Reasoning in large language models (LLMs) tends to produce substantially longer token generation sequences than simpler language modeling tasks. This extended generation length reflects the multi-step, compositional nature of reasoning and is often correlated with higher solution accuracy. From an efficiency perspective, longer token generation exacerbates the inherently sequential and memory-bound decoding phase of LLMs. However, not all parts of this expensive reasoning process are equally difficult to generate. We leverage this observation by offloading only the most challenging parts of the reasoning process to a larger, more capable model, while performing most of the generation with a smaller, more efficient model; furthermore, we teach the smaller model to identify these difficult segments and independently trigger offloading when needed. To enable this behavior, we annotate difficult segments across 18k reasoning traces from the OpenR1-Math-220k chain-of-thought (CoT) dataset. We then apply supervised fine-tuning (SFT) and reinforcement learning fine-tuning (RLFT) to a 1.5B-parameter reasoning model, training it to learn to offload the most challenging parts of its own reasoning process to a larger model. This approach improves AIME24 reasoning accuracy by 24% and 28.3% while offloading 1.35% and 5% of the generated tokens respectively. We open-source our SplitReason model, data, code and logs.

Automatic extraction of definitions from legal texts is critical for enhancing the comprehension and clarity of complex legal corpora such as the United States Code (U.S.C.). We present an advanced NLP system leveraging transformer-based architectures to automatically extract defined terms, their definitions, and their scope from the U.S.C. We address the challenges of automatically identifying legal definitions, extracting defined terms, and determining their scope within this complex corpus of over 200,000 pages of federal statutory law. Building upon previous feature-based machine learning methods, our updated model employs domain-specific transformers (Legal-BERT) fine-tuned specifically for statutory texts, significantly improving extraction accuracy. Our work implements a multi-stage pipeline that combines document structure analysis with state-of-the-art language models to process legal text from the XML version of the U.S. Code. Each paragraph is first classified using a fine-tuned legal domain BERT model to determine if it contains a definition. Our system then aggregates related paragraphs into coherent definitional units and applies a combination of attention mechanisms and rule-based patterns to extract defined terms and their jurisdictional scope. The definition extraction system is evaluated on multiple titles of the U.S. Code containing thousands of definitions, demonstrating significant improvements over previous approaches. Our best model achieves 96.8% precision and 98.9% recall (98.2% F1-score), substantially outperforming traditional machine learning classifiers. This work contributes to improving accessibility and understanding of legal information while establishing a foundation for downstream legal reasoning tasks.

--Vision systems are increasingly deployed in critical domains such as surveillance, law enforcement, and transportation. However, their vulnerabilities to rare or unforeseen scenarios pose significant safety risks. T o address these challenges, we introduce Context-A wareness and Interpretability of Rare Occurrences (CAIRO), an ontology-based human-assistive discovery framework for failure cases (or CP - Critical Phenomena) detection and formalization. CAIRO by design incentivizes human-in-the-loop for testing and evaluation of criticality that arises from misdetections, adversarial attacks, and hallucinations in AI black-box models. Our robust analysis of object detection model(s) failures in automated driving systems (ADS) showcases scalable and interpretable ways of formalizing the observed gaps between camera perception and real-world contexts, resulting in test cases stored as explicit knowledge graphs (in OWL/XML format) amenable for sharing, downstream analysis, logical reasoning, and accountability. I NTRODUCTION Formal verification techniques are a norm in chip design, but they remain elusive in computer vision (CV) applications. The reason being CV applications are deemed open-ended, often trained on millions of data and billions of parameters to learn a few hundreds of labels. Finetuning practices are commonly used to tailor them to specific needs, but with no standard testing procedures in place providing guidance for their application to ensure fail-safe behaviors, critical systems like Autonomous V ehicles (A V) are bound to fail [1].

Predominant LLMs focus on high-resource languages while leaving low-resource languages, particularly those in Southeast Asia (SEA), underrepresented. In addition, those models are general-purpose and pay limited attention to the e-commerce domain. To overcome these limitations, we introduce Compass-v2, a lightweight Mixture-of-Experts (MoE) model specifically designed for Southeast Asian languages and e-commerce applications. To balance model performance and inference cost, the model is designed with 30B total parameters and 5B active parameters, incorporating both fine-grained and shared expert modules. To enhance multilingual performance, we curated and constructed a high-quality, industry-leading SEA dataset, to the best of our knowledge. To boost performance in the e-commerce domain, we built a dataset comprising hundreds of billions of tokens, sourced through external data mining and internal platform collection. Besides, we pioneered a hybrid reasoning model that supports both fast thinking and deep thinking within a unified framework to enhance the reasoning capabilities, diverging from the conventional industry practice of deploying two separate models. Through extensive experimental evaluations, our model demonstrates state-of-the-art SEA multilingual and e-commerce performance among sub-30B models, while maintaining significantly lower inference cost.

Large Language Models (LLMs) have made remarkable breakthroughs in reasoning, yet continue to struggle with hallucinations, logical errors, and inability to self-correct during complex multi-step tasks. Current approaches like chain-of-thought prompting offer limited reasoning capabilities that fail when precise step validation is required. We propose Environment Augmented Generation (EAG), a framework that enhances LLM reasoning through: (1) real-time environmental feedback validating each reasoning step, (2) dynamic branch exploration for investigating alternative solution paths when faced with errors, and (3) experience-based learning from successful reasoning trajectories. Unlike existing methods, EAG enables deliberate backtracking and strategic replanning through tight integration of execution feedback with branching exploration. Our a1-32B model achieves state-of-the-art performance among similar-sized models across all benchmarks, matching larger models like o1 on competition mathematics while outperforming comparable models by up to 24.4 percentage points. Analysis reveals EAG's distinctive scaling pattern: initial token investment in environment interaction yields substantial long-term performance dividends, with advantages amplifying proportionally to task complexity. EAG's theoretical framework demonstrates how environment interactivity and systematic branch exploration together establish a new paradigm for reliable machine reasoning, particularly for problems requiring precise multi-step calculation and logical verification.

The rise of Large Language Models (LLMs) has redefined the AI landscape, particularly due to their ability to encode factual and commonsense knowledge, and their outstanding performance in tasks requiring reasoning. Despite these advances, hallucinations and reasoning errors remain a significant barrier to their deployment in high-stakes settings. In this work, we observe that even the most prominent LLMs, such as OpenAI-o1, suffer from high rates of reasoning errors and hallucinations on tasks requiring commonsense reasoning over obscure, long-tail entities. To investigate this limitation, we present a new dataset for Commonsense reasoning over Long-Tail entities (CoLoTa), that consists of 3,300 queries from question answering and claim verification tasks and covers a diverse range of commonsense reasoning skills. We remark that CoLoTa can also serve as a Knowledge Graph Question Answering (KGQA) dataset since the support of knowledge required to answer its queries is present in the Wikidata knowledge graph. However, as opposed to existing KGQA benchmarks that merely focus on factoid questions, our CoLoTa queries also require commonsense reasoning. Our experiments with strong LLM-based KGQA methodologies indicate their severe inability to answer queries involving commonsense reasoning. Hence, we propose CoLoTa as a novel benchmark for assessing both (i) LLM commonsense reasoning capabilities and their robustness to hallucinations on long-tail entities and (ii) the commonsense reasoning capabilities of KGQA methods.

Reasoning over Knowledge Graphs (KGs) plays a pivotal role in knowledge graph completion or question answering systems, providing richer and more accurate triples and attributes. As numerical attributes become increasingly essential in characterizing entities and relations in KGs, the ability to reason over these attributes has gained significant importance. Existing graph-based methods such as Graph Neural Networks (GNNs) and Knowledge Graph Embeddings (KGEs), primarily focus on aggregating homogeneous local neighbors and implicitly embedding diverse triples. However, these approaches often fail to fully leverage the potential of logical paths within the graph, limiting their effectiveness in exploiting the reasoning process. To address these limitations, we propose ChainsFormer, a novel chain-based framework designed to support numerical reasoning. Chainsformer not only explicitly constructs logical chains but also expands the reasoning depth to multiple hops. Specially, we introduces Relation-Attribute Chains (RA-Chains), a specialized logic chain, to model sequential reasoning patterns. ChainsFormer captures the step-by-step nature of multi-hop reasoning along RA-Chains by employing sequential in-context learning. To mitigate the impact of noisy chains, we propose a hyperbolic affinity scoring mechanism that selects relevant logic chains in a variable-resolution space. Furthermore, ChainsFormer incorporates an attention-based numerical reasoner to identify critical reasoning paths, enhancing both reasoning accuracy and transparency. Experimental results demonstrate that ChainsFormer significantly outperforms state-of-the-art methods, achieving up to a 20.0% improvement in performance. The implementations are available at https://github.com/zhaodazhuang2333/ChainsFormer.

Our work aims to advance the social reasoning of embodied artificial intelligence (AI) agents in real-world social interactions. Recently, language models (LMs) and foundational models (FMs) are being utilized as automatic evaluators of human-AI interactions with the goal of eventually being used to improve the policy of the AI agent. To enable further research in this direction, we introduce a large-scale real-world Human Robot Social Interaction (HSRI) Dataset to benchmark the capabilities of LMs and FMs to identify and reason about social interactions, specifically with regard to robot social errors and competencies . Our dataset consists of 400 real-world human social robot interaction videos and over 10K annotations, detailing the robot's social errors, competencies, rationale, and corrective actions, capturing unique aspects of human-AI interaction only present in real-world interactions. To further assess AI models' ability to reason about social interactions, we propose eight new benchmark tasks for evaluating centered around whether AI models can (1) evaluate social interactions via detecting social errors and competencies, (2) identify the explanatory factors associated to errors and competencies, (3) understand the flow of real-world social interactions, and (4) provide reasons and corrective actions for social errors. Human studies and experiments with modern LMs and FMs reveal that current models struggle with these tasks, demonstrating that our dataset and benchmark provides a step forward towards socially intelligent AI.

This work aligns deep learning (DL) with human reasoning capabilities and needs to enable more efficient, interpretable, and robust image classification. We approach this from three perspectives: explainability, causality, and biological vision. Introduction and background open this work before diving into operative chapters. First, we assess neural networks' visualization techniques for medical images and validate an explainable-by-design method for breast mass classification. A comprehensive review at the intersection of XAI and causality follows, where we introduce a general scaffold to organize past and future research, laying the groundwork for our second perspective. In the causality direction, we propose novel modules that exploit feature co-occurrence in medical images, leading to more effective and explainable predictions. We further introduce CROCODILE, a general framework that integrates causal concepts, contrastive learning, feature disentanglement, and prior knowledge to enhance generalization. Lastly, we explore biological vision, examining how humans recognize objects, and propose CoCoReco, a connectivity-inspired network with context-aware attention mechanisms. Overall, our key findings include: (i) simple activation maximization lacks insight for medical imaging DL models; (ii) prototypical-part learning is effective and radiologically aligned; (iii) XAI and causal ML are deeply connected; (iv) weak causal signals can be leveraged without a priori information to improve performance and interpretability; (v) our framework generalizes across medical domains and out-of-distribution data; (vi) incorporating biological circuit motifs improves human-aligned recognition. This work contributes toward human-aligned DL and highlights pathways to bridge the gap between research and clinical adoption, with implications for improved trust, diagnostic accuracy, and safe deployment.

Recent advancements in dialogue policy planning have emphasized optimizing system agent policies to achieve predefined goals, focusing on strategy design, trajectory acquisition, and efficient training paradigms. However, these approaches often overlook the critical role of user characteristics, which are essential in real-world scenarios like conversational search and recommendation, where interactions must adapt to individual user traits such as personality, preferences, and goals. To address this gap, we first conduct a comprehensive study utilizing task-specific user personas to systematically assess dialogue policy planning under diverse user behaviors. By leveraging realistic user profiles for different tasks, our study reveals significant limitations in existing approaches, highlighting the need for user-tailored dialogue policy planning. Building on this foundation, we present the User-Tailored Dialogue Policy Planning (UDP) framework, which incorporates an Intrinsic User World Model to model user traits and feedback. UDP operates in three stages: (1) User Persona Portraying, using a diffusion model to dynamically infer user profiles; (2) User Feedback Anticipating, leveraging a Brownian Bridge-inspired anticipator to predict user reactions; and (3) User-Tailored Policy Planning, integrating these insights to optimize response strategies. To ensure robust performance, we further propose an active learning approach that prioritizes challenging user personas during training. Comprehensive experiments on benchmarks, including collaborative and non-collaborative settings, demonstrate the effectiveness of UDP in learning user-specific dialogue strategies. Results validate the protocol's utility and highlight UDP's robustness, adaptability, and potential to advance user-centric dialogue systems.