhaustive coding scheme. We further show how performance of this framework improves with fine-tuning of the text embedding model, and how the method can be used to audit previously-analyzed datasets. These findings demonstrate that text embedding-assisted coding can flexibly scale to thousands of responses without sacrificing interpretability, opening avenues for deductive qualitative analysis at scale.

This thesis examines the obstacles and potential solutions for creating such bots, focusing on innovative techniques that enable bots to learn and adapt autonomously in constantly changing environments. End-to-end task bots, typically built using a static and limited corpus, face difficulties when deployed online due to three primary factors tied to this limitation. First, they might confront queries featuring unexpected linguistic patterns or slot values (i.e., unseen user behaviors). Second, they could potentially face requirements for new functions or tasks (i.e., task definition extensions). Third, even when equipped with relevant knowledge, these bots may produce responses that appear plausible but are actually incorrect (i.e., "hallucinations"). Addressing these challenges is vital for enhancing task bots' performance and reliability in real-world settings.