arXiv subjects

Yi Zheng

Publications and source records attributed to Yi Zheng.

At least 19 recordsLinked to original sources

arXiv2026

RTP-LLM: Building an Evolvable and Reliable LLM Inference Engine at Alibaba Scale

Large language models and inference techniques evolve rapidly, making model enablement a recurring production task rather than a one-time integration effort. At Alibaba, a shared inference infrastructure must keep pace with new models and serving capabilities while remaining reliable across heterogeneous production environments. To reconcile rapid evolution with production reliability, we present RTP-LLM, an open-source production inference engine developed and deployed at Alibaba. Its architecture separates model-specific execution from shared request processing, state management, and resource coordination. Explicit interfaces localize compatible model changes while allowing models and deployment configurations to reuse common serving mechanisms. RTP-LLM has evolved from an exploratory system into a shared production engine that supports $\sim$100 models and processes hundreds of trillions of tokens per day. We describe how its architecture and production practices accommodate model evolution and contain the impact of failures, and share lessons for building evolvable and reliable inference engines at enterprise scale.

cs.OS↗

arXiv2026

Beyond Modeling Order: Dual-Path Spatiotemporal Representation Learning for Cross-Subject EEG Emotion Recognition

Cross-subject generalization remains a major challenge for EEG-based emotion recognition due to pronounced inter-subject variability in neural activity and emotion-related spatiotemporal patterns. Existing methods mainly alleviate inter subject differences through distribution alignment or contrastive learning, while paying limited attention to representation bias associated with fixed spatiotemporal modeling orders. We hypothesize that high-level emotion representations of the same EEG signal should remain consistent across different spatiotemporal modeling paths. Based on this hypothesis, we propose a Dual-Path Spatiotemporal Contrastive Learning (DP-STCL) framework for cross-subject EEG emotion recognition. DP-STCL employs two complementary encoders with independent parameters, namely a Temporal-to-Spatial (TS) path and a Spatial to-Temporal (ST) path, to model EEG spatiotemporal information under different integration orders. During representation learning, the same emotional stimulus is used to establish cross subject associations, while three complementary relations jointly constrain representation variations across subjects and modeling paths. This promotes stable and discriminative stimulus-related emotion representations. To further handle limited labeled target subject data, we introduce a LoRA-based few-shot adaptation strategy with low parameter overhead. Extensive cross-subject experiments on SEED, SEED-IV, SEED-V, and FACED show that DP-STCL consistently outperforms the compared methods, demonstrating its effectiveness in improving cross-subject EEG emotion recognition.

cs.LG↗

arXiv2026

A geometric criterion for the dimension of Morgan-Scott spline spaces $S_{2r}^r(Δ_{\mathrm{MS}})$

We establish a geometric dimension criterion for spline spaces on nondegenerate standard Morgan-Scott triangulations with four distinct edge directions at each interior vertex. For every integer $r\ge1$, concurrence of the three auxiliary lines joining corresponding interior and boundary vertices implies that $\dim S_{2r}^r(Δ_{\mathrm{MS}})$ exceeds the Schumaker lower bound by one. The proof constructs a quadratic form through the six dual spoke points and uses its $r$th power to define a nonzero functional annihilating the image of the smoothing-cofactor map. Diener's upper bound then gives the dimension formula. We also give a matrix representation of this functional and an explicit example in the quadratic case. Combined with the known necessary conditions and the negative-branch exclusion of Deng, Feng, and Kozak, the result yields a necessary and sufficient concurrence criterion for $r>2$.

math.NA↗

arXiv2026

Model-independent parameters from DESI DR1 LRG2 with an explicitly varying power-spectrum shape

Distinguishing between dark energy and modified gravity as the origin of the accelerated expansion requires the cosmic distances and the growth of structure to be measured in a model-independent way. Such measurements are difficult to perform while the primordial shape parameters are varied at the same time, and those parameters are therefore usually held fixed at the values determined by cosmic microwave background experiments. In this work, we present a methodology that constrains the shape parameters simultaneously with the model-independent distance and growth parameters, using an emulator built upon a hybrid model in which perturbative predictions are supplemented by templates measured from $N$-body simulations. {Because these templates are measured rather than expanded, the model carries non-linear corrections beyond the order at which its perturbative part is truncated, so that its accuracy on small scales is set by the simulations rather than by the reach of the expansion.} A Gaussian process emulator supplies these templates rapidly across the space of scale-dependent parameters, which makes the joint analysis computationally feasible, and the emulator is trained and validated over a range spanning $5σ$ of the Planck constraints. Applying the method to the LRG2 sample of DESI DR1, we obtain measurements of the Alcock--Paczynski parameters $α_\perp$ and $α_{||}$ and of the growth rate $fσ_8$. {With the tight Planck prior adopted here the constraints differ from those obtained with the shape parameters held fixed by less than a twentieth of their own uncertainties, so that the choice of template contributes at that level to the error budget of the measurement. Projected onto flat $Λ$CDM after the fit, the three measurements are mutually consistent, with $χ^2=0.7$ for two degrees of freedom, and require $h=0.724\pm0.017$.

astro-ph.CO↗

arXiv2026

EmoTrace-Bench: Diagnosing Emotional Expression in Text-to-Video Generation through Controlled Cues and Intensity

Text-to-video (T2V) models are increasingly used to generate videos with targeted emotional expressions, yet existing evaluation methods mainly rely on holistic video-text alignment and provide limited insight into whether failures arise from target-emotion expression, intensity control, or specific affective cues. We introduce EmoTrace-Bench, a fine-grained diagnostic benchmark that controls emotion intensity and selected affective cues in prompts and evaluates generated videos at multiple levels. EmoTrace-Bench contains 448 controlled prompts covering eight emotions, two intensity levels, and diverse cue conditions. Using 12 T2V models, we generate 5,376 videos and evaluate target-emotion presence, perceived intensity, and ten audiovisual dimensions of emotional expression with an automated evaluator calibrated and validated against human annotations on a subset of the data. Results reveal substantial differences across models and prompt designs. Prompts that explicitly describe a character's emotional reaction are followed more reliably than prompts that describe only events or scenes, whereas adding more affective cues or using more explicit cinematic audiovisual language does not consistently improve emotional generation. These results illustrate how aggregate alignment scores may overlook localized differences in target-emotion expression, intensity control, and cue realization. EmoTrace-Bench provides a traceable and interpretable framework for diagnosing emotional video generation without modifying existing generation pipelines, and we release the prompts and the automated evaluation rules.

cs.CV↗

arXiv2026

Multimodal Image-based Robotic Assistance and Guidance Environment (MIRAGE): A Novel Unified Framework for Surgical Navigation and Visualization

This paper presents the computational architecture and quantitative preclinical evaluation of MIRAGE, a multimodal image-guided robotic surgery framework for real-time visualization and navigation. It integrates stereovision scene reconstruction, augmented-reality (AR) video overlay, quantified heads-up display (HUD) navigation, and intraoperative ultrasound within the da Vinci console's TilePro display (Intuitive Surgical). Additional modules include optical/electromagnetic tracker registration, stereo calibration, fiducial rigid registration with RMSE feedback, and 3D point-cloud reconstruction. Video I/O and rendering run in a C++ backend, while a Python GUI coordinates calibration, registration, reconstruction, visualization, and user interaction through an in-process zero-copy binding layer. The framework formalizes spatial transforms, module states, and synchronization of heterogeneous video and tracking streams. Preoperative calibration uses synchronized checkerboard image pairs and optimization, with estimated intrinsics, extrinsics, and rectification reused intraoperatively. MIRAGE was validated in phantom experiments across da Vinci S, Si, Xi, and SP platforms. Across clinically relevant working distances, mean 3D reprojection error was 0.6mm, with best 0.1mm RMSE at 100mm on Xi. Results support MIRAGE as a preclinical technical-validation platform for accurate, responsive, and extensible multimodal surgical guidance.

cs.RO↗

arXiv2026

Atomic Motion Coordinate for Language-Steerable and Force-Responsive Manipulation

Can changing only the language instruction redirect a VLA policy's end effector, or does the visually driven motion prior dominate? We present Atomic Motion Coordinate, a geometry-grounded coordinate for steerable and force-responsive manipulation. Each arm owns thirteen signed translation, rotation, and hold atoms grounded from text and forward kinematics with vision withheld, and the coordinate is injected into every action-expert block via weighted codebook alignment. Contact history modulates the same coordinate through a bounded spherical residual that is recomputed from a fixed nominal latent to regenerate only the unexecuted horizon suffix. Across 7,520 offline horizon interventions, opposite-atom separation reaches 92.5/83.1% (single/dual) versus 39.1/24.0% for LA4VLA-style. Across 50 real-robot trials per task, AMC raises OOD fruit progress from 60.5% to 87.8%; force adaptation raises Plug/Vase from 59.0/71.5% to 78.5/75.2%.

cs.RO↗

arXiv2026

ACPruner: Visual Token Pruning as Biased Attention Coverage Maximization in LVLMs

Large Vision-Language Models (LVLMs) face significant computational inefficiencies caused by the large number of visual tokens. Existing visual token pruning methods mainly focus on either retaining individually important tokens or selecting mutually diverse ones. In this work, we revisit visual token pruning from a coverage perspective and formulate it as a biased attention coverage maximization problem. The key idea is to select a compact token subset whose encoder-side outgoing attention can jointly cover the image while assigning higher coverage priority to more informative regions. From this perspective, we propose ACPruner, a training-free visual token pruning framework for efficient LVLM inference. ACPruner first estimates token importance by combining intra-modal saliency and inter-modal relevance, then derives token-wise coverage from attention patterns within the vision encoder, and finally performs greedy selection to maximize the proposed coverage objective. Extensive experiments across multiple LVLM backbones, including LLaVA-1.5-7B/13B, LLaVA-NeXT-7B/13B, Qwen2.5-VL-7B, and LLaVA-OneVision-7B, show that ACPruner consistently achieves strong performance retention while delivering substantial end-to-end inference speedups.

cs.CV↗

arXiv2026

ArenaFlow: From Trajectory Ranking to Hierarchical Credit Propagation for Open-Ended Agent RL

Reinforcement learning has substantially improved large language model (LLM) agents in verifiable domains, but remains difficult to apply to open-ended agent tasks, where solutions are diverse and reliable scalar rewards are hard to obtain. Recent pairwise evaluation methods alleviate reward discrimination collapse by replacing pointwise scoring with relative preferences. However, they still compress rich comparative feedback into a single trajectory-level reward, obscuring decisive intermediate steps and preventing successful behaviors from being consolidated into reusable skills. We propose ArenaFlow, a hierarchical credit propagation framework for open-ended agent reinforcement learning. ArenaFlow leverages tournament-based relative ranking to derive trajectory-level reward signals. Each comparison is further equipped with structured reflective evaluation, which reveals three types of supervision: pivotal success steps, reusable strategy skills, and usage attribution of retrieved skills. At the step level, ArenaFlow propagates trajectory-level advantages to high-confidence pivotal steps according to tournament survival depth, enabling more targeted optimization of local reasoning behaviors. At the skill level, ArenaFlow estimates skill utility from group-level usage attribution and maintains a global skill memory through utility-aware updating, pruning, and retrieval. The resulting high-utility skills further serve as policy priors for future exploration. Extensive experiments validate ArenaFlow's effectiveness on open-ended agent tasks.

cs.CL↗

arXiv2026

From Accuracy to Auditability: A Survey of Determinism in Financial AI Systems

Deploying machine learning in regulated financial environments -- credit risk, fraud detection, and anti-money laundering -- exposes critical vulnerabilities in algorithmic reproducibility. While early financial ML addressed statistical challenges such as backtest overfitting, deep neural networks and Generative AI have introduced mechanical nondeterminism rooted in hardware and architecture. This survey provides a systems perspective on reproducibility failures across three modalities now dominant in financial AI: tabular models (post-hoc explanation variance), graph networks (stochastic sampling and temporal asynchrony), and LLM-based agentic workflows (batch-dependent divergence and trajectory drift). We supplement the literature analysis with first-party experiments on public financial datasets -- quantifying explanation rank instability in credit scoring, prediction flip rates in GNN-based fraud detection, and tensor-parallel-induced output divergence in LLM entity extraction. We propose a layered evaluation framework linking modality-specific metrics (RBO, D_cos, TDI, PSD) to audit readiness, and report where these measures overlap rather than complement one another.

cs.AI↗

arXiv2026

Beyond Electrons: A Theoretical Framework for Near-Field Radiative Thermal Computing and Neural-Network-Inspired Processing

Near field radiative heat transfer provides a route for information processing in which thermal radiation, rather than charge transport, serves as the physical carrier of signals. Here, we propose and theoretically analyze a programmable near field radiative thermal computing framework in which radiative coupling, phase change nonlinearity, and thermal state memory are mapped onto neural network inspired operations. The framework is constructed from near field radiative thermal diodes, transistors, and multi terminal logic units separated by nanoscale gaps. Radiative heat flux represents the propagated thermal information, while geometry and material dependent radiative coupling provides physically constrained weighting, and the temperature dependent optical response of phase-change materials enables nonlinear modulation and logic state control. Based on these primitives, we formulate a radiative thermal convolutional network for spatial information processing and a radiative thermal recurrent network for history dependent computation. The recurrent response is associated with radiative feedback, thermal relaxation, and phase change hysteresis, with VO2 providing history dependent short term memory and GST offering a possible route toward non volatile phase storage. We further distinguish the physical radiative networks from a separate software based inverse identification study, in which recurrent machine-learning models are trained on simulated near field heat flux temperature characteristics to recover structural parameters. By establishing a bottom up connection between fluctuational electrodynamics, radiative thermal logic, programmable thermal states, and neural network inspired computation, this study provides a physically grounded basis for exploring non contact thermal information processing at the near field limit.

cond-mat.mes-hall↗

arXiv2026

EmotionDialogCN: A Spontaneous Multimodal Dataset for Mandarin Emotional Dialogue

Face-to-face audiovisual interaction is central to human communication, conveying rich emotional and social cues. However, existing multimodal dialogue datasets remain limited by inadequate emotion annotations, poor emotional diversity, and small scale. We introduce EmotionDialogCN, a large-scale audiovisual-emotional dataset designed to capture authentic face-to-face communication. It contains 21,880 dialogue sessions performed by 119 professional actors across 20 everyday scenarios, covering 18 emotion categories with over 400 hours of recordings, the largest and most comprehensive dataset of its kind. A novel data collection framework minimizes equipment interference, enabling natural and nuanced emotional expressions. EmotionDialogCN achieves an emotion distribution deviation of 0.64 from real human emotion statistics (versus 5.65 for prior datasets) and consistent subject framing (52-59% frame occupancy). Together, these properties translate into stable unimodal and multimodal performance across acoustic, lexical, and visual modalities, with fusion results further underscoring strong multimodal alignment and cross-modal complementarity.

cs.CV↗

arXiv2026

Harnessing thermo-optic dynamics for frequency-agile soliton microcombs

Dissipative Kerr soliton microcombs enable compact and scalable frequency comb sources for precision metrology, spectroscopy, communications and coherent LiDAR, where broad and reliable frequency tuning is essential. Thermo-optic response can support thermal locking during soliton operation, enabling resonance tracking and thereby extending the tuning range, albeit modestly. However, it also induces pronounced thermal instability during soliton initiation, hindering reliable access to this extended operating regime and limiting practical deployment in applications requiring frequency agility. Here we show that strong mode coupling reshapes the effective detuning trajectory governing soliton formation, establishing a distinct operating regime in which thermo-optic response is significantly reinforced and constructively harnessed. In this regime, soliton formation proceeds without the thermal instability inherent to conventional operation, enabling robust soliton generation in material platforms previously limited by strong thermal effects. Importantly, the enhanced thermo-optic response strengthens thermal locking during soliton operation, enabling more effective resonance tracking and substantially extending the tuning range. Leveraging this regime in AlGaAs-on-insulator multimode microresonators, we demonstrate soliton generation with a tuning range approaching 100 GHz at a pump power of 32 mW. The same mechanism further enables frequency-agile operation through direct pump-frequency tuning without auxiliary stabilization, allowing massively parallel chirped comb generation with more than 90 channels exhibiting frequency excursions exceeding 10 GHz. These results establish a general operating principle for transforming thermo-optic effects from a limiting factor into an active resource, enabling robust and frequency-agile integrated soliton microcombs.

physics.optics↗

arXiv2026

Fractals in rate-induced tipping

When parameters of a dynamical system change sufficiently fast, critical transitions can take place even in the absence of bifurcations. This phenomenon is known as rate-induced tipping and has been reported in a variety of systems, from simple ordinary differential equations and maps to mathematical models in climate sciences and ecology. In most examples, the transition happens at a critical rate of parameter change, a rate-induced tipping point, and is associated with a simple unstable orbit (edge state). In this work, we show how this simple picture changes when non-attracting fractal sets exist in the autonomous system, a ubiquitous situation in non-linear dynamics. We show that these fractals in phase space induce fractals in parameter space, which control the rates and parameter changes that result in tipping. We explain how such rate-induced fractals appear and how the fractal dimensions of the different sets are related to each other. We illustrate our general theory in three paradigmatic systems: a piecewise linear one-dimensional map, the two-dimensional Hénon map, and a forced pendulum.

nlin.CD↗

arXiv2026

AdaDINO: Context-Adaptive DINO-Distilled Vision Foundation Models for Efficient Open-Vocabulary Edge Inference

Always-on contextual AI runs language-aligned vision foundation models (VFMs) on edge devices, where the on-device model is the dominant continuous compute cost under strict latency and power limits. Due to an observed low-frequency shift in scene context and its relevant vocabulary, we present AdaDINO, an adaptive framework that makes on-device VFM inference efficient by matching execution to the current scene and task. We build on a known phenomenon, that the accuracy drop of shrinking model sizes depends on the task, and turn it into task-level adaptive execution. AdaDINO integrates neural architecture search (NAS) into a language-aligned VFM backbone distilled from DINOv2, training a single family of subnets for efficient execution during runtime. A multimodal large language model (LLM) on the cloud, invoked at low frequency, refines the candidate class set from scene context, while a learned selector activates the least-cost subnet predicted to retain a target fraction of accuracy. With the backbone and semantic pipeline held fixed, learned selection alone reduces average compute by $37\%$ over the best fixed subnet at equal segmentation accuracy. Across zero-shot classification and open-vocabulary segmentation, AdaDINO establishes a strong accuracy-efficiency frontier, improving over evaluated models of comparable sizes by up to $7.9\%$ in acc@1 on IN1K and $5.2\%$ mIoU on ADE20K, and reducing average FLOPs by up to $74.9\%$ at similar accuracy.

cs.CV↗

arXiv2026

Toward Annotation-Efficient Continuous Emotion Arousal Quantification via Group-Level EEG Dynamic Neural Synchrony

Continuous emotional arousal quantification remains bottlenecked by time-consuming and labor-intensive manual annotation. This work investigates group-level EEG dynamic neural synchrony (DNS) as a principled signal for continuous arousal quantification that bypasses per-subject manual labeling. Using Correlated Component Analysis (CorrCA) with sliding-window computation across four EEG datasets spanning 142 subjects and over 207 hours, we systematically evaluate DNS as a group-level marker for emotional arousal dynamics. Three key findings emerge. First, DNS exhibits significant emotion information from valence-dependent differences (all p<0.003), with positive emotions eliciting higher synchrony. Second, DNS correlates more strongly with the first-order derivative of arousal than with raw arousal values, revealing that neural synchrony captures the rate of emotional change rather than static intensity. Third, we provide the first systematic characterization of how DNS-arousal coupling depends on key methodological choices, finding that moderate windows (10-30 s), positive lags (0-10 steps), and First-order Difference feature of EEG from the dominant CorrCA component yield consistently strong coupling. Subject-split replication and block permutation tests confirm these associations are not statistical artifacts. Our findings establish DNS as an empirically validated group-level marker toward annotation-efficient continuous emotional arousal quantification.

cs.HC↗

arXiv2026

CRISP: Pre-LLM Yet Text-Driven Visual Token Pruning for Efficient LVLM Inference

Large Vision-Language Models (LVLMs) typically require processing hundreds to thousands of visual tokens, leading to substantial inference overhead. Existing visual token pruning methods either operate before the LLM using text-agnostic heuristics or prune inside the LLM at the cost of efficiency and noisy cross-modal attention. To address these limitations, we propose CRISP, a pre-LLM yet text-driven visual token pruning framework that preserves both instruction-relevant evidence and essential scene context. CRISP works in a two-stage pipeline: Stage 1 first identifies text-aligned visual tokens, and Stage 2 enhances contextual completeness through semantic diversity. Extensive experiments on LLaVA-1.5 and LLaVA-NeXT demonstrate that CRISP achieves superior performance retention under aggressive pruning ratios, maintaining up to 99.5% accuracy while reducing inference cost and latency by more than 2 times. CRISP serves as a practical solution for efficient LVLM inference, especially in resource-constrained scenarios.

cs.CV↗

arXiv2026

ReProAgent: Tool-Augmented Multi-Stage Agentic Generation of Bug Reproduction Tests from Issue Reports

Reproduction tests help developers confirm reported issues and provide executable feedback for issue resolution, yet issue reports in open-source projects rarely include such tests. Recent studies have explored generating issue reproduction tests from issue reports with large language models, but existing approaches largely rely on prompt-based pipelines that retrieve textual context and generate tests. This limits their ability to understand how reported issues behave in repository-scale codebases and to flexibly organize the construction of reproduction tests. In this paper, we propose ReProAgent, a multi-stage agent framework for reproduction test generation from issue reports. ReProAgent decomposes the task into four agent stages: bug localization, root cause analysis, test planning, and test generation. To support these stages, ReProAgent integrates task-specific tools for task decomposition and reflection, context retrieval from both textual sources and repository graphs, and runtime interaction with the execution environment. Experiments on SWT-bench-lite and SWT-bench-verified show that ReProAgent successfully reproduces 58.43% and 70.30% of issues, outperforming all baselines, with an average cost of $0.14 per instance. For example, when equipped with GPT-5-mini, ReProAgent exceeds OpenHands with the same backbone by 20.43 and 7.90 percentage points, respectively. ReProAgent also generalizes across multiple backbone LLMs and improves downstream issue resolution performance when integrated with existing repair approaches.

cs.SE↗