arXiv subjects

Yang Deng

Publications and source records attributed to Yang Deng.

At least 55 records · Page 3Linked to original sources

arXiv2025

When Less Language is More: Language-Reasoning Disentanglement Makes LLMs Better Multilingual Reasoners

Multilingual reasoning remains a significant challenge for large language models (LLMs), with performance disproportionately favoring high-resource languages. Drawing inspiration from cognitive neuroscience, which suggests that human reasoning functions largely independently of language processing, we hypothesize that LLMs similarly encode reasoning and language as separable components that can be disentangled to enhance multilingual reasoning. To evaluate this, we perform a causal intervention by ablating language-specific representations at inference time. Experiments on 10 open-weight LLMs spanning 11 typologically diverse languages show that this language-specific ablation consistently boosts multilingual reasoning performance. Layer-wise analyses further confirm that language and reasoning representations can be effectively disentangled throughout the model, yielding improved multilingual reasoning capabilities, while preserving top-layer language features remains essential for maintaining linguistic fidelity. Compared to post-training methods such as supervised fine-tuning or reinforcement learning, our training-free language-reasoning disentanglement achieves comparable or superior results with minimal computational overhead. These findings shed light on the internal mechanisms underlying multilingual reasoning in LLMs and suggest a lightweight and interpretable strategy for improving cross-lingual generalization.

cs.CL↗

arXiv2025

Mastering Diverse, Unknown, and Cluttered Tracks for Robust Vision-Based Drone Racing

Most reinforcement learning(RL)-based methods for drone racing target fixed, obstacle-free tracks, leaving the generalization to unknown, cluttered environments largely unaddressed. This challenge stems from the need to balance racing speed and collision avoidance, limited feasible space causing policy exploration trapped in local optima during training, and perceptual ambiguity between gates and obstacles in depth maps-especially when gate positions are only coarsely specified. To overcome these issues, we propose a two-phase learning framework: an initial soft-collision training phase that preserves policy exploration for high-speed flight, followed by a hard-collision refinement phase that enforces robust obstacle avoidance. An adaptive, noise-augmented curriculum with an asymmetric actor-critic architecture gradually shifts the policy's reliance from privileged gate-state information to depth-based visual input. We further impose Lipschitz constraints and integrate a track-primitive generator to enhance motion stability and cross-environment generalization. We evaluate our framework through extensive simulation and ablation studies, and validate it in real-world experiments on a computationally constrained quadrotor. The system achieves agile flight while remaining robust to gate-position errors, developing a generalizable drone racing framework with the capability to operate in diverse, partially unknown and cluttered environments. https://yufengsjtu.github.io/MasterRacing.github.io/

cs.RO↗

arXiv2025

Steady-State Drifting Equilibrium Analysis of Single-Track Two-Wheeled Robots for Controller Design

Drifting is an advanced driving technique where the wheeled robot's tire-ground interaction breaks the common non-holonomic pure rolling constraint. This allows high-maneuverability tasks like quick cornering, and steady-state drifting control enhances motion stability under lateral slip conditions. While drifting has been successfully achieved in four-wheeled robot systems, its application to single-track two-wheeled (STTW) robots, such as unmanned motorcycles or bicycles, has not been thoroughly studied. To bridge this gap, this paper extends the drifting equilibrium theory to STTW robots and reveals the mechanism behind the steady-state drifting maneuver. Notably, the counter-steering drifting technique used by skilled motorcyclists is explained through this theory. In addition, an analytical algorithm based on intrinsic geometry and kinematics relationships is proposed, reducing the computation time by four orders of magnitude while maintaining less than 6% error compared to numerical methods. Based on equilibrium analysis, a model predictive controller (MPC) is designed to achieve steady-state drifting and equilibrium points transition, with its effectiveness and robustness validated through simulations.

cs.RO↗

arXiv2025

RecToM: A Benchmark for Evaluating Machine Theory of Mind in LLM-based Conversational Recommender Systems

Large Language models are revolutionizing the conversational recommender systems through their impressive capabilities in instruction comprehension, reasoning, and human interaction. A core factor underlying effective recommendation dialogue is the ability to infer and reason about users' mental states (such as desire, intention, and belief), a cognitive capacity commonly referred to as Theory of Mind. Despite growing interest in evaluating ToM in LLMs, current benchmarks predominantly rely on synthetic narratives inspired by Sally-Anne test, which emphasize physical perception and fail to capture the complexity of mental state inference in realistic conversational settings. Moreover, existing benchmarks often overlook a critical component of human ToM: behavioral prediction, the ability to use inferred mental states to guide strategic decision-making and select appropriate conversational actions for future interactions. To better align LLM-based ToM evaluation with human-like social reasoning, we propose RecToM, a novel benchmark for evaluating ToM abilities in recommendation dialogues. RecToM focuses on two complementary dimensions: Cognitive Inference and Behavioral Prediction. The former focus on understanding what has been communicated by inferring the underlying mental states. The latter emphasizes what should be done next, evaluating whether LLMs can leverage these inferred mental states to predict, select, and assess appropriate dialogue strategies. Extensive experiments on state-of-the-art LLMs demonstrate that RecToM poses a significant challenge. While the models exhibit partial competence in recognizing mental states, they struggle to maintain coherent, strategic ToM reasoning throughout dynamic recommendation dialogues, particularly in tracking evolving intentions and aligning conversational strategies with inferred mental states.

cs.AI↗

arXiv2025

The Rise of Parameter Specialization for Knowledge Storage in Large Language Models

Over time, a growing wave of large language models from various series has been introduced to the community. Researchers are striving to maximize the performance of language models with constrained parameter sizes. However, from a microscopic perspective, there has been limited research on how to better store knowledge in model parameters, particularly within MLPs, to enable more effective utilization of this knowledge by the model. In this work, we analyze twenty publicly available open-source large language models to investigate the relationship between their strong performance and the way knowledge is stored in their corresponding MLP parameters. Our findings reveal that as language models become more advanced and demonstrate stronger knowledge capabilities, their parameters exhibit increased specialization. Specifically, parameters in the MLPs tend to be more focused on encoding similar types of knowledge. We experimentally validate that this specialized distribution of knowledge contributes to improving the efficiency of knowledge utilization in these models. Furthermore, by conducting causal training experiments, we confirm that this specialized knowledge distribution plays a critical role in improving the model's efficiency in leveraging stored knowledge.

cs.CL↗

arXiv2025

Trade-offs in Large Reasoning Models: An Empirical Analysis of Deliberative and Adaptive Reasoning over Foundational Capabilities

Recent advancements in Large Reasoning Models (LRMs), such as OpenAI's o1/o3 and DeepSeek-R1, have demonstrated remarkable performance in specialized reasoning tasks through human-like deliberative thinking and long chain-of-thought reasoning. However, our systematic evaluation across various model families (DeepSeek, Qwen, and LLaMA) and scales (7B to 32B) reveals that acquiring these deliberative reasoning capabilities significantly reduces the foundational capabilities of LRMs, including notable declines in helpfulness and harmlessness, alongside substantially increased inference costs. Importantly, we demonstrate that adaptive reasoning -- employing modes like Zero-Thinking, Less-Thinking, and Summary-Thinking -- can effectively alleviate these drawbacks. Our empirical insights underline the critical need for developing more versatile LRMs capable of dynamically allocating inference-time compute according to specific task characteristics.

cs.AI↗

arXiv2025

Do Retrieval Augmented Language Models Know When They Don't Know?

Existing large language models (LLMs) occasionally generate plausible yet factually incorrect responses, known as hallucinations. Two main approaches have been proposed to mitigate hallucinations: retrieval-augmented language models (RALMs) and refusal post-training. However, current research predominantly focuses on their individual effectiveness while overlooking the evaluation of the refusal capability of RALMs. Ideally, if RALMs know when they do not know, they should refuse to answer.In this study, we ask the fundamental question: Do RALMs know when they don't know? Specifically, we investigate three questions. First, are RALMs well calibrated with respect to different internal and external knowledge states? We examine the influence of various factors. Contrary to expectations, when all retrieved documents are irrelevant, RALMs still tend to refuse questions they could have answered correctly. Next, given the model's pronounced \textbf{over-refusal} behavior, we raise a second question: How does a RALM's refusal ability align with its calibration quality? Our results show that the over-refusal problem can be mitigated through in-context fine-tuning. However, we observe that improved refusal behavior does not necessarily imply better calibration or higher overall accuracy. Finally, we ask: Can we combine refusal-aware RALMs with uncertainty-based answer abstention to mitigate over-refusal? We develop a simple yet effective refusal mechanism for refusal-post-trained RALMs that improves their overall answer quality by balancing refusal and correct answers. Our study provides a more comprehensive understanding of the factors influencing RALM behavior. Meanwhile, we emphasize that uncertainty estimation for RALMs remains an open problem deserving deeper investigation.

cs.CL↗

arXiv2025

Towards Machine Learning-based Model Predictive Control for HVAC Control in Multi-Context Buildings at Scale via Ensemble Learning

The building thermodynamics model, which predicts real-time indoor temperature changes under potential HVAC (Heating, Ventilation, and Air Conditioning) control operations, is crucial for optimizing HVAC control in buildings. While pioneering studies have attempted to develop such models for various building environments, these models often require extensive data collection periods and rely heavily on expert knowledge, making the modeling process inefficient and limiting the reusability of the models. This paper explores a model ensemble perspective that utilizes existing developed models as base models to serve a target building environment, thereby providing accurate predictions while reducing the associated efforts. Given that building data streams are non-stationary and the number of base models may increase, we propose a Hierarchical Reinforcement Learning (HRL) approach to dynamically select and weight the base models. Our approach employs a two-tiered decision-making process: the high-level focuses on model selection, while the low-level determines the weights of the selected models. We thoroughly evaluate the proposed approach through offline experiments and an on-site case study, and the experimental results demonstrate the effectiveness of our method.

cs.AI↗

arXiv2025

Reinforced Strategy Optimization for Conversational Recommender Systems via Network-of-Experts

Conversational Recommender Systems (CRSs) aim to provide personalized recommendations through multi-turn natural language interactions with users. Given the strong interaction and reasoning skills of Large Language Models (LLMs), leveraging LLMs for CRSs has recently emerged as a promising direction. However, existing LLM-based methods often lack explicit optimization of interaction strategies, instead relying on unified prompts and the LLM's internal knowledge to decide how to interact, which can lead to suboptimal outcomes. In this paper, we propose a novel Reinforced Strategy Optimization (RSO) method for CRS, which decomposes the process of generating strategy-driven response decisions into the macro-level strategy planning and micro-level strategy adaptation through a network-of-experts architecture. At the macro level, a Planner expert selects macro-level interaction strategies (e.g., recommend, explain, encourage). At the micro level, an Actor expert generates detailed responses conditioned on the selected macro-level strategy, guided by auxiliary experts that provide complementary information such as user preferences and factual grounding. This hierarchical decomposition disentangles the optimization of different sub-tasks involved in CRS response generation, enabling more tractable learning at each level. To address the scarcity of high-quality multi-turn training data, we formulate strategy learning as a reinforcement learning problem, guided by an LLM-based reward model to achieve automatic strategy exploration. Extensive experiments show that RSO significantly improves interaction performance compared to state-of-the-art baselines, demonstrating the effectiveness of explicit hierarchical strategy optimization for CRS.

cs.CL↗

arXiv2025

Bounds of Chain-of-Thought Robustness: Reasoning Steps, Embed Norms, and Beyond

Existing research indicates that the output of Chain-of-Thought (CoT) is significantly affected by input perturbations. Although many methods aim to mitigate such impact by optimizing prompts, a theoretical explanation of how these perturbations influence CoT outputs remains an open area of research. This gap limits our in-depth understanding of how input perturbations propagate during the reasoning process and hinders further improvements in prompt optimization methods. Therefore, in this paper, we theoretically analyze the effect of input perturbations on the fluctuation of CoT outputs. We first derive an upper bound for input perturbations under the condition that the output fluctuation is within an acceptable range, based on which we prove that: (i) This upper bound is positively correlated with the number of reasoning steps in the CoT; (ii) Even an infinitely long reasoning process cannot eliminate the impact of input perturbations. We then apply these conclusions to the Linear Self-Attention (LSA) model, which can be viewed as a simplified version of the Transformer. For the LSA model, we prove that the upper bound for input perturbation is negatively correlated with the norms of the input embedding and hidden state vectors. To validate this theoretical analysis, we conduct experiments on three mainstream datasets and four mainstream models. The experimental results align with our theoretical analysis, empirically demonstrating the correctness of our findings.

cs.CL↗

arXiv2025

Chain of Strategy Optimization Makes Large Language Models Better Emotional Supporter

The growing emotional stress in modern society has increased the demand for Emotional Support Conversations (ESC). While Large Language Models (LLMs) show promise for ESC, they face two key challenges: (1) low strategy selection accuracy, and (2) preference bias, limiting their adaptability to emotional needs of users. Existing supervised fine-tuning (SFT) struggles to address these issues, as it rigidly trains models on single gold-standard responses without modeling nuanced strategy trade-offs. To overcome these limitations, we propose Chain-of-Strategy Optimization (CSO), a novel approach that optimizes strategy selection preferences at each dialogue turn. We first leverage Monte Carlo Tree Search to construct ESC-Pro, a high-quality preference dataset with turn-level strategy-response pairs. Training on ESC-Pro with CSO improves both strategy accuracy and bias mitigation, enabling LLMs to generate more empathetic and contextually appropriate responses. Experiments on LLaMA-3.1-8B, Gemma-2-9B, and Qwen2.5-7B demonstrate that CSO outperforms standard SFT, highlighting the efficacy of fine-grained, turn-level preference modeling in ESC.

cs.CL↗

arXiv2025

AdaSteer: Your Aligned LLM is Inherently an Adaptive Jailbreak Defender

Despite extensive efforts in safety alignment, large language models (LLMs) remain vulnerable to jailbreak attacks. Activation steering offers a training-free defense method but relies on fixed steering coefficients, resulting in suboptimal protection and increased false rejections of benign inputs. To address this, we propose AdaSteer, an adaptive activation steering method that dynamically adjusts model behavior based on input characteristics. We identify two key properties: Rejection Law (R-Law), which shows that stronger steering is needed for jailbreak inputs opposing the rejection direction, and Harmfulness Law (H-Law), which differentiates adversarial and benign inputs. AdaSteer steers input representations along both the Rejection Direction (RD) and Harmfulness Direction (HD), with adaptive coefficients learned via logistic regression, ensuring robust jailbreak defense while preserving benign input handling. Experiments on LLaMA-3.1, Gemma-2, and Qwen2.5 show that AdaSteer outperforms baseline methods across multiple jailbreak attacks with minimal impact on utility. Our results highlight the potential of interpretable model internals for real-time, flexible safety enforcement in LLMs.

cs.CR↗

arXiv2025

Why Stop at One Error? Benchmarking LLMs as Data Science Code Debuggers for Multi-Hop and Multi-Bug Errors

LLMs are transforming software development, yet current code generation and code repair benchmarks mainly assess syntactic and functional correctness in simple, single-error cases. LLMs' capabilities to autonomously find and fix runtime logical errors in complex data science code remain largely unexplored. To address this gap, we introduce DSDBench: the Data Science Debugging Benchmark, the first benchmark for systematic evaluation of LLMs on multi-hop error tracing and multi-bug detection in data science code debugging. DSDBench adapts datasets from existing data science task benchmarks, such as DABench and MatPlotBench, featuring realistic data science debugging tasks with automatically synthesized multi-hop, multi-bug code snippets. DSDBench includes 1,117 annotated samples with 741 cause-effect error pairs and runtime error messages. Evaluations of state-of-the-art LLMs on DSDBench show significant performance gaps, highlighting challenges in debugging logical runtime errors in data science code. DSDBench offers a crucial resource to evaluate and improve LLMs' debugging and reasoning capabilities, enabling more reliable AI-assisted data science in the future. DSDBench is publicly available at github.com/KevinCL16/DSDBench.

cs.CL↗

arXiv2025

From Personas to Talks: Revisiting the Impact of Personas on LLM-Synthesized Emotional Support Conversations

The rapid advancement of Large Language Models (LLMs) has revolutionized the generation of emotional support conversations (ESC), offering scalable solutions with reduced costs and enhanced data privacy. This paper explores the role of personas in the creation of ESC by LLMs. Our research utilizes established psychological frameworks to measure and infuse persona traits into LLMs, which then generate dialogues in the emotional support scenario. We conduct extensive evaluations to understand the stability of persona traits in dialogues, examining shifts in traits post-generation and their impact on dialogue quality and strategy distribution. Experimental results reveal several notable findings: 1) LLMs can infer core persona traits, 2) subtle shifts in emotionality and extraversion occur, influencing the dialogue dynamics, and 3) the application of persona traits modifies the distribution of emotional support strategies, enhancing the relevance and empathetic quality of the responses. These findings highlight the potential of persona-driven LLMs in crafting more personalized, empathetic, and effective emotional support dialogues, which has significant implications for the future design of AI-driven emotional support systems.

cs.CL↗

arXiv2025

Towards Goal-oriented Intelligent Tutoring Systems in Online Education

Interactive Intelligent Tutoring Systems (ITSs) enhance traditional ITSs by promoting effective learning through interactions and problem resolution in online education. Yet, proactive engagement, prioritizing resource optimization with planning and assessment capabilities, is often overlooked in current ITS designs. In this work, we investigate a new task, named Goal-oriented Intelligent Tutoring Systems (GITS), which aims to enable the student's mastery of a designated concept by strategically planning a customized sequence of exercises and assessment. To address the problem of goal-oriented policy learning in GITS, we propose a novel graph-based reinforcement learning framework, named Planning-Assessment-Interaction (PAI). Specifically, we first leverage cognitive structure information to improve state representation learning and action selection for planning the next action, which can be either to tutor an exercise or to assess the target concept. Further, we use a dynamically updated cognitive diagnosis model to simulate student responses to exercises and concepts. Three benchmark datasets across different subjects are constructed for enabling offline academic research on GITS. Experimental results demonstrate the effectiveness and efficiency of PAI and extensive analyses of various types of students are conducted to showcase the challenges in this task.

cs.CY↗

arXiv2025

Mapless Collision-Free Flight via MPC using Dual KD-Trees in Cluttered Environments

Collision-free flight in cluttered environments is a critical capability for autonomous quadrotors. Traditional methods often rely on detailed 3D map construction, trajectory generation, and tracking. However, this cascade pipeline can introduce accumulated errors and computational delays, limiting flight agility and safety. In this paper, we propose a novel method for enabling collision-free flight in cluttered environments without explicitly constructing 3D maps or generating and tracking collision-free trajectories. Instead, we leverage Model Predictive Control (MPC) to directly produce safe actions from sparse waypoints and point clouds from a depth camera. These sparse waypoints are dynamically adjusted online based on nearby obstacles detected from point clouds. To achieve this, we introduce a dual KD-Tree mechanism: the Obstacle KD-Tree quickly identifies the nearest obstacle for avoidance, while the Edge KD-Tree provides a robust initial guess for the MPC solver, preventing it from getting stuck in local minima during obstacle avoidance. We validate our approach through extensive simulations and real-world experiments. The results show that our approach significantly outperforms the mapping-based methods and is also superior to imitation learning-based methods, demonstrating reliable obstacle avoidance at up to 12 m/s in simulations and 6 m/s in real-world tests. Our method provides a simple and robust alternative to existing methods. The code is publicly available at https://github.com/SJTU-ViSYS-team/avoid-mpc.

cs.RO↗

arXiv2025

Multi-Layer Attention is the Amplifier of Demonstration Effectiveness

Numerous studies have investigated the underlying mechanisms of in-context learning (ICL) effectiveness to inspire the design of related methods. However, existing work predominantly assumes the effectiveness of the demonstrations provided within ICL, while many research indicates that not all demonstrations are effective, failing to yielding any performance improvement during ICL. Therefore, in this paper, we investigate the reasons behind demonstration ineffectiveness. Our analysis is based on gradient flow and linear self-attention models. By setting the gradient flow to zero, we deduce that a demonstration becomes ineffective if its information has either been learned by the model or is irrelevant to the user query. Furthermore, we demonstrate that in multi-layer models, the disparity in effectiveness among demonstrations is amplified with layer increasing, causing the model to focus more on effective ones. Considering that current demonstration selection methods primarily focus on the relevance to the user query while overlooking the information that the model has already assimilated, we propose a novel method called GradS, which leverages gradient flow for demonstration selection. We use the magnitude of the gradient flow of the demonstration with respect to a given user query as the criterion, thereby ensuring the effectiveness of the chosen ones. We validate our derivation and GradS on four prominent LLMs across five mainstream datasets. The experimental results confirm that the disparity in effectiveness among demonstrations is magnified as the model layer increases, substantiating our derivations. Moreover, GradS achieves a relative improvement of $6.8\%$ on average over the strongest baselines, demonstrating its effectiveness.

cs.CL↗

arXiv2025

Learning-to-Context Slope: Evaluating In-Context Learning Effectiveness Beyond Performance Illusions

In-context learning (ICL) has emerged as an effective approach to enhance the performance of large language models (LLMs). However, its effectiveness varies significantly across models and tasks, posing challenges for practitioners to determine when ICL reliably improves performance. Current evaluation approaches, reliant on performance change after applying ICL, suffer from low reliability, poor attribution, and impracticality in data-insufficient scenarios. We propose the Learning-to-Context Slope (LCS), a novel metric that quantifies ICL effectiveness by modeling the slope between learning gain (loss decrease from demonstrations) and contextual relevance (demonstration-input relevance). LCS addresses key limitations of performance-based metrics: (1) it captures continuous loss changes even when outputs are incorrect, improving reliability; (2) its formulation attributes ICL failures to weak contextual alignment (inability to adapt inputs to demonstrations) or strong output calibration (self-verification of correctness); and (3) it minimizes reliance on labeled data via synthetic evaluation. Extensive experiments demonstrate that LCS strongly correlates with performance improvements in labeled settings and reliably reflects true effectiveness in biased or data-scarce scenarios. Further analysis reveals actionable thresholds for LCS and identifies model capabilities critical to ICL success.

cs.CL↗