Model compression is crucial for minimizing memory storage and accelerating inference in deep learning (DL) models, including recent foundation models like large language models (LLMs). Users can access different compressed model versions according to their resources and budget. However, while existing compression operations primarily focus on optimizing the trade-off between resource efficiency and model performance, the privacy risks introduced by compression remain overlooked and insufficiently understood. In this work, through the lens of membership inference attack (MIA), we propose CompLeak, the first privacy risk evaluation framework examining three widely used compression configurations that are pruning, quantization, and weight clustering supported by the commercial model compression framework of Google's TensorFlow-Lite (TF-Lite) and Facebook's PyTorch Mobile. CompLeak has three variants, given available access to the number of compressed models and original model. CompLeakNR starts by adopting existing MIA methods to attack a single compressed model, and identifies that different compressed models influence members and non-members differently. When the original model and one compressed model are available, CompLeakSR leverages the compressed model as a reference to the original model and uncovers more privacy by combining meta information (e.g., confidence vector) from both models. When multiple compressed models are available with/without accessing the original model, CompLeakMR innovatively exploits privacy leakage info from multiple compressed versions to substantially signify the overall privacy leakage. We conduct extensive experiments on seven diverse model architectures (from ResNet to foundation models of BERT and GPT-2), and six image and textual benchmark datasets.

Multi-step reasoning is a fundamental challenge in artificial intelligence, with applications ranging from mathematical problem-solving to decision-making in dynamic environments. Reinforcement Learning (RL) has shown promise in enabling agents to perform multi-step reasoning by optimizing long-term rewards. However, conventional RL methods struggle with complex reasoning tasks due to issues such as credit assignment, high-dimensional state representations, and stability concerns. Recent advancements in Transformer architectures and hyperbolic geometry have provided novel solutions to these challenges. This paper introduces a new framework that integrates hyperbolic Transformers into RL for multi-step reasoning. The proposed approach leverages hyperbolic embeddings to model hierarchical structures effectively. We present theoretical insights, algorithmic details, and experimental results that include Frontier Math and nonlinear optimal control problems. Compared to RL with vanilla transformer, the hyperbolic RL largely improves accuracy by (32%~44%) on FrontierMath benchmark, (43%~45%) on nonlinear optimal control benchmark, while achieving impressive reduction in computational time by (16%~32%) on FrontierMath benchmark, (16%~17%) on nonlinear optimal control benchmark. Our work demonstrates the potential of hyperbolic Transformers in reinforcement learning, particularly for multi-step reasoning tasks that involve hierarchical structures.

Achieving human-like perception and reasoning in Multimodal Large Language Models (MLLMs) remains a central challenge in artificial intelligence. While recent research has primarily focused on enhancing reasoning capabilities in MLLMs, a fundamental question persists: Can Multimodal Large Language Models truly perceive the world as humans do? This paper shifts focus from reasoning to perception. Rather than constructing benchmarks specifically for reasoning, we introduce the Turing Eye Test (TET), a challenging perception-oriented benchmark comprising four diagnostic tasks that evaluate MLLMs' performance on synthetic images that humans process intuitively. Our findings reveal that state-of-the-art MLLMs exhibit catastrophic failures on our perceptual tasks trivial for humans. Both in-context learning and training on language backbone-effective for previous benchmarks-fail to improve performance on our tasks, while fine-tuning the vision tower enables rapid adaptation, suggesting that our benchmark poses challenges for vision tower generalization rather than for the knowledge and reasoning capabilities of the language backbone-a key gap between current MLLMs and human perception. We release a representative subset of TET tasks in this version, and will introduce more diverse tasks and methods to enhance visual generalization in future work.

Recent advances in Multimodal Large Language Models (MLLMs) have enabled the development of mobile agents that can understand visual inputs and follow user instructions, unlocking new possibilities for automating complex tasks on mobile devices. However, applying these models to real-world mobile scenarios remains a significant challenge due to the long-horizon task execution, difficulty in error recovery, and the cold-start problem in unfamiliar environments. To address these challenges, we propose MobileUse, a GUI agent designed for robust and adaptive mobile task execution. To improve resilience in long-horizon tasks and dynamic environments, we introduce a hierarchical reflection architecture that enables the agent to self-monitor, detect, and recover from errors across multiple temporal scales-ranging from individual actions to overall task completion-while maintaining efficiency through a reflection-on-demand strategy. To tackle cold-start issues, we further introduce a proactive exploration module, which enriches the agent's understanding of the environment through self-planned exploration. Evaluations on AndroidWorld and AndroidLab benchmarks demonstrate that MobileUse establishes new state-of-the-art performance, achieving success rates of 62.9% and 44.2%, respectively. To facilitate real-world applications, we release an out-of-the-box toolkit for automated task execution on physical mobile devices, which is available at https://github.com/MadeAgents/mobile-use.

Inspection of aquaculture net pens is essential for maintaining the structural integrity, biosecurity, and operational efficiency of fish farming systems. Traditional inspection approaches rely on pre-programmed missions or manual control, offering limited adaptability to dynamic underwater conditions and user-specific demands. In this study, we propose AquaChat, a novel Remotely Operated Vehicle (ROV) framework that integrates Large Language Models (LLMs) for intelligent and adaptive net pen inspection. The system features a multi-layered architecture: (1) a high-level planning layer that interprets natural language user commands using an LLM to generate symbolic task plans; (2) a mid-level task manager that translates plans into ROV control sequences; and (3) a low-level motion control layer that executes navigation and inspection tasks with precision. Real-time feedback and event-triggered replanning enhance robustness in challenging aquaculture environments. The framework is validated through experiments in both simulated and controlled aquatic environments representative of aquaculture net pens. Results demonstrate improved task flexibility, inspection accuracy, and operational efficiency. AquaChat illustrates the potential of integrating language-based AI with marine robotics to enable intelligent, user-interactive inspection systems for sustainable aquaculture operations.

The rapid advancement of large language models (LLMs) has enabled role-playing language agents to demonstrate significant potential in various applications. However, relying solely on prompts and contextual inputs often proves insufficient for achieving deep immersion in specific roles, particularly well-known fictional or public figures. On the other hand, fine-tuning-based approaches face limitations due to the challenges associated with data collection and the computational resources required for training, thereby restricting their broader applicability. To address these issues, we propose Test-Time-Matching (TTM), a training-free role-playing framework through test-time scaling and context engineering. TTM uses LLM agents to automatically decouple a character's features into personality, memory, and linguistic style. Our framework involves a structured, three-stage generation pipeline that utilizes these features for controlled role-playing. It achieves high-fidelity role-playing performance, also enables seamless combinations across diverse linguistic styles and even variations in personality and memory. We evaluate our framework through human assessment, and the results demonstrate that our method achieves the outstanding performance in generating expressive and stylistically consistent character dialogues.

Hausa Natural Language Processing (NLP) has gained increasing attention in recent years, yet remains understudied as a low-resource language despite having over 120 million first-language (L1) and 80 million second-language (L2) speakers worldwide. While significant advances have been made in high-resource languages, Hausa NLP faces persistent challenges, including limited open-source datasets and inadequate model representation. This paper presents an overview of the current state of Hausa NLP, systematically examining existing resources, research contributions, and gaps across fundamental NLP tasks: text classification, machine translation, named entity recognition, speech recognition, and question answering. We introduce HausaNLP (https://catalog.hausanlp.org), a curated catalog that aggregates datasets, tools, and research works to enhance accessibility and drive further development. Furthermore, we discuss challenges in integrating Hausa into large language models (LLMs), addressing issues of suboptimal tokenization and dialectal variation. Finally, we propose strategic research directions emphasizing dataset expansion, improved language modeling approaches, and strengthened community collaboration to advance Hausa NLP. Our work provides both a foundation for accelerating Hausa NLP progress and valuable insights for broader multilingual NLP research.

Language models for program synthesis are usually trained and evaluated on programming competition datasets (MBPP, APPS). However, these datasets are limited in size and quality, while these language models are extremely data hungry. Additionally, the language models have a misaligned program synthesis process compared to humans. While humans iteratively develop code with the help of a compiler, most program synthesis models currently produce code in one go. To solve these issues, we introduce a bootstrapping algorithm for program synthesis, that supports teaching models how to repair. We show that bootstrapping consistently outperforms regular fine-tuning. Compared to other work, our bootstrapped model performs on par with fine-tuned models that are 68\% larger. Notably, bootstrapping with repairing also improves non-repairing performance compared to regular bootstrapping during inference. However, on our models, repairing during inference is likely inferior to simply sampling the same number of solutions. Furthermore, we find that there are issues with the example test cases in the training portion of the APPS dataset that are valuable to the community, as many repairing and reinforcement learning methods rely on them.

The recent rise in youth suicide highlights the urgent need to understand how online experiences contribute to this public health issue. Our mixed-methods approach responds to this challenge by developing a set of themes focused on risk factors for suicide mortality in online spaces among youth ages 10-24, and a framework to model these themes at scale. Using 29,124 open text summaries of death investigations between 2013-2022, we conducted a thematic analysis to identify 12 types of online activities that were considered by investigators or next of kin to be relevant in contextualizing a given suicide death. We then develop a zero-shot learning framework to model these 12 themes at scale, and analyze variation in these themes by decedent characteristics and over time. Our work uncovers several online activities related to harm to self, harm to others, interpersonal interactions, activity levels online, and life events, which correspond to different phases of suicide risk from two prominent suicide theories. We find an association between these themes and decedent characteristics like age, means of death, and interpersonal problems, and many themes became more prevalent during the 2020 COVID-19 lockdowns. While digital spaces have taken some steps to address expressions of suicidality online, our work illustrates the opportunities for developing interventions related to less explicit indicators of suicide risk by combining suicide theories with computational research.

Large Language Models (LLMs) have shown promise in assisting developers with code-related questions; however, LLMs carry the risk of generating unreliable answers. To address this, Retrieval-Augmented Generation (RAG) has been proposed to reduce the unreliability (i.e., hallucinations) of LLMs. However, designing effective pipelines remains challenging due to numerous design choices. In this paper, we construct a retrieval corpus of over 3 million Java and Python related Stack Overflow posts with accepted answers, and explore various RAG pipeline designs to answer developer questions, evaluating their effectiveness in generating accurate and reliable responses. More specifically, we (1) design and evaluate 7 different RAG pipelines and 63 pipeline variants to answer questions that have historically similar matches, and (2) address new questions without any close prior matches by automatically lowering the similarity threshold during retrieval, thereby increasing the chance of finding partially relevant context and improving coverage for unseen cases. We find that implementing a RAG pipeline combining hypothetical-documentation-embedding (HyDE) with the full-answer context performs best in retrieving and answering similarcontent for Stack Overflow questions. Finally, we apply our optimal RAG pipeline to 4 open-source LLMs and compare the results to their zero-shot performance. Our findings show that RAG with our optimal RAG pipeline consistently outperforms zero-shot baselines across models, achieving higher scores for helpfulness, correctness, and detail with LLM-as-a-judge. These findings demonstrate that our optimal RAG pipelines robustly enhance answer quality for a wide range of developer queries including both previously seen and novel questions across different LLMs