arXiv subjects

Yan Ma

Publications and source records attributed to Yan Ma.

At least 19 recordsLinked to original sources

arXiv2026

AutoVend: Benchmarking LLM Agents on Long-Horizon Autonomous Vending Operations

Large language model (LLM) agents are extending agent-mediated commerce on the Web from shopping for consumers to operating businesses for sellers. In online-to-offline (O2O) commerce, decisions are issued through Web-based platforms but realized through costly physical execution. Evaluating such agents requires going beyond individual transactions to ask whether decisions made online turn into sustained economic returns. We introduce AutoVend, a benchmark in which one LLM agent manages 20 vending machines across five site types through a Web-based operating interface for 180 simulated days. Its data-grounded simulator combines a shared warehouse, procurement lead times, paid replenishment, inventory expiration, and recurring operating costs. We evaluate five LLM agents against a rule-based inventory-control baseline and diagnose their site coordination, execution, and performance over time. Four of five evaluated LLM agents fall short of this baseline, and returns per dollar of inference cost rank models differently from returns alone. Operational analyses show that gains in individual metrics do not consistently yield stronger economic growth, as inventory and execution costs erode these gains. Trajectory analysis further reveals difficulties in grounding feasible actions, managing inventory commitments, and converting feedback into effective policy revisions.

cs.AI↗

arXiv2026

Highly integrated quantum key distribution transmitter enabled by silicon photonics

Quantum key distribution (QKD) provides information-theoretic security independent of computational assumptions, yet the bulk and cost of current systems hinder large-scale deployment. Although integrated photonic technologies have enabled highly integrated QKD chips, practical QKD transmitters are still predominantly implemented as rack-mounted systems. Here, we demonstrate a highly integrated standalone QKD transmitter that integrates all essential functionalities required for practical QKD operation within a compact platform built around a silicon photonic encoding chip. The transmitter occupies only $167 \times 56 \times 21~\mathrm{mm}^3$ ($\sim 0.2~\mathrm{dm}^3$), comparable in size to a half-height, half-width PCIe card and more than 30 times smaller in volume than a conventional 1U rack-mounted system. Paired with a conventional discrete-component receiver, it achieves a secure key rate of 219.1 kbps over 51.3 km of standard single-mode fiber, with performance comparable to that of conventional discrete-component implementations. This work bridges the gap between photonic chip integration and deployable QKD hardware, marking an important step toward transitioning QKD transmitters from conventional rack-mounted equipment to compact board-level platforms.

quant-ph↗

arXiv2026

Query-Adaptive Neuron Steering for LLMs

Effectively eliciting the capabilities of large language models (LLMs) at inference time remains an important challenge. Existing studies address this challenge through context engineering, which shapes user queries, or fixed steering, which directly intervenes in model computation. However, predefined contexts and fixed intervention strategies may not adequately accommodate the diverse knowledge and reasoning requirements of individual queries. This raises a natural question: Can internal steering in LLMs be made query-adaptive? In this work, we investigate whether query-induced internal activations can guide adaptive steering in LLMs. Our pilot studies reveal subject- and layer-dependent neuron activation patterns and show that amplifying subject-relevant neurons can improve model performance. Based on these findings, we propose Q-NeuS, a query-adaptive neuron steering framework that dynamically determines which neurons to modulate and by how much for each input. Specifically, Q-NeuS constructs a compact subject-guided candidate neuron space and employs a lightweight Query-Adaptive Gate to generate neuron-wise modulation coefficients. To learn effective intervention strategies, we introduce Neu-GRPO, which adapts group-relative optimization to train the Query-Adaptive Gate using the relative performance of multiple neuron intervention configurations. The base LLM remains frozen throughout training and inference, and no task or subject labels are required at inference time. Extensive experiments across multiple LLMs and benchmarks demonstrate that Q-NeuS consistently improves model performance and exhibits broad applicability across model scales and strong cross-benchmark transferability.

cs.LG↗

arXiv2026

MOMAT: Mixture of Multiple Atlases for Low-Power Jailbreak Defense of Quantized LLMs

Quantized large language models are increasingly deployed on edge devices for their low latency and energy efficiency. However, model quantization weakens alignment safeguards, leaving qLLMs (quantized large language models) highly vulnerable to jailbreak attacks. To address this challenge, we present MOMAT (Mixture of Multiple Atlases), a hardware-enhanced safety framework that combines structured knowledge retrieval with low-power defense acceleration. Each atlas represents a semantic cluster of harmful or benign sample sets and policy templates, enabling domain-localized Retrieval-Augmented Generation guarding that mitigates the curse of dimensionality and the resulting semantic sparsity problem in large, heterogeneous safety databases. MOMAT retrieves top-$k$ similarity features from all atlases for each prompt and evaluates them using a lightweight MoE (Mixture of Experts) detector, while a CiM (Compute-in-Memory)-accelerated similarity engine performs fast, low-power atlas-local retrieval. MOMAT's CiM-based retrieval accelerates a 100-query batch from 15,052.44 ms to 3,207.21 ns (a $4.69 \times 10^6\times$ speedup) and reduces energy from $8.1 \times 10^7$ $μ$J to 3.32 $μ$J, yielding an approximately $2.5 \times 10^5\times$ energy reduction over DRAM-based (Raspberry Pi) baselines. Red-team evaluations across standard benchmarks show that MOMAT matches the defense performance of state-of-the-art methods while avoiding benign overkill and providing substantial efficiency gains, demonstrating that CiM-based modular defenses can make edge-deployed qLLMs both safer and more energy-efficient. We will release the full 223.2k-sample dataset to foster future research.

cs.CR↗

arXiv2026

Interior curvature estimates for graphical curvature quotient equations

We prove interior curvature estimates for admissible graphical solutions of curvature quotient equations in the range $3\leq k<n$ on the full Gårding cone $Γ_k$. No convexity, semiconvexity, or $Γ_{k+1}$-admissibility assumption is required. The proof combines a new quantitative compression inequality, a two-surface comparison and doubling argument, and a Pogorelov estimate.

math.AP↗

arXiv2026

VidaForge: Open Research Infrastructure for Video Pretraining Data Recipes

Video foundation models increasingly rely on large-scale pretraining data, yet the end-to-end data pipelines behind them remain largely closed and difficult to inspect or reuse. Researchers seeking to understand how video data recipes affect model pretraining often need to build substantial infrastructure before testing even a focused hypothesis. We present VIDAFORGE, an open research infrastructure that represents a video data recipe as an executable five-stage workflow from raw videos to training datasets. A decision in this workflow can be varied to construct alternative datasets while preserving how every sample was produced. To demon strate this research workflow, we compare data recipes with different coverage and quality in early from-scratch pretraining of Wan 2.1 and V-JEPA 2.1. Across both learning objectives, the broader-coverage recipe achieves the highest downstream benchmark scores, while loss-based evaluation favors different recipes. This study demonstrates how VidaForge connects data-recipe choices to downstream model performance. We further release VIDAFORGE-3M, containing 3.14 million scene level clips totaling 6,475 hours, with fine-grained annotations and curation signals for video data-recipe research.

cs.CV↗

arXiv2026

The role of disconnections in redox-induced phase transformation of metal oxides

Detailed atomic-level understanding of the phase transformation between magnetite (${\rm Fe_3 O_4}$) and haematite (${\rm Fe_2 O_3}$) is lacking, despite widespread interest in redox reactions of metal oxide systems. While the magnetite-to-haematite transformation entails the addition of oxygen to the system and thus a net diffusive flux, haematite formations have been reported to grow along well-defined habits on the close-packed $\{0001\}_{\rm Fe_2 O_3}//\{111\}_{\rm Fe_3 O_4}$ planes. The propagation of the phase interface in the bulk is thus thought to maintain the oxygen sublattice fixed up to a shear, and to require long-range diffusion of iron ions. In this letter we propose interfacial steps with dislocation character, i.e. disconnections, as the elementary defects propagating the transformation. We use the topological model of interfacial defects to predict three disconnection modes for the system, and we employ scanning transmission electron microscopy to ascertain their presence in a haematite/magnetite interface of a partially oxidised magnetite powder. Using atomistic simulations, we determine the equilibrium structures of the disconnections, and study their motion and how they accomplish the transformation, obtaining excellent agreement with atomic-resolution experimental observations.

cond-mat.mtrl-sci↗

arXiv2026

AnchorScore: A CLIP-Based Diagnostic of MLLM Annotation Difficulty

Multimodal large language models (MLLMs) are widely used for automated annotation, yet their per-class accuracy varies widely (e.g., 12%-98% across the 13 classes of three classroom sub-datasets) and is expensive to measure: evaluating one 27B MLLM on 5,416 validation images takes roughly 14 hours, whereas a frozen-CLIP pass over the same images completes in about 3 minutes. A low-cost signal for ranking classes by expected MLLM annotation difficulty a priori remains underexplored. Building on the AnchorProxy construct (per-class zero-shot CLIP accuracy) introduced in the companion study, this paper systematically evaluates its full-frame formulation, termed AnchorScore here, as an a priori diagnostic that flags the classes MLLMs are least likely to annotate reliably. On classroom behavior data (SCB5, 13 classes, 6 MLLMs), AnchorScore correlates with per-class MLLM accuracy (Spearman rho = 0.769, p = 0.002, n = 13). None of the alternative difficulty predictors (DINOv2, ResNet-50, SigLIP, or MLLM self-verbalized uncertainty) showed a significant class-level correlation at n = 13. A cross-model consensus control suggests AnchorScore primarily captures a shared class-difficulty factor rather than a CLIP-specific signal. An independent replication on Stanford40 Actions yields a nearly identical effect (rho = 0.817, p < 0.001); the association is strongest on activity-recognition data and attenuates on medical and satellite imagery. Three practical applications follow: a deployable hybrid CLIP/MLLM routing strategy (predicted-class routing: up to +23 pp over CLIP-only at roughly 44% fewer MLLM calls), prompt disambiguation on hard classes (exploratory), and review-priority prediction for human verification. AnchorScore does not estimate exact MLLM accuracy; it provides a low-cost ranking signal that directs expensive MLLM evaluation to the classes where it is most informative.

cs.CV↗

arXiv2026

SleepWalking: Privileged Representation Shaping for End-to-End Blind Locomotion in Legged Robots

Partially observable locomotion requires a policy to act when task-relevant properties of the robot--environment state are not fully specified by instantaneous observations. Existing approaches often address this challenge by explicitly estimating missing physical variables or processing extended observation histories through structured architectures. We take a different view: partial observability is fundamentally an information-retention problem. The decisive question is not how task-relevant information enters the network, but whether the policy's internal state retains it. Guided by this perspective, we propose SleepWalking for Robot Locomotion (SWAQ), a one-stage end-to-end framework that uses next-step privileged physical reconstruction to shape what a recurrent history representation retains during policy learning, while the deployed actor uses only a direct history-to-action pathway. Under aligned training settings, SWAQ achieves a 15.0\% higher peak mean terrain level than DWAQ, the strongest non-exteroceptive baseline, while using 44.4\% fewer inference MACs per control step. Layerwise probes further show that information associated with the reconstructed physical variables remains linearly decodable through the policy head up to the layer preceding the action output. Complementary theoretical analysis relates privileged-variable recoverability to the achievable-return gap between history-based and privileged-information policy classes. These results suggest that semantic objectives can structure learning without requiring a corresponding architectural decomposition of the deployed controller.

cs.RO↗

arXiv2026

Limited Structural Reliability in Public Educational Prediction Benchmarks: A Four-Dimension Audit of Seven Datasets

Across seven public educational prediction datasets, three passed all four pre-modeling reliability checks; the remaining four either failed group-aware generalization tests or lacked the provenance metadata needed to run them. One dataset was initially classified as failing but corrected after excluding group-identifier features from the holdout matrix, demonstrating that the audit can distinguish genuine cross-group confounding from feature-encoding artifacts. Each dataset was audited before model optimization using four checks: baseline gap, split instability, null separation, and metadata adequacy under group-aware holdout. The dominant failure mode was not weak iid performance alone but cross-group fragility: in the clearest case, UCI Student declined from iid R-squared 0.242 to group-holdout R-squared -0.097, while Higher Ed collapsed from 0.041 to -8.79. Increasing model complexity did not remove this pattern: ensemble models improved structurally sound datasets but amplified instability or failed under group holdout on fragile ones. An exploratory cross-dataset comparison further showed that stronger profiles clustered in larger, richer-grouped, performance-proximal datasets, while random-split performance severely overstated deployable signal in fragile datasets. Classification-metric sensitivity analyses reached the same substantive conclusions. The results show that benchmark reliability in educational AI is constrained less by algorithm choice than by data structure, group heterogeneity, and evaluation design. A reusable pre-modeling audit offers a minimum quality gate before public educational datasets support strong benchmark or deployment claims.

cs.LG↗

arXiv2026

HuBMAP Data Portal: a resource for multimodal spatial and single-cell data of healthy human tissues

The NIH Human BioMolecular Atlas Program (HuBMAP) Data Portal (https://portal.hubmapconsortium.org/) serves as a comprehensive repository for multimodal, multi-scale spatial and single-cell data from healthy human tissues. As of August 2026, the portal hosts 9,316 public datasets from 26 data types spanning 29 organ classes across 501 donors. Portal infrastructure and user interfaces support data search and discovery, visualization, and analysis directly in web browsers. These capabilities include metadata- and data-driven search, collaborative Workspaces with access to high-performance compute, and interactive Vitessce visualizations across non-spatial, 2D, and 3D spatial datasets. Data-type-specific uniform processing pipelines and rigorous quality control processes ensure comparability of results across laboratories, organs, and donors, while externally processed community-contributed datasets provide complementary perspectives. Here we describe portal functionality, infrastructure, and design, and highlight its role as a platform for large-scale spatial single-cell research across diverse data types, organs, and scales.

q-bio.QM↗

arXiv2026

Adaptively trained Physics-informed Radial Basis Function Neural Networks for Solving Multi-asset Option Pricing Problems

The present study investigates the numerical solution of Black-Scholes partial differential equation (PDE) for option valuation with multiple underlying assets. We develop a physics-informed (PI) machine learning algorithm based on a radial basis function neural network (RBFNN) that concurrently optimizes the network architecture and predicts the target option price. The physics-informed radial basis function neural network (PIRBFNN) combines the strengths of the traditional radial basis function collocation method and the physics-informed neural network machine learning approach to effectively solve PDE problems in the financial context. By employing a PDE residual-based technique to adaptively refine the distribution of hidden neurons during the training process, the PIRBFNN facilitates accurate and efficient handling of multidimensional option pricing models featuring non-smooth payoff conditions. The validity of the proposed method is demonstrated through a set of experiments encompassing a single-asset European put option, a double-asset exchange option, and a four-asset basket call option.

cs.LG↗

arXiv2026

How is Water released in Hydrogen-Based Metal Oxide Reduction? Unraveling the Kinetic Bottleneck in Sustainable Metal Production

Hydrogen-based direct reduction of metal oxides is a ubiquitous solid-gas redox process central to geophysics, sustainable metallurgy, redox energy cycles and catalysis. During this process, hydrogen removes lattice oxygen to form water, yet product water has long been regarded as a passive exhaust, and its nanoscale formation, trapping and removal remain poorly understood. Here, we directly observe redox-product water release from iron oxide during hydrogen-based direct reduction. Because water removal emerges from coupled structural, chemical and crystallographic evolution across multiple length-scales under realistic non-equilibrium reaction-conditions, we establish a correlative multiscale in-situ approach that links pore evolution, molecular water signatures, phase transformation and chemical-state evolution during hematite reduction. We uncover a mechanism in which oxygen removal induces closed nanopores spatially delocalized from reaction surfaces, causing transient trapping of water vapor. Water is released only when these pores coalesce into a percolating network connected to the surface, coinciding with and accelerating the onset of the hematite-to-magnetite transformation. These findings show that dynamically evolving pore topology governs mass transport and redox kinetics in solid-gas reactions, closing a critical mechanistic gap in product-water removal and providing nanoscale guidance for hydrogen-based metal extraction, reactor design, and sustainable redox energy technologies under practical conditions.

physics.chem-ph↗

arXiv2026

ClaimDiff-RL: Fine-Grained Caption Reinforcement Learning through Visual Claim Comparison

Long-form image captioning exposes a reward granularity problem in RL: captions are judged as whole sequences, while the important errors occur at the level of individual visual claims. A good dense caption should be both faithful and informative, avoiding hallucination without omitting salient details. Yet pairwise preferences, reference-based metrics, and holistic scalar rewards compress these local errors into a single sequence-level signal, obscuring the tradeoff between factuality and coverage. We introduce ClaimDiff-RL, a framework that uses reference-conditioned atomic claim differences as the reward unit for caption RL. Given an image, an actor caption, and a reference caption, a multimodal judge enumerates visually grounded differences, verifies each difference against the image, assigns open-vocabulary error types and severity levels, and produces per-difference statistics for reward composition. This makes hallucinated claims and omitted salient facts separately measurable and tunable. Experiments show that holistic scalar rewards can reduce hallucination by increasing missing facts, while ClaimDiff-RL exposes this faithfulness and coverage tradeoff and enables more balanced operating points. On a 160-image human-labeled diagnostic benchmark, public captioning benchmarks, and VQA benchmarks, ClaimDiff-RL improves the hallucination--missing-fact balance, preserves general capability, and even surpasses Gemini-3-Pro-Preview on several fine-grained Capability dimensions such as object counting, spatial relations, and scene recognition. These results suggest that typed, verifiable claim differences are an effective reward unit for fine-grained and diagnosable caption RL.

cs.LG↗

arXiv2026

What Does Vision Tool-Use Reinforcement Learning Really Learn? Disentangling Tool-Induced and Intrinsic Effects for Crop-and-Zoom

Vision tool-use reinforcement learning (RL) can equip vision language models with visual operators such as crop-and-zoom and achieves strong performance gains, yet it remains unclear whether these gains are driven by improvements in tool use or evolving intrinsic capabilities. We introduce MED (Measure--Explain--Diagnose), a coarse-to-fine framework that disentangles intrinsic capability changes from tool-induced effects, decomposes the tool-induced performance difference into gain and harm terms, and probes the mechanisms driving their evolution. Across checkpoint-level analyses in the crop-and-zoom setting on two VLMs with different tool priors and six benchmarks, we find that improvements are dominated by intrinsic learning, while tool-use RL mainly reduces tool-induced harm (e.g., fewer call-induced errors and weaker tool schema interference) and yields limited progress in tool-based correction of intrinsic failures. Overall, in the crop-and-zoom setting studied here, current vision tool-use RL learns to coexist safely with tools rather than master them.

cs.CV↗

arXiv2026

One RL to See Them All: Visual Triple Unified Reinforcement Learning

Reinforcement learning (RL) is becoming an important direction for post-training vision-language models (VLMs), but public training methodologies for unified multimodal RL remain much less mature, especially for heterogeneous reasoning and perception-heavy tasks. We propose V-Triune, a Visual Triple Unified Reinforcement Learning methodology for unified multimodal RL. It organizes training around three coordinated abstractions: Sample-Level Reward Routing, Verifier-Level Outcome Verification, and Source-Level Diagnostics. Within this methodology, Dynamic IoU provides localization-specific reward shaping that avoids reward ambiguity under loose thresholds and reward sparsity under strict ones. Built on V-Triune, we develop Orsta (7B, 32B), a family of models jointly trained on eight reasoning and perception tasks. Under matched budgets, unified training matches or outperforms specialist mixtures. The final Orsta models improve over their backbones on MEGA-Bench, compare favorably with strong multi-task RL-VLM baselines, and transfer these gains to a broad set of downstream benchmarks. These results show that unified RL can improve both reasoning and perception within a single VLM RL pipeline.The V-Triune system, along with the Orsta models, is publicly available at https://github.com/MiniMax-AI/One-RL-to-See-Them-All.

cs.CV↗

arXiv2026

Speed by Simplicity: A Single-Stream Architecture for Fast Audio-Video Generative Foundation Model

We present daVinci-MagiHuman, an open-source audio-video generative foundation model for human-centric generation. daVinci-MagiHuman jointly generates synchronized video and audio using a single-stream Transformer that processes text, video, and audio within a unified token sequence via self-attention only. This single-stream design avoids the complexity of multi-stream or cross-attention architectures while remaining easy to optimize with standard training and inference infrastructure. The model is particularly strong in human-centric scenarios, producing expressive facial performance, natural speech-expression coordination, realistic body motion, and precise audio-video synchronization. It supports multilingual spoken generation across Chinese (Mandarin and Cantonese), English, Japanese, Korean, German, and French. For efficient inference, we combine the single-stream backbone with model distillation, latent-space super-resolution, and a Turbo VAE decoder, enabling generation of a 5-second 256p video in 2 seconds on a single H100 GPU. In automatic evaluation, daVinci-MagiHuman achieves the highest visual quality and text alignment among leading open models, along with the lowest word error rate (14.60%) for speech intelligibility. In pairwise human evaluation, it achieves win rates of 80.0% against Ovi 1.1 and 60.9% against LTX 2.3 over 2000 comparisons. We open-source the complete model stack, including the base model, the distilled model, the super-resolution model, and the inference codebase.

cs.CV↗

arXiv2026

KeySense: LLM-Powered Hands-Down, Ten-Finger Typing on Commodity Touchscreens

Existing touchscreen software keyboards prevent users from resting their hands, forcing slow and fatiguing index-finger tapping ("chicken typing") instead of familiar hands-down ten-finger typing. We present KeySense, a purely software solution that preserves physical keyboard motor skills. KeySense isolates intentional taps from resting-finger noise using cognitive-motor timing patterns, and then uses a fine-tuned LLM decoder to convert the resulting noisy letter sequence into the intended word. In controlled component tests, the decoder substantially outperforms two statistical baselines (top-1 accuracy 84.8% vs 75.7% and 79.3%). A 12-participant study shows clear ergonomic and performance benefits: compared with the conventional hover-style keyboard, users rated KeySense as markedly less physically demanding (NASA-TLX median 1.5 vs 4.0), and after brief practice typed significantly faster (WPM 28.3 vs 26.2, p < 0.01). These results indicate that KeySense enables accurate, efficient, and comfortable ten-finger text entry on commodity touchscreens without any extra hardware.

cs.HC↗