arXiv subjects

Yusuke Iwasawa

Publications and source records attributed to Yusuke Iwasawa.

At least 73 records · Page 4Linked to original sources

arXiv2025

Topology of Reasoning: Understanding Large Reasoning Models through Reasoning Graph Properties

Recent large-scale reasoning models have achieved state-of-the-art performance on challenging mathematical benchmarks, yet the internal mechanisms underlying their success remain poorly understood. In this work, we introduce the notion of a reasoning graph, extracted by clustering hidden-state representations at each reasoning step, and systematically analyze three key graph-theoretic properties: cyclicity, diameter, and small-world index, across multiple tasks (GSM8K, MATH500, AIME 2024). Our findings reveal that distilled reasoning models (e.g., DeepSeek-R1-Distill-Qwen-32B) exhibit significantly more recurrent cycles (about 5 per sample), substantially larger graph diameters, and pronounced small-world characteristics (about 6x) compared to their base counterparts. Notably, these structural advantages grow with task difficulty and model capacity, with cycle detection peaking at the 14B scale and exploration diameter maximized in the 32B variant, correlating positively with accuracy. Furthermore, we show that supervised fine-tuning on an improved dataset systematically expands reasoning graph diameters in tandem with performance gains, offering concrete guidelines for dataset design aimed at boosting reasoning capabilities. By bridging theoretical insights into reasoning graph structures with practical recommendations for data construction, our work advances both the interpretability and the efficacy of large reasoning models.

cs.AI↗

arXiv2025

AIRoA MoMa Dataset: A Large-Scale Hierarchical Dataset for Mobile Manipulation

As robots transition from controlled settings to unstructured human environments, building generalist agents that can reliably follow natural language instructions remains a central challenge. Progress in robust mobile manipulation requires large-scale multimodal datasets that capture contact-rich and long-horizon tasks, yet existing resources lack synchronized force-torque sensing, hierarchical annotations, and explicit failure cases. We address this gap with the AIRoA MoMa Dataset, a large-scale real-world multimodal dataset for mobile manipulation. It includes synchronized RGB images, joint states, six-axis wrist force-torque signals, and internal robot states, together with a novel two-layer annotation schema of sub-goals and primitive actions for hierarchical learning and error analysis. The initial dataset comprises 25,469 episodes (approx. 94 hours) collected with the Human Support Robot (HSR) and is fully standardized in the LeRobot v2.1 format. By uniquely integrating mobile manipulation, contact-rich interaction, and long-horizon structure, AIRoA MoMa provides a critical benchmark for advancing the next generation of Vision-Language-Action models. The first version of our dataset is now available at https://huggingface.co/datasets/airoa-org/airoa-moma .

cs.RO↗

arXiv2025

Leave No Observation Behind: Real-time Correction for VLA Action Chunks

To improve efficiency and temporal coherence, Vision-Language-Action (VLA) models often predict action chunks; however, this action chunking harms reactivity under inference delay and long horizons. We introduce Asynchronous Action Chunk Correction (A2C2), which is a lightweight real-time chunk correction head that runs every control step and adds a time-aware correction to any off-the-shelf VLA's action chunk. The module combines the latest observation, the predicted action from VLA (base action), a positional feature that encodes the index of the base action within the chunk, and some features from the base policy, then outputs a per-step correction. This preserves the base model's competence while restoring closed-loop responsiveness. The approach requires no retraining of the base policy and is orthogonal to asynchronous execution schemes such as Real Time Chunking (RTC). On the dynamic Kinetix task suite (12 tasks) and LIBERO Spatial, our method yields consistent success rate improvements across increasing delays and execution horizons (+23% point and +7% point respectively, compared to RTC), and also improves robustness for long horizons even with zero injected delay. Since the correction head is small and fast, there is minimal overhead compared to the inference of large VLA models. These results indicate that A2C2 is an effective, plug-in mechanism for deploying high-capacity chunking policies in real-time control.

cs.RO↗

arXiv2025

Unlocking Noise-Resistant Vision: Key Architectural Secrets for Robust Models

While the robustness of vision models is often measured, their dependence on specific architectural design choices is rarely dissected. We investigate why certain vision architectures are inherently more robust to additive Gaussian noise and convert these empirical insights into simple, actionable design rules. Specifically, we performed extensive evaluations on 1,174 pretrained vision models, empirically identifying four consistent design patterns for improved robustness against Gaussian noise: larger stem kernels, smaller input resolutions, average pooling, and supervised vision transformers (ViTs) rather than CLIP ViTs, which yield up to 506 rank improvements and 21.6\%p accuracy gains. We then develop a theoretical analysis that explains these findings, converting observed correlations into causal mechanisms. First, we prove that low-pass stem kernels attenuate noise with a gain that decreases quadratically with kernel size and that anti-aliased downsampling reduces noise energy roughly in proportion to the square of the downsampling factor. Second, we demonstrate that average pooling is unbiased and suppresses noise in proportion to the pooling window area, whereas max pooling incurs a positive bias that grows slowly with window size and yields a relatively higher mean-squared error and greater worst-case sensitivity. Third, we reveal and explain the vulnerability of CLIP ViTs via a pixel-space Lipschitz bound: The smaller normalization standard deviations used in CLIP preprocessing amplify worst-case sensitivity by up to 1.91 times relative to the Inception-style preprocessing common in supervised ViTs. Our results collectively disentangle robustness into interpretable modules, provide a theory that explains the observed trends, and build practical, plug-and-play guidelines for designing vision models more robust against Gaussian noise.

cs.CV↗

arXiv2025

When Instructions Multiply: Measuring and Estimating LLM Capabilities of Multiple Instructions Following

As large language models (LLMs) are increasingly applied to real-world scenarios, it becomes crucial to understand their ability to follow multiple instructions simultaneously. To systematically evaluate these capabilities, we introduce two specialized benchmarks for fundamental domains where multiple instructions following is important: Many Instruction-Following Eval (ManyIFEval) for text generation with up to ten instructions, and Style-aware Mostly Basic Programming Problems (StyleMBPP) for code generation with up to six instructions. Our experiments with the created benchmarks across ten LLMs reveal that performance consistently degrades as the number of instructions increases. Furthermore, given the fact that evaluating all the possible combinations of multiple instructions is computationally impractical in actual use cases, we developed three types of regression models that can estimate performance on both unseen instruction combinations and different numbers of instructions which are not used during training. We demonstrate that a logistic regression model using instruction count as an explanatory variable can predict performance of following multiple instructions with approximately 10% error, even for unseen instruction combinations. We show that relatively modest sample sizes (500 for ManyIFEval and 300 for StyleMBPP) are sufficient for performance estimation, enabling efficient evaluation of LLMs under various instruction combinations.

cs.CL↗

arXiv2025

Dynamic Injection of Entity Knowledge into Dense Retrievers

Dense retrievers often struggle with queries involving less-frequent entities due to their limited entity knowledge. We propose the Knowledgeable Passage Retriever (KPR), a BERT-based retriever enhanced with a context-entity attention layer and dynamically updatable entity embeddings. This design enables KPR to incorporate external entity knowledge without retraining. Experiments on three datasets demonstrate that KPR consistently improves retrieval accuracy, with particularly large gains on the EntityQuestions dataset. When built on the off-the-shelf bge-base retriever, KPR achieves state-of-the-art performance among similarly sized models on two datasets. Models and code are released at https://github.com/knowledgeable-embedding/knowledgeable-embedding.

cs.CL↗

arXiv2025

Enhancing Reference-based Sketch Colorization via Separating Reference Representations

Reference-based sketch colorization methods have garnered significant attention for the potential application in animation and digital illustration production. However, most existing methods are trained with image triplets of sketch, reference, and ground truth that are semantically and spatially similar, while real-world references and sketches often exhibit substantial misalignment. This mismatch in data distribution between training and inference leads to overfitting, consequently resulting in artifacts and signif- icant quality degradation in colorization results. To address this issue, we conduct an in-depth analysis of the reference representations, defined as the intermedium to transfer information from reference to sketch. Building on this analysis, we introduce a novel framework that leverages distinct reference representations to optimize different aspects of the colorization process. Our approach decomposes colorization into modular stages, al- lowing region-specific reference injection to enhance visual quality and reference similarity while mitigating spatial artifacts. Specifically, we first train a backbone network guided by high-level semantic embeddings. We then introduce a background encoder and a style encoder, trained in separate stages, to enhance low-level feature transfer and improve reference similar- ity. This design also enables flexible inference modes suited for a variety of use cases. Extensive qualitative and quantitative evaluations, together with a user study, demonstrate the superior performance of our proposed method compared to existing approaches. Code and pre-trained weight will be made publicly available upon paper acceptance.

cs.GR↗

arXiv2025

AGENTiGraph: A Multi-Agent Knowledge Graph Framework for Interactive, Domain-Specific LLM Chatbots

AGENTiGraph is a user-friendly, agent-driven system that enables intuitive interaction and management of domain-specific data through the manipulation of knowledge graphs in natural language. It gives non-technical users a complete, visual solution to incrementally build and refine their knowledge bases, allowing multi-round dialogues and dynamic updates without specialized query languages. The flexible design of AGENTiGraph, including intent classification, task planning, and automatic knowledge integration, ensures seamless reasoning between diverse tasks. Evaluated on a 3,500-query benchmark within an educational scenario, the system outperforms strong zero-shot baselines (achieving 95.12% classification accuracy, 90.45% execution success), indicating potential scalability to compliance-critical or multi-step queries in legal and medical domains, e.g., incorporating new statutes or research on the fly. Our open-source demo offers a powerful new paradigm for multi-turn enterprise knowledge management that bridges LLMs and structured graphs.

cs.AI↗

arXiv2025

Retrieve-Augmented Generation for Speeding up Diffusion Policy without Additional Training

Diffusion Policies (DPs) have attracted attention for their ability to achieve significant accuracy improvements in various imitation learning tasks. However, DPs depend on Diffusion Models, which require multiple noise removal steps to generate a single action, resulting in long generation times. To solve this problem, knowledge distillation-based methods such as Consistency Policy (CP) have been proposed. However, these methods require a significant amount of training time, especially for difficult tasks. In this study, we propose RAGDP (Retrieve-Augmented Generation for Diffusion Policies) as a novel framework that eliminates the need for additional training using a knowledge base to expedite the inference of pre-trained DPs. In concrete, RAGDP encodes observation-action pairs through the DP encoder to construct a vector database of expert demonstrations. During inference, the current observation is embedded, and the most similar expert action is extracted. This extracted action is combined with an intermediate noise removal step to reduce the number of steps required compared to the original diffusion step. We show that by using RAGDP with the base model and existing acceleration methods, we improve the accuracy and speed trade-off with no additional training. Even when accelerating the models 20 times, RAGDP maintains an advantage in accuracy, with a 7% increase over distillation models such as CP.

cs.LG↗

arXiv2025

LLMs-guided adaptive compensator: Bringing Adaptivity to Automatic Control Systems with Large Language Models

With rapid advances in code generation, reasoning, and problem-solving, Large Language Models (LLMs) are increasingly applied in robotics. Most existing work focuses on high-level tasks such as task decomposition. A few studies have explored the use of LLMs in feedback controller design; however, these efforts are restricted to overly simplified systems, fixed-structure gain tuning, and lack real-world validation. To further investigate LLMs in automatic control, this work targets a key subfield: adaptive control. Inspired by the framework of model reference adaptive control (MRAC), we propose an LLM-guided adaptive compensator framework that avoids designing controllers from scratch. Instead, the LLMs are prompted using the discrepancies between an unknown system and a reference system to design a compensator that aligns the response of the unknown system with that of the reference, thereby achieving adaptivity. Experiments evaluate five methods: LLM-guided adaptive compensator, LLM-guided adaptive controller, indirect adaptive control, learning-based adaptive control, and MRAC, on soft and humanoid robots in both simulated and real-world environments. Results show that the LLM-guided adaptive compensator outperforms traditional adaptive controllers and significantly reduces reasoning complexity compared to the LLM-guided adaptive controller. The Lyapunov-based analysis and reasoning-path inspection demonstrate that the LLM-guided adaptive compensator enables a more structured design process by transforming mathematical derivation into a reasoning task, while exhibiting strong generalizability, adaptability, and robustness. This study opens a new direction for applying LLMs in the field of automatic control, offering greater deployability and practicality compared to vision-language models.

cs.RO↗

arXiv2025

Which Programming Language and What Features at Pre-training Stage Affect Downstream Logical Inference Performance?

Recent large language models (LLMs) have demonstrated remarkable generalization abilities in mathematics and logical reasoning tasks. Prior research indicates that LLMs pre-trained with programming language data exhibit high mathematical and reasoning abilities; however, this causal relationship has not been rigorously tested. Our research aims to verify which programming languages and features during pre-training affect logical inference performance. Specifically, we pre-trained decoder-based language models from scratch using datasets from ten programming languages (e.g., Python, C, Java) and three natural language datasets (Wikipedia, Fineweb, C4) under identical conditions. Thereafter, we evaluated the trained models in a few-shot in-context learning setting on logical reasoning tasks: FLD and bAbi, which do not require commonsense or world knowledge. The results demonstrate that nearly all models trained with programming languages consistently outperform those trained with natural languages, indicating that programming languages contain factors that elicit logic inference performance. In addition, we found that models trained with programming languages exhibit a better ability to follow instructions compared to those trained with natural languages. Further analysis reveals that the depth of Abstract Syntax Trees representing parsed results of programs also affects logical reasoning performance. These findings will offer insights into the essential elements of pre-training for acquiring the foundational abilities of LLMs.

cs.CL↗

arXiv2025

SPARK: Graph-Based Online Semantic Integration System for Robot Task Planning

The ability to update information acquired through various means online during task execution is crucial for a general-purpose service robot. This information includes geometric and semantic data. While SLAM handles geometric updates on 2D maps or 3D point clouds, online updates of semantic information remain unexplored. We attribute the challenge to the online scene graph representation, for its utility and scalability. Building on previous works regarding offline scene graph representations, we study online graph representations of semantic information in this work. We introduce SPARK: Spatial Perception and Robot Knowledge Integration. This framework extracts semantic information from environment-embedded cues and updates the scene graph accordingly, which is then used for subsequent task planning. We demonstrate that graph representations of spatial relationships enhance the robot system's ability to perform tasks in dynamic environments and adapt to unconventional spatial cues, like gestures.

cs.RO↗

arXiv2025

Beyond Induction Heads: In-Context Meta Learning Induces Multi-Phase Circuit Emergence

Transformer-based language models exhibit In-Context Learning (ICL), where predictions are made adaptively based on context. While prior work links induction heads to ICL through a sudden jump in accuracy, this can only account for ICL when the answer is included within the context. However, an important property of practical ICL in large language models is the ability to meta-learn how to solve tasks from context, rather than just copying answers from context; how such an ability is obtained during training is largely unexplored. In this paper, we experimentally clarify how such meta-learning ability is acquired by analyzing the dynamics of the model's circuit during training. Specifically, we extend the copy task from previous research into an In-Context Meta Learning setting, where models must infer a task from examples to answer queries. Interestingly, in this setting, we find that there are multiple phases in the process of acquiring such abilities, and that a unique circuit emerges in each phase, contrasting with the single-phases change in induction heads. The emergence of such circuits can be related to several phenomena known in large language models, and our analysis lead to a deeper understanding of the source of the transformer's ICL ability.

cs.CL↗

arXiv2025

Answer When Needed, Forget When Not: Language Models Pretend to Forget via In-Context Knowledge Unlearning

As large language models (LLMs) are applied across diverse domains, the ability to selectively unlearn specific information is becoming increasingly essential. For instance, LLMs are expected to selectively provide confidential information to authorized internal users, such as employees or trusted partners, while withholding it from external users, including the general public and unauthorized entities. Therefore, we propose a novel method termed ``in-context knowledge unlearning'', which enables the model to selectively forget information in test-time based on the query context. Our method fine-tunes pre-trained LLMs to enable prompt unlearning of target knowledge within the context, while preserving unrelated information. Experiments on TOFU, AGE and RWKU datasets using Llama2-7B/13B and Mistral-7B models demonstrate that our method achieves up to 95% forget accuracy while retaining 80% of unrelated knowledge, significantly outperforming baselines in both in-domain and out-of-domain scenarios. Further investigation of the model's internal behavior revealed that while fine-tuned LLMs generate correct predictions in the middle layers and preserve them up to the final layer. However, the decision to forget is made only at the last layer, i.e. ``LLMs pretend to forget''. Our findings offer valuable insight into the improvement of the robustness of the unlearning mechanisms in LLMs, laying a foundation for future research in the field.

cs.CL↗

arXiv2025

A Comprehensive Survey on Physical Risk Control in the Era of Foundation Model-enabled Robotics

Recent Foundation Model-enabled robotics (FMRs) display greatly improved general-purpose skills, enabling more adaptable automation than conventional robotics. Their ability to handle diverse tasks thus creates new opportunities to replace human labor. However, unlike general foundation models, FMRs interact with the physical world, where their actions directly affect the safety of humans and surrounding objects, requiring careful deployment and control. Based on this proposition, our survey comprehensively summarizes robot control approaches to mitigate physical risks by covering all the lifespan of FMRs ranging from pre-deployment to post-accident stage. Specifically, we broadly divide the timeline into the following three phases: (1) pre-deployment phase, (2) pre-incident phase, and (3) post-incident phase. Throughout this survey, we find that there is much room to study (i) pre-incident risk mitigation strategies, (ii) research that assumes physical interaction with humans, and (iii) essential issues of foundation models themselves. We hope that this survey will be a milestone in providing a high-resolution analysis of the physical risks of FMRs and their control, contributing to the realization of a good human-robot relationship.

cs.RO↗

arXiv2025

ReAgent: Reversible Multi-Agent Reasoning for Knowledge-Enhanced Multi-Hop QA

Recent advances in large language models (LLMs) have significantly improved multi-hop question answering (QA) through direct Chain-of-Thought (CoT) reasoning. However, the irreversible nature of CoT leads to error accumulation, making it challenging to correct mistakes in multi-hop reasoning. This paper introduces ReAgent: a Reversible multi-Agent collaborative framework augmented with explicit backtracking mechanisms, enabling reversible multi-hop reasoning. By incorporating text-based retrieval, information aggregation and validation, our system can detect and correct errors mid-reasoning, leading to more robust and interpretable QA outcomes. The framework and experiments serve as a foundation for future work on error-tolerant QA systems. Empirical evaluations across three benchmarks indicate ReAgent's efficacy, yielding average about 6\% improvements against baseline models.

cs.AI↗

arXiv2025

MMLU-ProX: A Multilingual Benchmark for Advanced Large Language Model Evaluation

Existing large language model (LLM) evaluation benchmarks primarily focus on English, while current multilingual tasks lack parallel questions that specifically assess cross-linguistic reasoning abilities. This dual limitation makes it challenging to comprehensively assess LLMs' performance in the multilingual setting. To fill this gap, we introduce MMLU-ProX, a comprehensive benchmark covering 29 languages, built on an English benchmark. Each language version consists of 11,829 identical questions, enabling direct cross-linguistic comparisons. Additionally, to meet efficient evaluation needs, we provide a lite version containing 658 questions per language. To ensure the high quality of MMLU-ProX, we employ a rigorous development process that involves multiple powerful LLMs for translation, followed by expert review to ensure accurate expression, consistent terminology, and cultural relevance. Building on this, we systematically evaluate 36 state-of-the-art LLMs, including reasoning-enhanced and multilingual-optimized LLMs. The results reveal significant disparities in the multilingual capabilities of LLMs: While they perform well in high-resource languages, their performance declines markedly in low-resource languages, with gaps of up to 24.3%. Through MMLU-ProX, we aim to advance the development of more inclusive AI systems and promote equitable access to technology across global contexts.

cs.CL↗

arXiv2025

Inconsistent Tokenizations Cause Language Models to be Perplexed by Japanese Grammar

Typical methods for evaluating the performance of language models evaluate their ability to answer questions accurately. These evaluation metrics are acceptable for determining the extent to which language models can understand and reason about text in a general sense, but fail to capture nuanced capabilities, such as the ability of language models to recognize and obey rare grammar points, particularly in languages other than English. We measure the perplexity of language models when confronted with the "first person psych predicate restriction" grammar point in Japanese. Weblab is the only tested open source model in the 7-10B parameter range which consistently assigns higher perplexity to ungrammatical psych predicate sentences than grammatical ones. We give evidence that Weblab's uniformly bad tokenization is a possible root cause for its good performance, and show that Llama 3's perplexity on grammatical psych predicate sentences can be reduced by orders of magnitude (28x difference) by restricting test sentences to those with uniformly well-behaved tokenizations. We show in further experiments on machine translation tasks that language models will use alternative grammar patterns in order to produce grammatical sentences when tokenization issues prevent the most natural sentence from being output.

cs.CL↗