Model merging has achieved significant success, with numerous innovative methods proposed to enhance capabilities by combining multiple models. However, challenges persist due to the lack of a unified framework for classification and systematic comparative analysis, leading to inconsistencies in terminologies and categorizations. Meanwhile, as an increasing number of fine-tuned models are publicly available, their original training data often remain inaccessible due to privacy concerns or intellectual property restrictions. This makes traditional multi-task learning based on shared training data impractical. In scenarios where direct access to training data is infeasible, merging model parameters to create a unified model with broad generalization across multiple domains becomes crucial, further underscoring the importance of model merging techniques. Despite the rapid progress in this field, a comprehensive taxonomy and survey summarizing recent advances and predicting future directions are still lacking. This paper addresses these gaps by establishing a new taxonomy of model merging methods, systematically comparing different approaches, and providing an overview of key developments. By offering a structured perspective on this evolving area, we aim to help newcomers quickly grasp the field's landscape and inspire further innovations.

Reinforcement learning from human feedback (RLHF) has become essential for improving language model capabilities, but traditional approaches rely on the assumption that human preferences follow a transitive Bradley-Terry model. This assumption fails to capture the non-transitive nature of populational human preferences. Nash learning from human feedback (NLHF), targeting non-transitive preferences, is a problem of computing the Nash equilibrium (NE) of the two-player constant-sum game defined by the human preference. We introduce Extragradient preference optimization (EGPO), a novel algorithm for NLHF achieving last-iterate linear convergence to the NE of KL-regularized games and polynomial convergence to the NE of original games, while being robust to noise. Unlike previous approaches that rely on nested optimization, we derive an equivalent implementation using gradients of an online variant of the identity preference optimization (IPO) loss, enabling more faithful implementation for neural networks. Our empirical evaluations demonstrate EGPO's superior performance over baseline methods when training for the same number of epochs, as measured by pairwise win-rates using the ground truth preference.

Source-free domain adaptation (SFDA) has been exploited for cross-domain bearing fault diagnosis without access to source data. Current methods select partial target samples with reliable pseudo-labels for model adaptation, which is sub-optimal due to the ignored target samples. We argue that every target sample can contribute to model adaptation, and accordingly propose in this paper a novel SFDA-based approach for bearing fault diagnosis that exploits both reliable and unreliable pseudo-labels. We develop a data-augmentation-based label voting strategy to divide the target samples into reliable and unreliable ones. We propose to explore the underlying relation between feature space and label space by using the reliable pseudo-labels as ground-truth labels, meanwhile, alleviating negative transfer by maximizing the entropy of the unreliable pseudo-labels. The proposed method achieves well-balance between discriminability and diversity by taking advantage of reliable and unreliable pseudo-labels. Extensive experiments are conducted on two bearing fault benchmarks, demonstrating that our approach achieves significant performance improvements against existing SFDA-based bearing fault diagnosis methods. Our code is available at https://github.com/BdLab405/SDALR.

In today's increasingly digital interactions, robust Identity Verification (IDV) is crucial for security and trust. Artificial Intelligence (AI) is transforming IDV, enhancing accuracy and fraud detection. This paper introduces ``Zero to One,'' a holistic conceptual framework for developing AI-powered IDV products. This paper outlines the foundational problem and research objectives that necessitate a new framework for IDV in the age of AI. It details the evolution of identity verification and the current regulatory landscape to contextualize the need for a robust conceptual model. The core of the paper is the presentation of the ``Zero to One'' framework itself, dissecting its four essential components: Document Verification, Biometric Verification, Risk Assessment, and Orchestration. The paper concludes by discussing the implications of this conceptual model and suggesting future research directions focused on the framework's further development and application. The framework addresses security, privacy, UX, and regulatory compliance, offering a structured approach to building effective IDV solutions. Successful IDV platforms require a balanced conceptual understanding of verification methods, risk management, and operational scalability, with AI as a key enabler. This paper presents the ``Zero to One'' framework as a refined conceptual model, detailing verification layers, and AI's transformative role in shaping next-generation IDV products.

Large Language Models (LLMs) are prone to hallucinations, e.g., factually incorrect information, in their responses. These hallucinations present challenges for LLM-based applications that demand high factual accuracy. Existing hallucination detection methods primarily depend on external resources, which can suffer from issues such as low availability, incomplete coverage, privacy concerns, high latency, low reliability, and poor scalability. There are also methods depending on output probabilities, which are often inaccessible for closed-source LLMs like GPT models. This paper presents MetaQA, a self-contained hallucination detection approach that leverages metamorphic relation and prompt mutation. Unlike existing methods, MetaQA operates without any external resources and is compatible with both open-source and closed-source LLMs. MetaQA is based on the hypothesis that if an LLM's response is a hallucination, the designed metamorphic relations will be violated. We compare MetaQA with the state-of-the-art zero-resource hallucination detection method, SelfCheckGPT, across multiple datasets, and on two open-source and two closed-source LLMs. Our results reveal that MetaQA outperforms SelfCheckGPT in terms of precision, recall, and f1 score. For the four LLMs we study, MetaQA outperforms SelfCheckGPT with a superiority margin ranging from 0.041 - 0.113 (for precision), 0.143 - 0.430 (for recall), and 0.154 - 0.368 (for F1-score). For instance, with Mistral-7B, MetaQA achieves an average F1-score of 0.435, compared to SelfCheckGPT's F1-score of 0.205, representing an improvement rate of 112.2%. MetaQA also demonstrates superiority across all different categories of questions.

Until recently, robotics safety research was predominantly about collision avoidance and hazard reduction in the immediate vicinity of a robot. Since the advent of large vision and language models (VLMs), robots are now also capable of higher-level semantic scene understanding and natural language interactions with humans. Despite their known vulnerabilities (e.g. hallucinations or jail-breaking), VLMs are being handed control of robots capable of physical contact with the real world. This can lead to dangerous behaviors, making semantic safety for robots a matter of immediate concern. Our contributions in this paper are two fold: first, to address these emerging risks, we release the ASIMOV Benchmark, a large-scale and comprehensive collection of datasets for evaluating and improving semantic safety of foundation models serving as robot brains. Our data generation recipe is highly scalable: by leveraging text and image generation techniques, we generate undesirable situations from real-world visual scenes and human injury reports from hospitals. Secondly, we develop a framework to automatically generate robot constitutions from real-world data to steer a robot's behavior using Constitutional AI mechanisms. We propose a novel auto-amending process that is able to introduce nuances in written rules of behavior; this can lead to increased alignment with human preferences on behavior desirability and safety. We explore trade-offs between generality and specificity across a diverse set of constitutions of different lengths, and demonstrate that a robot is able to effectively reject unconstitutional actions. We measure a top alignment rate of 84.3% on the ASIMOV Benchmark using generated constitutions, outperforming no-constitution baselines and human-written constitutions. Data is available at asimov-benchmark.github.io

It has been roughly three years since the open-source release of Stable Diffusion ignited a Generative AI (GenAI) boom [Bengesi et al., 2023]. The proliferation of these technologies has since reshaped design practice and research. From early ideation to final implementation, these developments have significantly altered how design work is conceived, conducted, and evaluated [Hou et al., 2024]. This essay examines the critical juncture at which the design research community finds itself, seeking to understand and shape these developments while grappling with their implications for creative practice, design education, and professional identities. Popular discourse around GenAI often centers on simplified unequivocal narratives: AI as a threat to humanity, as a solution to global challenges, as a force of disruption, or as a replacement for humans [Gilardi et al., 2024]. While these narratives have sparked debate and interest, they can function as "semantic stopsigns"--conceptual framings that oversimplify complex issues, providing an illusion of resolution that hinders deeper inquiry [LessWrong Community, n.d., Lifton, 1961]. For instance, claims like "AI is unreliable" can lead to outright dismissal of its potential,

Evaluating Large Language Models (LLMs) free-form generated responses remains a challenge due to their diverse and open-ended nature. Traditional supervised signal-based automatic metrics fail to capture semantic equivalence or handle the variability of open-ended responses, while human evaluation, though reliable, is resource-intensive. Leveraging LLMs as evaluators offers a promising alternative due to their strong language understanding and instruction-following capabilities. Taking advantage of these capabilities, we propose the Dynamic Arbitration Framework for Evaluation (DAFE), which employs two primary LLM-as-judges and engages a third arbitrator only in cases of disagreements. This selective arbitration prioritizes evaluation reliability while reducing unnecessary computational demands compared to conventional majority voting. DAFE utilizes task-specific reference answers with dynamic arbitration to enhance judgment accuracy, resulting in significant improvements in evaluation metrics such as Macro F1 and Cohen's Kappa. Through experiments, including a comprehensive human evaluation, we demonstrate DAFE's ability to provide consistent, scalable, and resource-efficient assessments, establishing it as a robust framework for evaluating free-form model outputs.

This paper introduces JurisTCU, a Brazilian Portuguese dataset for legal information retrieval (LIR). The dataset is freely available and consists of 16,045 jurisprudential documents from the Brazilian Federal Court of Accounts, along with 150 queries annotated with relevance judgments. It addresses the scarcity of Portuguese-language LIR datasets with query relevance annotations. The queries are organized into three groups: real user keyword-based queries, synthetic keyword-based queries, and synthetic question-based queries. Relevance judgments were produced through a hybrid approach combining LLM-based scoring with expert domain validation. We used JurisTCU in 14 experiments using lexical search (document expansion methods) and semantic search (BERT-based and OpenAI embeddings). We show that the document expansion methods significantly improve the performance of standard BM25 search on this dataset, with improvements exceeding 45% in P@10, R@10, and nDCG@10 metrics when evaluating short keyword-based queries. Among the embedding models, the OpenAI models produced the best results, with improvements of approximately 70% in P@10, R@10, and nDCG@10 metrics for short keyword-based queries, suggesting that these dense embeddings capture semantic relationships in this domain, surpassing the reliance on lexical terms. Besides offering a dataset for the Portuguese-language IR research community, suitable for evaluating search systems, the results also contribute to enhancing a search system highly relevant to Brazilian citizens.

Long-form writing agents require flexible integration and interaction across information retrieval, reasoning, and composition. Current approaches rely on predetermined workflows and rigid thinking patterns to generate outlines before writing, resulting in constrained adaptability during writing. In this paper we propose a general agent framework that achieves human-like adaptive writing through recursive task decomposition and dynamic integration of three fundamental task types, i.e. retrieval, reasoning, and composition. Our methodology features: 1) a planning mechanism that interleaves recursive task decomposition and execution, eliminating artificial restrictions on writing workflow; and 2) integration of task types that facilitates heterogeneous task decomposition. Evaluations on both fiction writing and technical report generation show that our method consistently outperforms state-of-the-art approaches across all automatic evaluation metrics, which demonstrate the effectiveness and broad applicability of our proposed framework.