Searcharxiv⌕ Search

arXiv subjects

Zeyu Huang

Publications and source records attributed to Zeyu Huang.

At least 19 recordsLinked to original sources
arXiv2026

Moving Phones, Active Peers: Exploring the Effect of Animated Phones as Facilitators in In-Person Group Discussion

In today's in-person group discussions, smartphones are integrated as intelligent workstations; yet given their co-presence in such face-to-face interactions, whether and how they may enhance people's behavioral engagement with others remains underexplored. This work investigates how animating personal smartphones to move expressively, without compromising regular functions, can transform them into embodied facilitators for co-located group interaction. In the four-stranger small-group discussion setting, guided by Tuckman's group-development theory, we conducted a design workshop (n=12) to identify problematic group-work circumstances and design expressive, attention-efficient animated phone facilitations. Subsequently, we developed AnimaStand, a movement-enabled phone stand that animates phones to deliver facilitation cues according to conversation dynamics. In a between-subjects Wizard-of-Oz study (n=56) with four-stranger group discussions, where everyone's phone was on an AnimaStand, the facilitations re-engaged inactive members, balancing interpersonal dynamics and potentially enhancing relationships, without improving task performance. We finally discuss prospects for more generalizable animated personal device facilitation.

cs.HC↗
arXiv2026

ProcArena: A Multi-Scenario Benchmark for LLMs on Direct and Interactive PL/SQL Development from Natural Language

Large language models (LLMs) have shown strong potential for translating natural-language (NL) requirements into PL/SQL programs, attracting increasing attention from the database community. However, existing NL-to-PL/SQL efforts primarily focus on directly generating PL/SQL from complete NL requirements. In practice, PL/SQL development involves diverse scenarios, such as from-scratch development, code modification, debugging, and optimization, and may require either direct generation or multi-turn interaction. Yet, no comprehensive benchmark evaluates multi-scenario, direct and interactive, and multi-dialect NL-to-PL/SQL development. In this paper, we present ProcArena, an execution-based benchmark covering both Direct and Interactive modes. ProcArena comprises 3,998 executable tasks over 157 databases, spanning nine development subscenarios in PostgreSQL and Oracle. We construct challenging Direct tasks through Iterative Logic Enhancement and scenario-specific adapters, and derive paired Interactive tasks through Knowledge Integration and Requirement Perturbation while preserving executable targets. We further design a controlled Solver-User Simulator protocol that allows models to clarify user intent and inspect the database environment without exposing hidden execution feedback. Evaluating seven language models, we find that the best average scores are only 62.2% and 57.8% in Direct and Interactive, respectively, demonstrating that realistic NL-to-PL/SQL development remains challenging, particularly in interactive settings.

cs.CL↗
arXiv2026

Blending Supervised and Reinforcement Fine-Tuning with Prefix Sampling

Existing LLMs-post-training techniques are broadly categorized into supervised fine-tuning (SFT) and reinforcement fine-tuning (RFT). Each paradigm presents a distinct trade-off: (1) SFT excels at mimicking demonstration data, but can lead to problematic generalization as a form of behavior cloning. (2) Conversely, RFT can significantly enhance a model's performance but is prone to learning unexpected behaviors, and its performance is sensitive to the initial policy. In this paper, we propose a unified view of these methods and introduce Prefix-RFT, a hybrid approach that synergizes learning from both demonstration and exploration. Using mathematical reasoning problems as a test bed, we empirically demonstrate that Prefix-RFT is simple yet effective. Not only does it surpass the performance of standalone SFT and RFT, but it also outperforms parallel mixed-policy RFT methods. Our analysis highlights the complementary nature of SFT and RFT, validating that Prefix-RFT effectively harmonizes them. Further ablation studies confirm the method's robustness to variations in the quality and quantity of demonstration data.

cs.LG↗
arXiv2026

Context Training with Active Information Seeking

Most existing large language models (LLMs) are expensive to adapt after deployment, especially when a task requires newly produced information or niche domain knowledge. Recent work has shown that, by manipulating and optimizing their context, LLMs can be tailored to downstream tasks without updating their weights. However, most existing methods remain closed-loop, relying solely on the model's intrinsic knowledge. In this paper, we equip these context optimizers with Wikipedia search and browser tools for active information seeking. We show that naively adding these tools to a standard sequential context optimization pipeline can actually degrade performance compared to baselines. However, when paired with a search-based training procedure that maintains and prunes multiple candidate contexts, active information seeking delivers consistent and substantial gains. We demonstrate these improvements across diverse domains, including low-resource translation (Flores+), health scenarios (HealthBench), and reasoning-heavy tasks (LiveCodeBench and Humanity's Last Exam). Furthermore, our method proves to be data-efficient, robust across different hyperparameters, and capable of generating effective textual contexts that generalize well across different models.

cs.CL↗
arXiv2026

The Cancellation Hypothesis in Critic-Free RL: From Outcome Rewards to Token Credits

A commonly accepted explanation of critic-free RL for LLMs, based on sequence-level rewards, is that it reinforces successful rollouts with a positive advantage while penalizing failed ones. In contrast, we study critic-free RL from a token-level perspective, revealing the token-flipping phenomenon: positive and negative rollouts exhibit remarkably similar proportions of tokens whose probabilities are boosted or suppressed during RL training. To explain this phenomenon, we further show that a token's change in probability is not fully determined by its own advantage; coupled gradient interactions with other tokens also play a non-negligible role. Specifically, these token coupling effects occur primarily between identical tokens that are both predicted with low confidence. Building upon this analysis, we propose the cancellation hypothesis: as a result of coupling, opposing signals cancel out for tokens shared by positive and negative rollouts, while tokens more specific to successful rollouts receive stronger reinforcement, thereby inducing hidden token-level credit assignment from rollout-level rewards. We support this hypothesis with complementary empirical evidence. (1) Compared with training on only positive rollouts, critic-free RL shifts updates from template and formatting tokens toward reasoning tokens; (2) Tokens boosted by critic-free RL consistently demonstrate higher value than suppressed tokens, regardless of whether they originate from positive or negative rollouts. Guided by this view, we implement two batching interventions to encourage or preserve cancellation in critic-free RL training: query-preserved mini-batching and reward-balanced batching. Despite their simplicity, these interventions improve RLVR training across multiple model scales, supporting cancellation as both an explanatory principle and a practical design criterion for critic-free RL training.

cs.LG↗
arXiv2026

HeartSway: Exploring Biodata as Poetic Traces in Public Space

Human traces scattered across urban landscapes can signify our everyday lives and societal vibrancy in subtle and poetic forms. In this paper, we explore how designed technology can engage biodata as evocative traces. To this end, we present the design, implementation, and evaluation of HeartSway, an interactive hammock that captures a user's heart rate and micro-movements as traces and replays them as an embodied experience for the next visitor. Through a qualitative field study (N=10), we find that HeartSway evokes feelings of connection, curiosity about prior users, and appreciation for shared human vitality. Our work contributes to understanding anonymous archival biodata as a design material for experiential urban traces. We offer design considerations for intimate asynchronous encounters between strangers in public spaces and for reimagining public amenities.

cs.HC↗
arXiv2026

Exploring the Grassroots Understanding and Practices of Collective Memory Co-Contribution in a University Community

Collective memory -- community members' interconnected memories and impressions of the group -- is essential to the community's culture and identity. Its development requires members' continuous participatory contribution and sensemaking. However, existing works mainly adopt a holistic sociological perspective to analyze well-developed collective memory, less focusing on member-level conceptualization of this possession or what the co-contribution practices can be. Therefore, this work alternatively adopts the latter perspective and probes such interpretative and interactional patterns with two mobile systems. With one being a locative narrative and exploration system condensed from existing literature's design frameworks, and the other being a conventional online forum representing current practices, they served as the anchors of observation for our two-week, mixed-methods field study (n=38) on a university campus. A core debate we have identified was to retrospectively contemplate or document the presence as a history for the future. This also subsequently impacted the narrative focuses, expectations of collective memory constituents, and the ways participants seek inspiration from the group. We further extracted design considerations that could better embrace the diverse conceptualizations of collective memory and bond different community members together. Lastly, revisiting and reflecting on our design, we provided extra insights on designing devoted locative narrative experiences for community-driven UGC platforms.

cs.HC↗
arXiv2026

A Unified View of Attention and Residual Sinks: Outlier-Driven Rescaling is Essential for Transformer Training

We investigate the functional role of emergent outliers in large language models, specifically attention sinks (a few tokens that consistently receive large attention logits) and residual sinks (a few fixed dimensions with persistently large activations across most tokens). We hypothesize that these outliers, in conjunction with the corresponding normalizations (\textit{e.g.}, softmax attention and RMSNorm), effectively rescale other non-outlier components. We term this phenomenon \textit{outlier-driven rescaling} and validate this hypothesis across different model architectures and training token counts. This view unifies the origin and mitigation of both sink types. Our main conclusions and observations include: (1) Outliers function jointly with normalization: removing normalization eliminates the corresponding outliers but degrades training stability and performance; directly clipping outliers while retaining normalization leads to degradation, indicating that outlier-driven rescaling contributes to training stability. (2) Outliers serve more as rescale factors rather than contributors, as the final contributions of attention and residual sinks are significantly smaller than those of non-outliers. (3) Outliers can be absorbed into learnable parameters or mitigated via explicit gated rescaling, leading to improved training performance (average gain of 2 points) and enhanced quantization robustness (1.2 points degradation under W4A4 quantization).

cs.CL↗
arXiv2026

AR-LIF: Adaptive reset leaky integrate-and-fire neuron for spiking neural networks

Spiking neural networks offer low energy consumption due to their event-driven nature. Beyond binary spike outputs, their intrinsic floating-point dynamics merit greater attention. Neuronal threshold levels and reset modes critically determine spike count and timing. Hard reset cause information loss, while soft reset apply uniform treatment to neurons. To address these issues, we design an adaptive reset neuron that establishes relationships between inputs, outputs, and reset, while integrating a simple yet effective threshold adjustment strategy. Experimental results demonstrate that our method achieves excellent performance while maintaining lower energy consumption. In particular, it attains state-of-the-art accuracy on Tiny-ImageNet and CIFAR10-DVS. Codes are available at https://github.com/2ephyrus/AR-LIF.

cs.NE↗
arXiv2026

Mixed Reality Scenic Live Streaming for Cultural Heritage: Visual Interactions in a Historic Landscape

Scenic Live Streams (SLS), capturing real-world scenic sites from fixed cameras without streamers, have gained increasing popularity recently. They afford unique real-time lenses into remote sites for viewers' synchronous and collective engagement. Foregrounding its lack of dynamism and interactivity, we aim to maximize the potential of SLS by making it interactive. Namely MRSLS, we overlaid plain SLS with interactive Mixed Reality content that matches the site's geographical structures and local cultural backgrounds. We further highlight the substantial benefit of MRSLS to cultural heritage site interactions, and we demonstrate this design proposal with an MRSLS prototype at a UNESCO-listed heritage site in China. The design process includes an interview (N=6) to pinpoint local scenery and culture, as well as two iterative design studies (N=15, 14). A mixed-methods, between-subjects study (N=43, 37) shows that MRSLS affords immersive scenery appreciation, effective cultural imprints, and vivid shared experience. With its balance between cultural, participatory, and authentic attributes, we appeal for more HCI attention to (MR)SLS as an under-explored design space.

cs.HC↗
arXiv2025

Improving Low-Latency Learning Performance in Spiking Neural Networks via a Change-Perceptive Dendrite-Soma-Axon Neuron

Spiking neurons, the fundamental information processing units of Spiking Neural Networks (SNNs), have the all-or-zero information output form that allows SNNs to be more energy-efficient compared to Artificial Neural Networks (ANNs). However, the hard reset mechanism employed in spiking neurons leads to information degradation due to its uniform handling of diverse membrane potentials. Furthermore, the utilization of overly simplified neuron models that disregard the intricate biological structures inherently impedes the network's capacity to accurately simulate the actual potential transmission process. To address these issues, we propose a dendrite-soma-axon (DSA) neuron employing the soft reset strategy, in conjunction with a potential change-based perception mechanism, culminating in the change-perceptive dendrite-soma-axon (CP-DSA) neuron. Our model contains multiple learnable parameters that expand the representation space of neurons. The change-perceptive (CP) mechanism enables our model to achieve competitive performance in short time steps utilizing the difference information of adjacent time steps. Rigorous theoretical analysis is provided to demonstrate the efficacy of the CP-DSA model and the functional characteristics of its internal parameters. Furthermore, extensive experiments conducted on various datasets substantiate the significant advantages of the CP-DSA model over state-of-the-art approaches.

cs.NE↗
arXiv2025

A Controllable Examination for Long-Context Language Models

Existing frameworks for evaluating long-context language models (LCLM) can be broadly categorized into real-world applications (e.g, document summarization) and synthetic tasks (e.g, needle-in-a-haystack). Despite their utility, both approaches are accompanied by certain intrinsic limitations. Real-world tasks often involve complexity that makes interpretation challenging and suffer from data contamination, whereas synthetic tasks frequently lack meaningful coherence between the target information (needle) and its surrounding context (haystack), undermining their validity as proxies for realistic applications. In response to these challenges, we posit that an ideal long-context evaluation framework should be characterized by three essential features: 1) seamless context 2) controllable setting and 3) sound evaluation. This study introduces $\textbf{LongBioBench}$, a benchmark that utilizes artificially generated biographies as a controlled environment for assessing LCLMs across dimensions of understanding, reasoning, and trustworthiness. Our experimental evaluation, which includes 18 LCLMs in total, demonstrates that most models still exhibit deficiencies in semantic understanding and elementary reasoning over retrieved results and are less trustworthy as context length increases. Our further analysis indicates some design choices employed by existing synthetic benchmarks, such as contextual non-coherence, numerical needles, and the absence of distractors, rendering them vulnerable to test the model's long-context capabilities. To sum up, compared to previous synthetic benchmarks, LongBioBench achieves a better trade-off between mirroring authentic language tasks and maintaining controllability, and is highly interpretable and configurable.

cs.CL↗
arXiv2025

Post-hoc Reward Calibration: A Case Study on Length Bias

Reinforcement Learning from Human Feedback aligns the outputs of Large Language Models with human values and preferences. Central to this process is the reward model (RM), which translates human feedback into training signals for optimising LLM behaviour. However, RMs can develop biases by exploiting spurious correlations in their training data, such as favouring outputs based on length or style rather than true quality. These biases can lead to incorrect output rankings, sub-optimal model evaluations, and the amplification of undesirable behaviours in LLMs alignment. This paper addresses the challenge of correcting such biases without additional data and training, introducing the concept of Post-hoc Reward Calibration. We first propose an intuitive approach to estimate the bias term and, thus, remove it to approximate the underlying true reward. We then extend the approach to a more general and robust form with the Locally Weighted Regression. Focusing on the prevalent length bias, we validate our proposed approaches across three experimental settings, demonstrating consistent improvements: (1) a 3.11 average performance gain across 33 reward models on the RewardBench dataset; (2) enhanced alignment of RM rankings with GPT-4 evaluations and human preferences based on the AlpacaEval benchmark; and (3) improved Length-Controlled win rate of the RLHF process in multiple LLM--RM combinations. Our method is computationally efficient and generalisable to other types of bias and RMs, offering a scalable and robust solution for mitigating biases in LLM alignment. Our code and results are available at https://github.com/ZeroYuHuang/Reward-Calibration.

cs.AI↗
arXiv2025

MVISU-Bench: Benchmarking Mobile Agents for Real-World Tasks by Multi-App, Vague, Interactive, Single-App and Unethical Instructions

Given the significant advances in Large Vision Language Models (LVLMs) in reasoning and visual understanding, mobile agents are rapidly emerging to meet users' automation needs. However, existing evaluation benchmarks are disconnected from the real world and fail to adequately address the diverse and complex requirements of users. From our extensive collection of user questionnaire, we identified five tasks: Multi-App, Vague, Interactive, Single-App, and Unethical Instructions. Around these tasks, we present \textbf{MVISU-Bench}, a bilingual benchmark that includes 404 tasks across 137 mobile applications. Furthermore, we propose Aider, a plug-and-play module that acts as a dynamic prompt prompter to mitigate risks and clarify user intent for mobile agents. Our Aider is easy to integrate into several frameworks and has successfully improved overall success rates by 19.55\% compared to the current state-of-the-art (SOTA) on MVISU-Bench. Specifically, it achieves success rate improvements of 53.52\% and 29.41\% for unethical and interactive instructions, respectively. Through extensive experiments and analysis, we highlight the gap between existing mobile agents and real-world user expectations.

cs.CL↗
arXiv2025

Sign-patterns of Certain Infinite Products

The signs of Fourier coefficients of certain eta quotients are determined by dissecting expansions for theta functions and by applying a general dissection formula for certain classes of quintuple products. A characterization is given for the coefficient sign patterns for \[ \frac{(q^i;q^i)_{\infty}}{(q^p;q^p)_{\infty}} \] for integers \( i > 1 \) and primes \( p > 3 \). The sign analysis for this quotient addresses and extends a conjecture of Bringmann et al. for the coefficients of \( (q^2;q^2)_{\infty}(q^5;q^5)_{\infty}^{-1} \). The sign distribution for additional classes of eta quotients is considered. This addresses multiple conjectures posed by Bringmann et al.

math.NT↗
arXiv2025

Extended Non-Markovian Stochastic Schrödinger Equation with Complex Frequency Modes for General Basis Functions

We introduce an extended formulation of the non-Markovian stochastic Schrödinger equation with complex frequency modes (extended cNMSSE), designed for simulating open quantum system dynamics under arbitrary spectral densities. This extension employs non-exponential basis sets to expand the bath correlation functions, overcoming the reliance of the original cNMSSE on exponential decompositions of the spectral density. Consequently, the extended cNMSSE is applicable to environments beyond those characterized by Debye-type spectral densities. The flexibility to employ general basis functions is particularly advantageous for handling spectral densities with higher-order poles, for which exponential decompositions are often inaccurate or unavailable. The extended cNMSSE is implemented in a pseudo-Fock space using conventional ladder operators and solved efficiently via matrix product state (MPS) techniques, preserving the favorable linear-scaling and wavefunction-based nature of the original method. Benchmark simulations across four representative cases, including discrete spectral density, Ohmic spectral density with exponential and algebraic cutoffs, and critically damped Brownian spectral density, demonstrate excellent agreement with results of hierarchy of forward-backward stochastic Schrödinger equations (HFB-SSE) and extended hierarchical equation of motion (HEOM).

quant-ph↗
arXiv2025

A Closer Look into Mixture-of-Experts in Large Language Models

Mixture-of-experts (MoE) is gaining increasing attention due to its unique properties and remarkable performance, especially for language tasks. By sparsely activating a subset of parameters for each token, MoE architecture could increase the model size without sacrificing computational efficiency, achieving a better trade-off between performance and training costs. However, the underlying mechanism of MoE still lacks further exploration, and its modularization degree remains questionable. In this paper, we make an initial attempt to understand the inner workings of MoE-based large language models. Concretely, we comprehensively study the parametric and behavioral features of three popular MoE-based models and reveal some intriguing observations, including 1) Neurons act like fine-grained experts; 2) The router of MoE usually selects experts with larger output norms; 3) The expert diversity increases as the layer increases, while the last layer is an outlier, which is further validated by an initial experiment. Based on the observations, we also provide suggestions for a broad spectrum of MoE practitioners, such as router design and expert allocation. We hope this work could shed light on future research on the MoE framework and other modular architectures. Code is available at https://github.com/kamanphoebe/Look-into-MoEs.

cs.CL↗