Data privacy remains a critical concern in educational research, necessitating Institutional Review Board (IRB) certification and stringent data handling protocols to ensure compliance with ethical standards. Traditional approaches rely on anonymization and controlled data-sharing mechanisms to facilitate research while mitigating privacy risks. However, these methods still involve direct access to raw student data, posing potential vulnerabilities and being time-consuming. This study proposes a federated learning (FL) framework for automatic scoring in educational assessments, eliminating the need to share raw data. Our approach leverages client-side model training, where student responses are processed locally on edge devices, and only optimized model parameters are shared with a central aggregation server. To effectively aggregate heterogeneous model updates, we introduce an adaptive weighted averaging strategy, which dynamically adjusts weight contributions based on client-specific learning characteristics. This method ensures robust model convergence while preserving privacy. We evaluate our framework using assessment data from nine middle schools, comparing the accuracy of federated learning-based scoring models with traditionally trained centralized models. A statistical significance test (paired t-test, $t(8) = 2.29, p = 0.051$) confirms that the accuracy difference between the two approaches is not statistically significant, demonstrating that federated learning achieves comparable performance while safeguarding student data. Furthermore, our method significantly reduces data collection, processing, and deployment overhead, accelerating the adoption of AI-driven educational assessments in a privacy-compliant manner.

Large Language Models (LLMs) have demonstrated unprecedented generative capabilities, yet their alignment with human values remains critical for ensuring helpful and harmless deployments. While Reinforcement Learning from Human Feedback (RLHF) has emerged as a powerful paradigm for aligning LLMs with human preferences, its reliance on complex reward modeling introduces inherent trade-offs in computational efficiency and training stability. In this context, Direct Preference Optimization (DPO) has recently gained prominence as a streamlined alternative that directly optimizes LLMs using human preferences, thereby circumventing the need for explicit reward modeling. Owing to its theoretical elegance and computational efficiency, DPO has rapidly attracted substantial research efforts exploring its various implementations and applications. However, this field currently lacks systematic organization and comparative analysis. In this survey, we conduct a comprehensive overview of DPO and introduce a novel taxonomy, categorizing previous works into four key dimensions: data strategy, learning framework, constraint mechanism, and model property. We further present a rigorous empirical analysis of DPO variants across standardized benchmarks. Additionally, we discuss real-world applications, open challenges, and future directions for DPO. This work delivers both a conceptual framework for understanding DPO and practical guidance for practitioners, aiming to advance robust and generalizable alignment paradigms. All collected resources are available and will be continuously updated at https://github.com/liushunyu/awesome-direct-preference-optimization.

We introduce Test-Time Discovery (TTD) as a novel task that addresses class shifts during testing, requiring models to simultaneously identify emerging categories while preserving previously learned ones. A key challenge in TTD is distinguishing newly discovered classes from those already identified. To address this, we propose a training-free, hash-based memory mechanism that enhances class discovery through fine-grained comparisons with past test samples. Leveraging the characteristics of unknown classes, our approach introduces hash representation based on feature scale and directions, utilizing Locality-Sensitive Hashing (LSH) for efficient grouping of similar samples. This enables test samples to be easily and quickly compared with relevant past instances. Furthermore, we design a collaborative classification strategy, combining a prototype classifier for known classes with an LSH-based classifier for novel ones. To enhance reliability, we incorporate a self-correction mechanism that refines memory labels through hash-based neighbor retrieval, ensuring more stable and accurate class assignments. Experimental results demonstrate that our method achieves good discovery of novel categories while maintaining performance on known classes, establishing a new paradigm in model testing. Our code is available at https://github.com/fanlyu/ttd.

Medical large language models (LLMs) research often makes bold claims, from encoding clinical knowledge to reasoning like a physician. These claims are usually backed by evaluation on competitive benchmarks; a tradition inherited from mainstream machine learning. But how do we separate real progress from a leaderboard flex? Medical LLM benchmarks, much like those in other fields, are arbitrarily constructed using medical licensing exam questions. For these benchmarks to truly measure progress, they must accurately capture the real-world tasks they aim to represent. In this position paper, we argue that medical LLM benchmarks should (and indeed can) be empirically evaluated for their construct validity. In the psychological testing literature, "construct validity" refers to the ability of a test to measure an underlying "construct", that is the actual conceptual target of evaluation. By drawing an analogy between LLM benchmarks and psychological tests, we explain how frameworks from this field can provide empirical foundations for validating benchmarks. To put these ideas into practice, we use real-world clinical data in proof-of-concept experiments to evaluate popular medical LLM benchmarks and report significant gaps in their construct validity. Finally, we outline a vision for a new ecosystem of medical LLM evaluation centered around the creation of valid benchmarks.

In this paper, we address the problem of human trajectory forecasting, which aims to predict the inherently multi-modal future movements of humans based on their past trajectories and other contextual cues. We propose a novel motion prediction conditional flow matching model, termed MoFlow, to predict K-shot future trajectories for all agents in a given scene. We design a novel flow matching loss function that not only ensures at least one of the $K$ sets of future trajectories is accurate but also encourages all $K$ sets of future trajectories to be diverse and plausible. Furthermore, by leveraging the implicit maximum likelihood estimation (IMLE), we propose a novel distillation method for flow models that only requires samples from the teacher model. Extensive experiments on the real-world datasets, including SportVU NBA games, ETH-UCY, and SDD, demonstrate that both our teacher flow model and the IMLE-distilled student model achieve state-of-the-art performance. These models can generate diverse trajectories that are physically and socially plausible. Moreover, our one-step student model is $\textbf{100}$ times faster than the teacher flow model during sampling. The code, model, and data are available at our project page: https://moflow-imle.github.io

Vision-and-language navigation (VLN) tasks require agents to navigate three-dimensional environments guided by natural language instructions, offering substantial potential for diverse applications. However, the scarcity of training data impedes progress in this field. This paper introduces PanoGen++, a novel framework that addresses this limitation by generating varied and pertinent panoramic environments for VLN tasks. PanoGen++ incorporates pre-trained diffusion models with domain-specific fine-tuning, employing parameter-efficient techniques such as low-rank adaptation to minimize computational costs. We investigate two settings for environment generation: masked image inpainting and recursive image outpainting. The former maximizes novel environment creation by inpainting masked regions based on textual descriptions, while the latter facilitates agents' learning of spatial relationships within panoramas. Empirical evaluations on room-to-room (R2R), room-for-room (R4R), and cooperative vision-and-dialog navigation (CVDN) datasets reveal significant performance enhancements: a 2.44% increase in success rate on the R2R test leaderboard, a 0.63% improvement on the R4R validation unseen set, and a 0.75-meter enhancement in goal progress on the CVDN validation unseen set. PanoGen++ augments the diversity and relevance of training environments, resulting in improved generalization and efficacy in VLN tasks.

Generative AI (GAI) technologies are disrupting professional writing, challenging traditional practices. Recent studies explore GAI adoption experiences of creative practitioners, but we know little about how these experiences evolve into established practices and how GAI resistance alters these practices. To address this gap, we conducted 25 semi-structured interviews with writing professionals who adopted and/or resisted GAI. Using the theoretical lens of Job Crafting, we identify four strategies professionals employ to reshape their roles. Writing professionals employed GAI resisting strategies to maximize human potential, reinforce professional identity, carve out a professional niche, and preserve credibility within their networks. In contrast, GAI-enabled strategies allowed writers who embraced GAI to enhance desirable workflows, minimize mundane tasks, and engage in new AI-managerial labor. These strategies amplified their collaborations with GAI while reducing their reliance on other people. We conclude by discussing implications of GAI practices on writers' identity and practices as well as crafting theory.

Queer people are often discussed as targets of bias, harm, or discrimination in research on generative AI. However, the specific ways that queer people engage with generative AI, and thus possible uses that support queer people, have yet to be explored. We conducted a workshop study with 13 queer artists, during which we gave participants access to GPT-4 and DALL-E 3 and facilitated group sensemaking activities. We found our participants struggled to use these models due to various normative values embedded in their designs, such as hyper-positivity and anti-sexuality. We describe various strategies our participants developed to overcome these models' limitations and how, nevertheless, our participants found value in these highly-normative technologies. Drawing on queer feminist theory, we discuss implications for the conceptualization of "state-of-the-art" models and consider how FAccT researchers might support queer alternatives.

Automatic scoring of student responses enhances efficiency in education, but deploying a separate neural network for each task increases storage demands, maintenance efforts, and redundant computations. To address these challenges, this paper introduces the Gromov-Wasserstein Scoring Model Merging (GW-SMM) method, which merges models based on feature distribution similarities measured via the Gromov-Wasserstein distance. Our approach begins by extracting features from student responses using individual models, capturing both item-specific context and unique learned representations. The Gromov-Wasserstein distance then quantifies the similarity between these feature distributions, identifying the most compatible models for merging. Models exhibiting the smallest pairwise distances, typically in pairs or trios, are merged by combining only the shared layers preceding the classification head. This strategy results in a unified feature extractor while preserving separate classification heads for item-specific scoring. We validated our approach against human expert knowledge and a GPT-o1-based merging method. GW-SMM consistently outperformed both, achieving a higher micro F1 score, macro F1 score, exact match accuracy, and per-label accuracy. The improvements in micro F1 and per-label accuracy were statistically significant compared to GPT-o1-based merging (p=0.04, p=0.01). Additionally, GW-SMM reduced storage requirements by half without compromising much accuracy, demonstrating its computational efficiency alongside reliable scoring performance.

Supervised learning relies on annotated data, which is expensive to obtain. A longstanding strategy to reduce annotation costs is active learning, an iterative process, in which a human annotates only data instances deemed informative by a model. Large language models (LLMs) have pushed the effectiveness of active learning, but have also improved methods such as few- or zero-shot learning, and text synthesis - thereby introducing potential alternatives. This raises the question: has active learning become obsolete? To answer this fully, we must look beyond literature to practical experiences. We conduct an online survey in the NLP community to collect previously intangible insights on the perceived relevance of data annotation, particularly focusing on active learning, including best practices, obstacles and expected future developments. Our findings show that annotated data remains a key factor, and active learning continues to be relevant. While the majority of active learning users find it effective, a comparison with a community survey from over a decade ago reveals persistent challenges: setup complexity, estimation of cost reduction, and tooling. We publish an anonymized version of the collected dataset