The rapid advancement of Large Language Models (LLMs) has spurred significant progress in Large Speech-Language Models (LSLMs), enhancing their capabilities in both speech understanding and generation. While existing LSLMs often concentrate on augmenting speech generation or tackling a diverse array of short-speech tasks, the efficient processing of long-form speech remains a critical yet underexplored challenge. This gap is primarily attributed to the scarcity of long-speech training datasets and the high computational costs associated with long sequences. To address these limitations, we introduce FastLongSpeech, a novel framework designed to extend LSLM capabilities for efficient long-speech processing without necessitating dedicated long-speech training data. FastLongSpeech incorporates an iterative fusion strategy that can compress excessively long-speech sequences into manageable lengths. To adapt LSLMs for long-speech inputs, it introduces a dynamic compression training approach, which exposes the model to short-speech sequences at varying compression ratios, thereby transferring the capabilities of LSLMs to long-speech tasks. To assess the long-speech capabilities of LSLMs, we develop a long-speech understanding benchmark called LongSpeech-Eval. Experiments show that our method exhibits strong performance in both long-speech and short-speech tasks, while greatly improving inference efficiency.

Red teaming has evolved from its origins in military applications to become a widely adopted methodology in cybersecurity and AI. In this paper, we take a critical look at the practice of AI red teaming. We argue that despite its current popularity in AI governance, there exists a significant gap between red teaming's original intent as a critical thinking exercise and its narrow focus on discovering model-level flaws in the context of generative AI. Current AI red teaming efforts focus predominantly on individual model vulnerabilities while overlooking the broader sociotechnical systems and emergent behaviors that arise from complex interactions between models, users, and environments. To address this deficiency, we propose a comprehensive framework operationalizing red teaming in AI systems at two levels: macro-level system red teaming spanning the entire AI development lifecycle, and micro-level model red teaming. Drawing on cybersecurity experience and systems theory, we further propose a set of six recommendations. In these, we emphasize that effective AI red teaming requires multifunctional teams that examine emergent risks, systemic vulnerabilities, and the interplay between technical and social factors.

Complex information needs in real-world search scenarios demand deep reasoning and knowledge synthesis across diverse sources, which traditional retrieval-augmented generation (RAG) pipelines struggle to address effectively. Current reasoning-based approaches suffer from a fundamental limitation: they use a single model to handle both high-level planning and detailed execution, leading to inefficient reasoning and limited scalability. In this paper, we introduce HiRA, a hierarchical framework that separates strategic planning from specialized execution. Our approach decomposes complex search tasks into focused subtasks, assigns each subtask to domain-specific agents equipped with external tools and reasoning capabilities, and coordinates the results through a structured integration mechanism. This separation prevents execution details from disrupting high-level reasoning while enabling the system to leverage specialized expertise for different types of information processing. Experiments on four complex, cross-modal deep search benchmarks demonstrate that HiRA significantly outperforms state-of-the-art RAG and agent-based systems. Our results show improvements in both answer quality and system efficiency, highlighting the effectiveness of decoupled planning and execution for multi-step information seeking tasks. Our code is available at https://github.com/ignorejjj/HiRA.

The "end-to-end" label for LLMs is a misnomer. In practice, they depend on a non-differentiable decoding process that requires laborious, hand-tuning of hyperparameters like temperature and top-p. This paper introduces AutoDeco, a novel architecture that enables truly "end-to-end" generation by learning to control its own decoding strategy. We augment the standard transformer with lightweight heads that, at each step, dynamically predict context-specific temperature and top-p values alongside the next-token logits. This approach transforms decoding into a parametric, token-level process, allowing the model to self-regulate its sampling strategy within a single forward pass. Through extensive experiments on eight benchmarks, we demonstrate that AutoDeco not only significantly outperforms default decoding strategies but also achieves performance comparable to an oracle-tuned baseline derived from "hacking the test set"-a practical upper bound for any static method. Crucially, we uncover an emergent capability for instruction-based decoding control: the model learns to interpret natural language commands (e.g., "generate with low randomness") and adjusts its predicted temperature and top-p on a token-by-token basis, opening a new paradigm for steerable and interactive LLM decoding.

We introduce Aeolus, a large-scale Multi-modal Flight Delay Dataset designed to advance research on flight delay prediction and support the development of foundation models for tabular data. Existing datasets in this domain are typically limited to flat tabular structures and fail to capture the spatiotemporal dynamics inherent in delay propagation. Aeolus addresses this limitation by providing three aligned modalities: (i) a tabular dataset with rich operational, meteorological, and airportlevel features for over 50 million flights; (ii) a flight chain module that models delay propagation along sequential flight legs, capturing upstream and downstream dependencies; and (iii) a flight network graph that encodes shared aircraft, crew, and airport resource connections, enabling cross-flight relational reasoning. The dataset is carefully constructed with temporal splits, comprehensive features, and strict leakage prevention to support realistic and reproducible machine learning evaluation. Aeolus supports a broad range of tasks, including regression, classification, temporal structure modeling, and graph learning, serving as a unified benchmark across tabular, sequential, and graph modalities. We release baseline experiments and preprocessing tools to facilitate adoption. Aeolus fills a key gap for both domain-specific modeling and general-purpose structured data research.Our source code and data can be accessed at https://github.com/Flnny/Delay-data

Normative reasoning is a type of reasoning that involves normative or deontic modality, such as obligation and permission. While large language models (LLMs) have demonstrated remarkable performance across various reasoning tasks, their ability to handle normative reasoning remains underexplored. In this paper, we systematically evaluate LLMs' reasoning capabilities in the normative domain from both logical and modal perspectives. Specifically, to assess how well LLMs reason with normative modals, we make a comparison between their reasoning with normative modals and their reasoning with epistemic modals, which share a common formal structure. To this end, we introduce a new dataset covering a wide range of formal patterns of reasoning in both normative and epistemic domains, while also incorporating non-formal cognitive factors that influence human reasoning. Our results indicate that, although LLMs generally adhere to valid reasoning patterns, they exhibit notable inconsistencies in specific types of normative reasoning and display cognitive biases similar to those observed in psychological studies of human reasoning. These findings highlight challenges in achieving logical consistency in LLMs' normative reasoning and provide insights for enhancing their reliability. All data and code are released publicly at https://github.com/kmineshima/NeuBAROCO.

Traffic accidents represent a critical public health challenge, claiming over 1.35 million lives annually worldwide. Traditional accident prediction models treat road segments independently, failing to capture complex spatial relationships and temporal dependencies in urban transportation networks. This study develops MDAS-GNN, a Multi-Dimensional Attention-based Spatial-diffusion Graph Neural Network integrating three core risk dimensions: traffic safety, infrastructure, and environmental risk. The framework employs feature-specific spatial diffusion mechanisms and multi-head temporal attention to capture dependencies across different time horizons. Evaluated on UK Department for Transport accident data across Central London, South Manchester, and SE Birmingham, MDASGNN achieves superior performance compared to established baseline methods. The model maintains consistently low prediction errors across short, medium, and long-term periods, with particular strength in long-term forecasting. Ablation studies confirm that integrated multi-dimensional features outperform singlefeature approaches, reducing prediction errors by up to 40%. This framework provides civil engineers and urban planners with advanced predictive capabilities for transportation infrastructure design, enabling data-driven decisions for road network optimization, infrastructure resource improvements, and strategic safety interventions in urban development projects.

Most work interpreting reasoning models studies only a single chain-of-thought (CoT), yet these models define distributions over many possible CoTs. We argue that studying a single sample is inadequate for understanding causal influence and the underlying computation. Though fully specifying this distribution is intractable, it can be understood by sampling. We present case studies using resampling to investigate model decisions. First, when a model states a reason for its action, does that reason actually cause the action? In "agentic misalignment" scenarios, we resample specific sentences to measure their downstream effects. Self-preservation sentences have small causal impact, suggesting they do not meaningfully drive blackmail. Second, are artificial edits to CoT sufficient for steering reasoning? These are common in literature, yet take the model off-policy. Resampling and selecting a completion with the desired property is a principled on-policy alternative. We find off-policy interventions yield small and unstable effects compared to resampling in decision-making tasks. Third, how do we understand the effect of removing a reasoning step when the model may repeat it post-edit? We introduce a resilience metric that repeatedly resamples to prevent similar content from reappearing downstream. Critical planning statements resist removal but have large effects when eliminated. Fourth, since CoT is sometimes "unfaithful", can our methods teach us anything in these settings? Adapting causal mediation analysis, we find that hints that have a causal effect on the output without being explicitly mentioned exert a subtle and cumulative influence on the CoT that persists even if the hint is removed. Overall, studying distributions via resampling enables reliable causal analysis, clearer narratives of model reasoning, and principled CoT interventions.

Distributed machine learning is critical for training deep learning models on large datasets with numerous parameters. Current research primarily focuses on leveraging additional hardware resources and powerful computing units to accelerate the training process. As a result, larger batch sizes are often employed to speed up training. However, training with large batch sizes can lead to lower accuracy due to poor generalization. To address this issue, we propose the dual-batch learning scheme, a distributed training method built on the parameter server framework. This approach maximizes training efficiency by utilizing the largest batch size that the hardware can support while incorporating a smaller batch size to enhance model generalization. By using two different batch sizes simultaneously, this method improves accuracy with minimal additional training time. Additionally, to mitigate the time overhead caused by dual-batch learning, we propose the cyclic progressive learning scheme. This technique repeatedly and gradually increases image resolution from low to high during training, thereby reducing training time. By combining cyclic progressive learning with dual-batch learning, our hybrid approach improves both model generalization and training efficiency. Experimental results with ResNet-18 demonstrate that, compared to conventional training methods, our approach improves accuracy by 3.3% while reducing training time by 10.1% on CIFAR-100, and further achieves a 34.8% reduction in training time on ImageNet.

Although mathematics is often considered culturally neutral, the way mathematical problems are presented can carry implicit cultural context. Existing benchmarks like GSM8K are predominantly rooted in Western norms, including names, currencies, and everyday scenarios. In this work, we create culturally adapted variants of the GSM8K test set for five regions Africa, India, China, Korea, and Japan using prompt-based transformations followed by manual verification. We evaluate six large language models (LLMs), ranging from 8B to 72B parameters, across five prompting strategies to assess their robustness to cultural variation in math problem presentation. Our findings reveal a consistent performance gap: models perform best on the original US-centric dataset and comparatively worse on culturally adapted versions. However, models with reasoning capabilities are more resilient to these shifts, suggesting that deeper reasoning helps bridge cultural presentation gaps in mathematical tasks