arXiv subjects

Zheng Hu

Publications and source records attributed to Zheng Hu.

At least 19 recordsLinked to original sources

arXiv2026

A/B Agent: A Self-Evolving Agent for Strategy Iteration in Industrial A/B Testing

Industrial recommendation strategy iteration heavily relies on large-scale A/B experimentation. Traditional tuning requires experts to repeatedly design strategies, configure experiments, analyze results, and adjust parameters, making the process labor-intensive and time-consuming. Meanwhile, valuable knowledge from historical experiments is often fragmented, making systematic reuse difficult through manual expert effort alone. Existing RAG agents partially alleviate this burden by retrieving prior strategies, but typically organize experience in a flat manner, overlooking the hierarchical relationships among business scenarios, recommendation stages, optimization objectives, and experimental contexts. This often results in mismatched retrieval and limited cross-scenario transfer, while preventing agents from continuously refining strategies and parameters through sequential A/B feedback. % To address these limitations, we propose A/B Agent, a closed-loop A/B agent for industrial recommendation strategy optimization. The framework comprises three tightly coupled core components: Historical Strategy Knowledge Organization, Autonomous Target-Aware Strategy Generation, and Experiment-Guided Strategy Self-Evolution. It organizes historical strategies into a hierarchical experience tree, retrieves transferable evidence through multi-path Tree-RAG to generate executable strategies, and continuously analyzes online A/B feedback to guide autonomous tuning and update the experience tree for self-evolution. Extensive offline and online evaluations demonstrate its effectiveness, including a 4.829% improvement in GMV in a real-world short-video e-commerce recommendation system while maintaining positive gains across all guardrail metrics.

cs.AI↗

arXiv2026

Instance-Level Post Hoc Uncertainty Quantification in Object Detection

Object detection is a safety-critical component of autonomous driving. It is essential to quantify the uncertainty in bounding-box predictions for safety assurance. Post hoc uncertainty quantification without retraining aligns with real-world deployment requirements; therefore, we employ the Laplace approximation. Because instance-level uncertainty is needed, linearized inference methods that require multiple backpropagations are not time-efficient, and sampling-based methods are not fully post hoc. We propose Monte-Carlo generalized linearized model (MC-GLM), which provides instance-level and approximately post hoc uncertainty quantification. The number of samples required in the Monte Carlo step is constant and independent of the number of output instances, so it can be parallelized. Experiments on the nuScenes dataset with the CenterPoint detector validate the effectiveness of our method, and the resulting uncertainties exhibit good quality.

cs.CV↗

arXiv2026

Calibrating Overconfidence Without Sacrificing Confidence: Probe-Conditioned Head Intervention for LLMs

Large language models often express high confidence in answers that are wrong. Standard calibration remedies typically act globally or at the score level, reducing unwarranted confidence but also risking erosion of warranted confidence on correct answers. We introduce Probe-Conditioned Head Intervention (PCHI), an inference-time method that uses a frozen probe to detect likely wrong-but-confident responses and conditionally rescales downstream attention-head outputs during confidence generation. On Qwen3-4B-Instruct solving OpenMathInstruct problems with a structured binary confidence field, readout-token PCHI converts 82.2% of originally wrong-yes confidence readouts to $\texttt{no}$, while a joint intervention across upstream confidence-template tokens reduces ECE from 21.9% to 9.2% and damages only 5.1% of originally correct-yes readouts. The readout-token effect also appears on Gemma3-4B, though upstream interventions are weaker and more mask-dependent. These results show that verbalized overconfidence can be selectively reduced through conditionally applied internal intervention, partially decoupling the suppression of unwarranted confidence from the loss of warranted confidence.

cs.LG↗

arXiv2026

Beyond Static Collision Handling: Adaptive Semantic ID Learning for Multimodal Recommendation at Industrial Scale

Modern recommendation systems involve massive catalogs of multimodal items, where scalable item identification must balance compactness, semantic fidelity, and downstream effectiveness. Semantic IDs (SIDs) address this need by representing items as short discrete token sequences derived from multimodal signals, providing a compact interface for retrieval, ranking, and generative recommendation. However, effective SID learning is hindered by collisions, where different items are assigned identical or highly confusable codes. Existing methods mainly rely on improved quantization or fixed overlap regularization, but they do not adaptively distinguish whether an overlap should be suppressed or preserved. We propose AdaSID, an adaptive semantic ID learning framework for recommendation. AdaSID regulates SID overlaps through a two-stage process. First, it relaxes repulsion for observed overlaps when the involved items are semantically compatible, preserving admissible sharing rather than uniformly separating all collisions. Second, it allocates the remaining regulation pressure according to local collision load and training progress, strengthening control in congested regions while gradually rebalancing optimization toward recommendation alignment. This design adaptively decides which overlaps to penalize, how strongly to regulate them, and when to shift the learning focus. Extensive offline and online experiments validate AdaSID. On two public benchmarks, AdaSID improves Recall and NDCG by about 4.5% on average over strong baselines, while improving codebook utilization and SID diversity. In Kuaishou e-commerce, an online A/B test on short-video retrieval covering tens of millions of users achieves statistically significant gains, including a 0.98% GMV improvement, and industrial ranking evaluation shows consistent AUC improvements.

cs.IR↗

arXiv2026

Quantifying and Mitigating Self-Preference Bias of LLM Judges

LLM-as-a-Judge has become a dominant approach in automated evaluation systems, playing critical roles in model alignment, leaderboard construction, quality control, and so on. However, the scalability and trustworthiness of this approach can be substantially distorted by Self-Preference Bias (SPB), which is a directional evaluative deviation in which LLMs systematically favor or disfavor their own generated outputs during evaluation. Existing measurements rely on costly human annotations and conflate generative capability with evaluative stance, and thus are impractical for large-scale deployment in real-world systems. To address this issue, we introduce a fully automated framework to quantifying and mitigating SPB, which constructs equal-quality pairs of responses with negligible quality differences, enabling statistical disentanglement of discriminability from bias propensity without human gold standards. Empirical analysis across 20 mainstream LLMs reveals that advanced capabilities are often uncorrelated, or even negatively correlated, with low SPB. To mitigate this bias, we propose a structured multi-dimensional evaluation strategy grounded in cognitive load decomposition, which reduces SPB by 31.5\% on average.

cs.LG↗

arXiv2026

Stop Treating Collisions Equally: Qualification-Aware Semantic ID Learning for Recommendation at Industrial Scale

Semantic IDs (SIDs) are compact discrete representations derived from multimodal item features, serving as a unified abstraction for ID-based and generative recommendation. However, learning high-quality SIDs remains challenging due to two issues. (1) Collision problem: the quantized token space is prone to collisions, in which semantically distinct items are assigned identical or overly similar SID compositions, resulting in semantic entanglement. (2) Collision-signal heterogeneity: collisions are not uniformly harmful. Some reflect genuine conflicts between semantically unrelated items, while others stem from benign redundancy or systematic data effects. To address these challenges, we propose Qualification-Aware Semantic ID Learning (QuaSID), an end-to-end framework that learns collision-qualified SIDs by selectively repelling qualified conflict pairs and scaling the repulsion strength by collision severity. QuaSID consists of two mechanisms: Hamming-guided Margin Repulsion, which translates low-Hamming SID overlaps into explicit, severity-scaled geometric constraints on the encoder space; and Conflict-Aware Valid Pair Masking, which masks protocol-induced benign overlaps to denoise repulsion supervision. In addition, QuaSID incorporates a dual-tower contrastive objective to inject collaborative signals into tokenization. Experiments on public benchmarks and industrial data validate QuaSID. On public datasets, QuaSID consistently outperforms strong baselines, improving top-K ranking quality by 5.9% over the best baseline while increasing SID composition diversity. In an online A/B test on Kuaishou e-commerce with a 5% traffic split, QuaSID increases ranking GMV-S2 by 2.38% and improves completed orders on cold-start retrieval by up to 6.42%. Finally, we show that the proposed repulsion loss is plug-and-play and enhances a range of SID learning frameworks across datasets.

cs.IR↗

arXiv2025

Watch Closely: Mitigating Object Hallucinations in Large Vision-Language Models with Disentangled Decoding

Large Vision-Language Models (LVLMs) bridge the gap between visual and linguistic modalities, demonstrating strong potential across a variety of domains. However, despite significant progress, LVLMs still suffer from severe hallucination issues in object recognition tasks. These models often fail to accurately identify certain objects, leading to text generation that appears fluent but does not correspond to the visual content, which can have serious consequences in real-world applications. Recently, several methods have been proposed to alleviate LVLM hallucinations, but most focus solely on reducing hallucinations in the language modality. To mitigate hallucinations in both the language and visual modalities, we introduce Hallucination Disentangled Decoding (HDD) method that requires no training. HDD enhances the original image by segmenting it and selecting images that augment the original, while also utilizing a blank image to eliminate language prior hallucinations in both the original and segmented images. This design not only reduces the model's dependence on language priors but also enhances its visual performance. (Code: https://github.com/rickeyhhh/Hallucination-Disentangled-Decoding)

cs.CV↗

arXiv2025

The Safety Challenge of World Models for Embodied AI Agents: A Review

The rapid progress in embodied artificial intelligence has highlighted the necessity for more advanced and integrated models that can perceive, interpret, and predict environmental dynamics. In this context, World Models (WMs) have been introduced to provide embodied agents with the abilities to anticipate future environmental states and fill in knowledge gaps, thereby enhancing agents' ability to plan and execute actions. However, when dealing with embodied agents it is fundamental to ensure that predictions are safe for both the agent and the environment. In this article, we conduct a comprehensive literature review of World Models in the domains of autonomous driving and robotics, with a specific focus on the safety implications of scene and control generation tasks. Our review is complemented by an empirical analysis, wherein we collect and examine predictions from state-of-the-art models, identify and categorize common faults (herein referred to as pathologies), and provide a quantitative evaluation of the results.

cs.AI↗

arXiv2025

Cognitive Duality for Adaptive Web Agents

Web navigation represents a critical and challenging domain for evaluating artificial general intelligence (AGI), demanding complex decision-making within high-entropy, dynamic environments with combinatorially explosive action spaces. Current approaches to building autonomous web agents either focus on offline imitation learning or online exploration, but rarely integrate both paradigms effectively. Inspired by the dual-process theory of human cognition, we derive a principled decomposition into fast System 1 and slow System 2 cognitive processes. This decomposition provides a unifying perspective on existing web agent methodologies, bridging the gap between offline learning of intuitive reactive behaviors and online acquisition of deliberative planning capabilities. We implement this framework in CogniWeb, a modular agent architecture that adaptively toggles between fast intuitive processing and deliberate reasoning based on task complexity. Our evaluation on WebArena demonstrates that CogniWeb achieves competitive performance (43.96% success rate) while maintaining significantly higher efficiency (75% reduction in token usage).

cs.AI↗

arXiv2025

FlashThink: An Early Exit Method For Efficient Reasoning

Large Language Models (LLMs) have shown impressive performance in reasoning tasks. However, LLMs tend to generate excessively long reasoning content, leading to significant computational overhead. Our observations indicate that even on simple problems, LLMs tend to produce unnecessarily lengthy reasoning content, which is against intuitive expectations. Preliminary experiments show that at a certain point during the generation process, the model is already capable of producing the correct solution without completing the full reasoning content. Therefore, we consider that the reasoning process of the model can be exited early to achieve the purpose of efficient reasoning. We introduce a verification model that identifies the exact moment when the model can stop reasoning and still provide the correct answer. Comprehensive experiments on four different benchmarks demonstrate that our proposed method, FlashThink, effectively shortens the reasoning content while preserving the model accuracy. For the Deepseek-R1 and QwQ-32B models, we reduced the length of reasoning content by 77.04% and 77.47%, respectively, without reducing the accuracy.

cs.CL↗

arXiv2025

Augmented Body Communicator: Enhancing daily body expression for people with upper limb limitations through LLM and a robotic arm

Individuals with upper limb movement limitations face challenges in interacting with others. Although robotic arms are currently used primarily for functional tasks, there is considerable potential to explore ways to enhance users' body language capabilities during social interactions. This paper introduces an Augmented Body Communicator system that integrates robotic arms and a large language model. Through the incorporation of kinetic memory, disabled users and their supporters can collaboratively design actions for the robot arm. The LLM system then provides suggestions on the most suitable action based on contextual cues during interactions. The system underwent thorough user testing with six participants who have conditions affecting upper limb mobility. Results indicate that the system improves users' ability to express themselves. Based on our findings, we offer recommendations for developing robotic arms that support disabled individuals with body language capabilities and functional tasks.

cs.HC↗

arXiv2025

Instruct-of-Reflection: Enhancing Large Language Models Iterative Reflection Capabilities via Dynamic-Meta Instruction

Self-reflection for Large Language Models (LLMs) has gained significant attention. Existing approaches involve models iterating and improving their previous responses based on LLMs' internal reflection ability or external feedback. However, recent research has raised doubts about whether intrinsic self-correction without external feedback may even degrade performance. Based on our empirical evidence, we find that current static reflection methods may lead to redundant, drift, and stubborn issues. To mitigate this, we introduce Instruct-of-Reflection (IoRT), a novel and general reflection framework that leverages dynamic-meta instruction to enhance the iterative reflection capability of LLMs. Specifically, we propose the instructor driven by the meta-thoughts and self-consistency classifier, generates various instructions, including refresh, stop, and select, to guide the next reflection iteration. Our experiments demonstrate that IoRT achieves an average improvement of 10.1% over established baselines in mathematical and commonsense reasoning tasks, highlighting its efficacy and applicability.

cs.CL↗

arXiv2025

Dimension Effect of Nanocarbon Precursors on Diamond Synthesis and Transformation Mechanism under Extreme Conditions

Diamond holds significant promise for a wide range of applications due to its exceptional physicochemical properties. Investigating the controlled diamond preparation from nanocarbon precursors with varying dimensions is crucial to optimize the transition conditions and even elucidate the daunting transformation mechanism, however, this remains outstanding challenge despite considerable effort. Herein, we report the imperative dimension effect of nanocarbon precursors on diamond synthesis and physical mechanism under high temperature and high pressure, by comparing the distinct transition processes of zero-dimensional (0D) carbon nanocages (CNCs) and one-dimensional (1D) carbon nanotubes (CNTs) from conventional graphite. The optical and structural characterizations evidently demonstrated that both 0D CNCs and 1D CNTs first undergo collapse and graphitization, followed by the formation of mixed amorphous carbon with embedded diamond clusters, eventually leading to cubic diamond. The plotted pressure-temperature diagram exhibits the unique dimension effect of carbon nanomaterials to diamond transformation. These results provide valuable insights into the phase transition mechanisms of diamond synthesis and its derivatives under extreme conditions.

cond-mat.mtrl-sci↗

arXiv2025

HGTUL: A Hypergraph-based Model For Trajectory User Linking

Trajectory User Linking (TUL), which links anonymous trajectories with users who generate them, plays a crucial role in modeling human mobility. Despite significant advancements in this field, existing studies primarily neglect the high-order inter-trajectory relationships, which represent complex associations among multiple trajectories, manifested through multi-location co-occurrence patterns emerging when trajectories intersect at various Points of Interest (POIs). Furthermore, they also overlook the variable influence of POIs on different trajectories, as well as the user class imbalance problem caused by disparities in user activity levels and check-in frequencies. To address these limitations, we propose a novel HyperGraph-based multi-perspective Trajectory User Linking model (HGTUL). Our model learns trajectory representations from both relational and spatio-temporal perspectives: (1) it captures high-order associations among trajectories by constructing a trajectory hypergraph and leverages a hypergraph attention network to learn the variable impact of POIs on trajectories; (2) it models the spatio-temporal characteristics of trajectories by incorporating their temporal and spatial information into a sequential encoder. Moreover, we design a data balancing method to effectively address the user class imbalance problem and experimentally validate its significance in TUL. Extensive experiments on three real-world datasets demonstrate that HGTUL outperforms state-of-the-art baselines, achieving improvements of 2.57%~20.09% and 5.68%~26.00% in ACC@1 and Macro-F1 metrics, respectively.

cs.LG↗

arXiv2025

Living Bento: Heartbeat-Driven Noodles for Enriched Dining Dynamics

To enhance focused eating and dining socialization, previous Human-Food Interaction research has indicated that external devices can support these dining objectives and immersion. However, methods that focus on the food itself and the diners themselves have remained underdeveloped. In this study, we integrated biofeedback with food, utilizing diners' heart rates as a source of the food's appearance to promote focused eating and dining socialization. By employing LED lights, we dynamically displayed diners' real-time physiological signals through the transparency of the food. Results revealed significant effects on various aspects of dining immersion, such as awareness perceptions, attractiveness, attentiveness to each bite, and emotional bonds with the food. Furthermore, to promote dining socialization, we established a "Sharing Bio-Sync Food" dining system to strengthen emotional connections between diners. Based on these findings, we developed tableware that integrates biofeedback into the culinary experience.

cs.HC↗

arXiv2025

Fixed-budget simulation method for growing cell populations

Investigating the dynamics of growing cell populations is crucial for unraveling key biological mechanisms in living organisms, with many important applications in therapeutics and biochemical engineering. Classical agent-based simulation algorithms are often inefficient for these systems because they track each individual cell, making them impractical for fast (or even exponentially) growing cell populations. To address this challenge, we introduce a novel stochastic simulation approach based on a Feynman-Kac-like representation of the population dynamics. This method, named the Feynman-Kac-inspired Gillespie's Stochastic Simulation Algorithm (FKG-SSA), always employs a fixed number of independently simulated cells for Monte Carlo computation of the system, resulting in a constant computational complexity regardless of the population size. Furthermore, we theoretically show the statistical consistency of the proposed method, indicating its accuracy and reliability. Finally, a couple of biologically relevant numerical examples are presented to illustrate the approach. Overall, the proposed FKG-SSA effectively addresses the challenge of simulating growing cell populations, providing a solid foundation for better analysis of these systems.

q-bio.QM↗

arXiv2024

Bridging the User-side Knowledge Gap in Knowledge-aware Recommendations with Large Language Models

In recent years, knowledge graphs have been integrated into recommender systems as item-side auxiliary information, enhancing recommendation accuracy. However, constructing and integrating structural user-side knowledge remains a significant challenge due to the improper granularity and inherent scarcity of user-side features. Recent advancements in Large Language Models (LLMs) offer the potential to bridge this gap by leveraging their human behavior understanding and extensive real-world knowledge. Nevertheless, integrating LLM-generated information into recommender systems presents challenges, including the risk of noisy information and the need for additional knowledge transfer. In this paper, we propose an LLM-based user-side knowledge inference method alongside a carefully designed recommendation framework to address these challenges. Our approach employs LLMs to infer user interests based on historical behaviors, integrating this user-side information with item-side and collaborative data to construct a hybrid structure: the Collaborative Interest Knowledge Graph (CIKG). Furthermore, we propose a CIKG-based recommendation framework that includes a user interest reconstruction module and a cross-domain contrastive learning module to mitigate potential noise and facilitate knowledge transfer. We conduct extensive experiments on three real-world datasets to validate the effectiveness of our method. Our approach achieves state-of-the-art performance compared to competitive baselines, particularly for users with sparse interactions.

cs.IR↗

arXiv2024

PLPP: Prompt Learning with Perplexity Is Self-Distillation for Vision-Language Models

Pre-trained Vision-Language (VL) models such as CLIP have demonstrated their excellent performance across numerous downstream tasks. A recent method, Context Optimization (CoOp), further improves the performance of VL models on downstream tasks by introducing prompt learning. CoOp optimizes a set of learnable vectors, aka prompt, and freezes the whole CLIP model. However, relying solely on CLIP loss to fine-tune prompts can lead to models that are prone to overfitting on downstream task. To address this issue, we propose a plug-in prompt-regularization method called PLPP (Prompt Learning with PerPlexity), which use perplexity loss to regularize prompt learning. PLPP designs a two-step operation to compute the perplexity for prompts: (a) calculating cosine similarity between the weight of the embedding layer and prompts to get labels, (b) introducing a language model (LM) head that requires no training behind text encoder to output word probability distribution. Meanwhile, we unveil that the essence of PLPP is inherently a form of self-distillation. To further prevent overfitting as well as to reduce the additional computation introduced by PLPP, we turn the hard label to soft label and choose top-$k$ values for calculating the perplexity loss. For accelerating model convergence, we introduce mutual self-distillation learning, that is perplexity and inverted perplexity loss. The experiments conducted on four classification tasks indicate that PLPP exhibits superior performance compared to existing methods.

cs.CL↗