Searcharxiv⌕ Search

arXiv subjects

Ming Li

Publications and source records attributed to Ming Li.

At least 91 records · Page 5Linked to original sources
arXiv2026

A Quantitative Approximation Framework for Flow Distillation in Diffusion Models

We develop a quantitative framework for diffusion distillation by viewing few step sampling as approximation through compositions of learned flow maps. For trajectory distillation of the probability flow ODE, we show that low noise multimodal regimes separate score approximability from dynamical stability: the score remains efficiently approximable, while small local errors may be strongly amplified by stiff flow dynamics. In a Gaussian mixture Ornstein--Uhlenbeck model, we prove time uniform \(L^p(p_t)\) score approximation by ReLU and ReQU networks with explicit polylogarithmic complexity, and derive a computable Lipschitz bound \(L(t)\) for the flow velocity. The stability factor \(\exp\bigl(\int_s^t L(u)\mathrm du\bigr)\) can grow exponentially as noise decreases and mixture separation increases. Comparing this certificate with a certified local Lipschitz budget for one step students identifies regimes of direct distillation difficulty, without implying an approximation lower bound. We also show that deep residual compositions control global transport error through propagated local errors, and that equalizing cumulative stability yields an optimal nonuniform segmentation. With eight segments, this grid reduces final mean relative MSE by up to \(51.9\%\) versus uniform grids.

stat.ML↗
arXiv2026

How Much Sensing Information Is Needed to Control an Unstable Linear System?

Modern control systems increasingly rely on sensing to infer the system state before control actions can be taken. Yet a given observation mechanism may fail to preserve sufficient information about the unstable modes, regardless of the downstream estimator or controller. This paper asks how much sensing information is needed to estimate and control an unstable linear system, whose measurements are generated by a prescribed, possibly nonlinear and non-Gaussian, observation law p(y_t|x_t). To address this question, we first quantify sensing information using directed information, thereby accounting for causal feedback. We then establish necessary and sufficient information rate conditions for estimating and controlling this linear system. For necessity, keeping either the estimation error or the closed-loop state bounded in mean square requires a directed information rate of at least the open-loop expansion rate R_exp. This lower bound remains valid under additive process noise. Since this rate is difficult to evaluate, we derive computable bounds for nonlinear observations with additive noise. An upper bound below R_exp certifies infeasibility, whereas a lower bound above R_exp + R_NG certifies sufficiency under posterior covariance regularity. For linear Gaussian observations, the tight upper bound is determined by the steady-state Riccati equation. For sufficiency, the posterior non-Gaussianity rate R_NG measures the divergence rate from the covariance-matched Gaussian. Under uniform posterior covariance regularity, a rate above R_exp + R_NG guarantees mean-square convergence of the estimation error. For a stabilizable plant, certainty-equivalence feedback also guarantees mean-square convergence of the closed-loop state. Finally, verifiable curvature conditions on the likelihood and prior make R_NG vanish, so the sufficient threshold equals R_exp.

eess.SY↗
arXiv2026

Beyond Representational Similarity: Source-Conditioned Description-Length Gain for Generative Plagiarism Detection and Candidate Source Reranking

Large language models (LLMs) pose challenges to academic integrity and peer review. Yet generative plagiarism detection remains an underexplored and largely unresolved challenge. Prior work on LLM-generated-text detection targets AI involvement, which may be permissible, rather than source reuse, while similarity-based methods struggle after extensive rewriting and multi-source synthesis. Motivated by the description-length view of probabilistic prediction, in which relevant side information can reduce a target sequence's code length, we introduce Source-Conditioned Description-Length Gain (SCDG), a directional, training-free framework that contrasts a frozen language model's description length of a suspicious document $P$ with and without a candidate source $S$. This contrast yields token-level log-likelihood gains that measure the incremental predictive evidence supplied by $S$. We evaluate SCDG on the PAN at CLEF benchmarks for generative plagiarism. On a PAN 2025-derived pairwise benchmark, SCDG achieves 0.92 Precision, 0.97 Recall, and 0.94 F1, outperforming all baselines; on PAN 2026's multi-source retrieval task, it reaches 0.83 nDCG@10 and 0.96 Recall@100, surpassing all baselines. On a same-topic, same-event Multi-News test, the calibrated gain-distribution SCDG classifier predicts source reuse for only $0.125\%$ of pairs, supporting robustness to topical overlap under this evaluation protocol. These results establish SCDG as a unified and token-decomposable signal for source-specific content reuse under extensive transformation.

cs.CL↗
arXiv2026

Thinking Hard, Not Smart: Reasoning Models Fail to Ration Test-Time Compute Across Questions

Reasoning language models increasingly use test-time compute to improve performance, but existing evaluations typically study this compute one question at a time. Yet when multiple problems share an end-to-end cost or latency constraint, models must decide how to divide limited inference compute among them. We introduce an exam-style evaluation framework for studying this setting, in which a model must distribute one shared token budget across questions with different difficulty and point values to maximize its total score. Across several open and frontier reasoning models, we find that models fail to allocate a shared budget strategically across questions of varying difficulties and values. Models behave largely as greedy sequential solvers: they prioritize questions by presentation order, front-load effort on early questions, and remain insensitive to value, with these tendencies becoming more pronounced as the number of questions grows. Explicit planning prompts spread compute more evenly but do not produce value- or difficulty-aware prioritization. The same behavioral pattern extends from mathematical to code reasoning. These findings establish global budget allocation as a distinct capability that is not captured by conventional per-question evaluation and remains a challenge for current reasoning models.

cs.CL↗
arXiv2026

From Neural Intent to Cryptographic Authorization: Securing AI-Driven Enterprise Workflows

The rapid adoption of artificial intelligence (AI)-driven workflows is transforming high-consequence government and enterprise systems into language-based, tool-using and increasingly autonomous infrastructures. While these workflows can delegate planning autonomously, security-critical execution should be strictly mediated. Conventional identity management services authenticate who may invoke a primitive, but remain agnostic to which workflow steps are authorized at runtime. An AI-driven workflow can still be hijacked by injection attacks into executing malicious actions that satisfy identity checks yet violate user intent. We propose Neural Cryptographic Services (NCS), a neuro-symbolic security enforcement plane interposed between neural planners and privileged tools. NCS decouples cognitive planning from execution authority: an untrusted neural planner drafts structured plans, while a deterministic symbolic controller gates execution using an offline-signed, hash-chained instruction stream. Specifically, NCS validates cryptographic signatures and hash chains incrementally, releasing a single instruction template at a time, and admitting a tool call only when its proposed parameters satisfy the constraints of the signed template. Out-of-order or altered tool calls fail-closed, and state transitions are logged for post-hoc auditing. NCS does not attempt to prevent neural planner compromise under injection; it guarantees that a compromised planner cannot dispatch actions outside the authorization. We evaluate NCS using AgentDojo, a custom argument-hijacking dataset, adaptive adversarial instructions, and TheAgentCompany. NCS drives attack success rates to near zero while preserving acceptable utility on benign workflows.

cs.CR↗
arXiv2026

How Can Rhetoric Reward-Hack AI Reviewers? Dissecting Rhetorical Sensitivity in AI-Based Peer Review

As large language models increasingly participate in scientific evaluation, we investigate a potential form of reward hacking: how rhetorical choices shape AI-review judgments when reported scientific content is preserved and how these effects vary across evaluation conditions. We construct a controlled corpus of 4,200 full-paper manuscripts derived from 120 anonymized ICLR 2026 submissions. Two LLM rewriters transform six rhetorical dimensions in opposing directions, and five LLM reviewers evaluate the resulting manuscripts under standard and strict protocols. We also test joint, recursive, and reviewer-guided rewriting. Our results show that rhetorical sensitivity is structured rather than uniform. Evidence framing and novelty stance produce the largest positive-negative contrasts in overall assessment, with scope framing forming a weaker second tier; the remaining dimensions have smaller or less stable effects. This hierarchy persists across human-assessed quality levels, but score movement depends strongly on the AI reviewer's original score: lower scores tend to rise, higher scores tend to fall, and directional contrasts are clearest in the middle ranges. More elaborate workflows do not reliably yield larger gains. Joint rewriting is strongly rewriter-dependent, reviewer guidance does not consistently outperform an unguided second pass, and repeated rewriting yields diminishing, configuration-dependent returns. Across conditions, the rewriter primarily determines the separation between opposing variants, whereas the reviewer determines the magnitude and sign of their score effects. Strict review lowers mean OA by 1.36 points without consistently changing rhetorical sensitivity. These findings identify when rhetorical presentation influences AI scientific review and motivate evaluation systems robust to content-preserving variation in scientific writing.

cs.CL↗
arXiv2026

Cognitive Episodes in LLM Reasoning Traces Enable Interpretable Human Item Difficulty Prediction

Predicting human item difficulty is central to educational assessment, where reliable estimates support fairness and effective test construction. Existing methods often depend on costly human calibration or item-level textual representations, providing limited evidence about the cognitive processes that make items difficult. We argue that difficulty should be viewed not only as a property of item text, but also as an observable consequence of the problem-solving burden an item induces. Large Reasoning Models (LRMs) offer scalable process evidence through reasoning traces, but such evidence must be structured to support interpretable modeling. To this end, we introduce Epi2Diff (Episode to Difficulty), a framework that maps LRM reasoning traces into cognitively grounded episode sequences. These episodes group trace segments into functional problem-solving states, enabling difficulty to be modeled through reasoning scale, effort allocation, and state transitions. Epi2Diff extracts compact episode-dynamic features and combines them with semantic item representations for human difficulty prediction. Experiments on four real-world human difficulty datasets show that Epi2Diff consistently outperforms strong baselines, including fine-tuned small language models, LLM in-context learning, and supervised LLM adaptation. On SAT-derived classification benchmarks, Epi2Diff achieves an 8.1% average relative gain over supervised LLM fine-tuning baselines. Further analyses show that harder items induce more effortful, iterative, and implementation-centered episode dynamics, rather than merely longer responses. These results demonstrate that cognitive episodes in LRM reasoning traces provide a predictive and interpretable process representation for human item difficulty, offering a new lens for educational measurement with reasoning models.

cs.CL↗
arXiv2026

Identity-Faithful Audio-Visual Target Speaker Extraction with REAL-2MIX and VOXBLINK2-AVSE

Audio-visual target speaker extraction should return the speaker indicated by the video, yet a separator can ignore the visual cue and repeatedly output the acoustically dominant voice. We introduce REAL-2MIX, a Mandarin AV-TSE benchmark of jointly recorded real two-speaker mixtures with synchronized multi-view video. Each scene also contains preceding A-only and B-only stages that provide in-scene speaker references. It contains 77 scenes and 7,598 clips (11.84 hours), including 6,042 dual-annotated mixtures. After processing there leave 6,038 evaluable mixtures and 12,076 target-speaker rows. We additionally curate VOXBLINK2-AVSE from VoxBlink2, comprising 250,828 synchronized audio--lip-ROI pairs from 28,421 identities and 766.17 hours of speech. Our extractor uses frozen, 1,280-dimensional projected AV-HuBERT features, target-conditioned training, and layer-wise feature modulation. We jointly evaluate content with Qwen3-ASR-1.7B CER and target identity with WeSpeaker ResNet34 plus Overlapped Speech Detection (OSD). On the complete manifest, the best archived checkpoint obtains 0.2261 CER, 82.22% strict output correctness, and 69.53% both-output strict success.

eess.AS↗
arXiv2026

DVAR: Adversarial Multi-Agent Debate for Video Authenticity Detection

The rapid evolution of video generation technologies poses a significant challenge to media forensics, as conventional detection methods often fail to generalize beyond their training distributions. To address this, we propose DVAR (Debate-based Video Authenticity Reasoning), a training-free framework that reformulates video detection as a structured multi-agent forensic reasoning process. Moving beyond the paradigm of pattern matching, DVAR orchestrates a competition between a Generative Hypothesis Agent and a Natural Mechanism Agent. Through iterative rounds of cross-examination, these agents defend their respective explanations against abnormal evidence, driving a logical convergence where the truth emerges from rigorous stress-testing. To adjudicate these conflicting claims, we apply Occam's Razor through the Minimum Description Length (MDL) framework, defining an Explanatory Cost to quantify the "logical burden" of each reasoning path. Furthermore, we integrate GenVideoKB, a dynamic knowledge repository that provides high-level reasoning heuristics on generative boundaries and failure modes. Extensive experiments demonstrate that DVAR achieves competitive performance against supervised state-of-the-art methods while exhibiting superior generalization to unseen generative architectures. By transforming detection into a transparent debate, DVAR provides explicit, interpretable reasoning traces for robust video authenticity assessment.

cs.CV↗
arXiv2026

LLMs Struggle to Measure What Distinguishes Students of Different Proficiency Levels: A Study of Item Discrimination in Reading Comprehension Assessment

Existing work on LLM-based educational assessment has focused largely on item difficulty, but difficulty alone does not indicate whether an item meaningfully distinguishes higher- from lower-proficiency students. Item discrimination captures this complementary and fundamental psychometric property. We investigate whether LLMs can predict human item discrimination from assessment content. We evaluate 42 proprietary and open-weight LLMs using two complementary approaches. Direct discrimination prediction asks models to explicitly predict an item's discrimination value, while response-based proxy estimation treats LLM answers as synthetic responses and applies a Classical Test Theory (CTT)-inspired item-rest calculation. Direct predictions show weak alignment with human item discrimination. The response-based proxy provides a stronger but still limited ranking signal, reaching a CEFR-stratified rank correlation of 0.231. Further analysis shows that this correlation comes mainly from differences across models rather than proficiency prompts that reliably simulate students at different ability levels. Current LLMs therefore contain some discrimination-relevant information, but they do not yet reliably model the ability-conditioned human response behavior that gives item discrimination its psychometric meaning.

cs.CL↗
arXiv2026

Representing Visual Evidence for Item Difficulty Prediction: Visual Textualization and Image-Native Modeling

Predicting item difficulty from content can provide an initial estimate for newly developed questions before sufficient student responses are available. Existing approaches typically represent the question stem and answer choices as text. When mathematics items contain visual components, a common pipeline first textualizes that evidence and then applies a text predictor. We ask: how should visual evidence be represented for item difficulty prediction? We compare question text alone, visual textualization, which expresses visual evidence in language, and image-native modeling, which retains the original image. Using Eedi items with difficulty calibrated from student responses, we train large language models (LLMs) and vision-language models (VLMs) directly for difficulty regression. Both visual interfaces achieve the lowest point estimates, although the leading systems cannot be reliably ordered. Open-VLM textualization yields lower RMSE point estimates for all evaluated LLMs, while broader adaptation does so for all image-native VLMs. Test-time interventions show dependence on the paired full-item image, but do not isolate the additional visual component. The two visual interfaces also make partially complementary item-level errors and differ substantially in computational workflow. Thus, textualization should not be treated as the only practical interface: image-native modeling is a competitive alternative whose effectiveness depends on how the VLM is adapted.

cs.CL↗
arXiv2026

A Comprehensive FP8 Training Recipe for Reasoning-Enhanced Language Models

The immense computational cost of training Large Language Models (LLMs) presents a major barrier to innovation. While FP8 training offers a promising solution with significant theoretical efficiency gains, its widespread adoption has been hindered by the lack of a comprehensive, open-source training recipe. To bridge this gap, we introduce an end-to-end FP8 training recipe that seamlessly integrates continual pre-training and supervised fine-tuning. Our methodology employs a fine-grained, hybrid-granularity quantization strategy to maintain numerical fidelity while maximizing computational efficiency. Through extensive experiments, including the continue pre-training of models on a 160B-token corpus, we demonstrate that our recipe is not only remarkably stable but also essentially lossless, achieving performance on par with the BF16 baseline across a suite of reasoning benchmarks. Crucially, this is achieved with substantial efficiency improvements, including up to a 22% reduction in training time, a 14% decrease in peak memory usage, and a 19% increase in throughput. Our results establish FP8 as a practical and robust alternative to BF16, and we will release the accompanying code to further democratize large-scale model training.

cs.CL↗
arXiv2026

The MiniMax-M2 Series: Mini Activations Unleashing Max Real-World Intelligence

We introduce the MiniMax-M2 series, a family of Mixture-of-Experts language models built around the principle that mini activations can unleash maximum real-world intelligence. The flagship M2 contains 229.9B total parameters with only 9.8B activated per token. Designed end-to-end for agentic deployment, the M2 series rests on three components: (i) agent-driven data pipelines producing large-scale, verifiable trajectories across agentic coding and agentic cowork, each grounded in an executable workspace and an artifact-aligned reward; (ii) Forge, a scalable agent-native RL system that adapts to long-horizon agent trajectories, paired with windowed-FIFO scheduling, prefix-tree merging, inference optimization, and a clean training-inference-agent decoupling that supports both white-box and black-box agents; (iii) the latest M2.7 checkpoint takes an early step toward self-evolution -- autonomously debugging training runs and modifying its own scaffold. Across M2 through M2.7, this combination translates a mini-activation footprint into frontier-tier performance on agentic coding, deep search, office-task, and reasoning benchmarks.

cs.AI↗
arXiv2026

A Framework for Developing University Policies on Generative AI Governance: A Cross-national Comparative Study

As generative AI (GAI) becomes increasingly embedded in higher education, universities worldwide are developing policies to govern its ethical, pedagogical, and institutional use. However, these policies vary across national and institutional contexts. We undertake a cross-nationalanalysis of GAI guidelines issued by leading universities in the United States, Japan, and China, identifying key policy orientations and proposing a structured framework to support policy development. Using an extended Technology Acceptance Model as an analytical lens, we examine five domains Perceived Usefulness and Perceived Ease of Use, Perceived Risk, Facilitating Conditions, Social Influence, and Self-Efficacy, and identify 20 key themes through thematic coding. Together, these findings inform the development of the University Policy Development Framework for Generative AI (UPDF-GAI). U.S. universities emphasize faculty autonomy, practical application, and policy adaptability, reflecting environments shaped by cutting-edge research and peer collaboration. The Japanese universities analyzed adopt a more government-aligned approach, prioritizing ethics and risk management, but offering comparatively limited guidance on AI implementation and flexibility. The Chinese universities in the sample reflect a centralized, government-led model, focusing on technology application rather than early policy formulation, while actively exploring GAI integration in education and research. Based on these insights, the study proposes the UPDF-GAI, integrating technological, organizational, and social dimensions of policy formation. The framework provides a structured approach for universities to assess policy priorities, navigate tensions between innovation and risk, and strengthen institutional capacity for sustainable GAI governance, contributing to the evolving discourse on AI governance in higher education.

cs.CY↗
arXiv2026

Heisenberg scaling in optical magnetometry with measurement-induced correlations as a quantum resource

Theoretical proposals to reach the Heisenberg scaling of the measurement precision typically require carefully engineered interactions or initial entanglement. In studying optical magnetometry, we show that the continuous collective measurement process itself can generate the necessary many-body quantum correlations to achieve the elusive Heisenberg scaling of the quantum Fisher information in a dissipative, steady-state system without direct inter-atomic interactions. By contrasting a correlation-neglecting but otherwise consistent semiclassical model, which can violate the quantum Cramér-Rao bound (QCRB) by several orders of magnitude, with a collective quantum model, we isolate measurement-induced correlations as the essential mechanism. The violation of the QCRB serves thereby as a fundamental sanity check for semiclassical spectroscopic theories. This work reveals measurement-induced correlations as a widely unexplored quantum resource for quantum-enhanced sensing, establishes a new paradigm for achieving Heisenberg scaling in open quantum systems, and provides a direct path to test the foundations of quantum mechanics using macroscopic atom ensembles.

quant-ph↗
arXiv2026

The Hidden Footprint: Making Storage a First-Class Metric for LLM Agent Evaluation

LLM agent benchmarks measure task completion, reliability, and inference cost, but not the persistent data an agent run leaves on disk, including logs, context snapshots, checkpoints, and debug traces. We introduce AgentFootprint, a cross-framework benchmark of post-run agent storage footprint. Its serialization-aware metric suite measures total retention, channel composition, duplication, growth, compressibility, and conversation-history reconstructability. It addresses a measurement trap: naive byte-level measurement understates duplication by an order of magnitude because database paging and JSON escaping obscure repeated content. A fixed-trace control separates agent-generated logical volume from persistence-layer amplification: replaying the same trajectory through seven persisting frameworks yields a 6.7x spread. Under identical models, tools, and tasks, configurations with 100% accuracy differ by 15.7x in retained bytes, although their defaults support different recovery and audit capabilities. Three full-history configurations grow superlinearly on a repeated-observation stress task. Exported trajectories from 108 instance-normalized SWE-bench Verified submissions span three orders of magnitude per instance, with no detectable correlation with resolve rate. A content-addressed store reduces retention by 4.8x-32.7x while preserving every reconstructability score. These results establish persistent storage as a resource metric to report jointly with accuracy and reconstructability.

cs.AI↗
arXiv2026

Compile, Then Page: Executable SOP Programs and a Capability-Gated Runtime for Procedural LLM Agents

Enterprise agents must follow long-horizon, conditional, safety-critical standard operating procedures (SOPs). We compile machine-readable SOP constraints into executable pseudo-code and run them with a program-guided (PG) stack machine that pages the active frame while an LLM performs semantic execution. A three-arm SOPBench study across six models separates representation from runtime: compiled text never significantly hurts and gains up to 16.0 points where official prose underperforms. Runtime guidance is capability-gated. Two strong models independently show positive seven-domain PG contrasts (58:19 and 75:31 discordant pairs), whereas weak models are harmed. A full-program cursor ablation (active frame first, complete program retained) recovers much of the strong-model refusal gain; selective visibility adds a smaller improvement. Paired probe and audit measurements track this divide to spontaneous state discipline rather than reconstruction ability. On Bank the three primary arms rise from 70.4 to 86.4 to 92.8, with 100% refusal correctness. Practical guidance: compile first; enable active-frame paging only after a model-level discipline check.

cs.AI↗
arXiv2026

Investigating Codec-Internal Latent Audio Watermarking for Neural Codec Robustness

Neural audio codecs are challenging transformations for audio watermarking because they re-encode, quantize, and resynthesize speech. This paper investigates continuous latent-space watermarking for codec robustness. Instead of adding a watermark only to the waveform or spectrogram, we embed a 32-bit message into the continuous latent representation of a codec-like speech autoencoder. The pipeline uses a SEANet-style encoder-decoder, a Conformer-based message embedder, RVQ-guided latent decomposition, and a latent-domain detector trained under signal-processing and neural-codec transformations. Rather than proposing a final universal watermarking baseline, we characterize the trade-offs that appear when the watermark carrier is moved before neural decoding. On 48 kHz speech, EnCodec-aware training improves EnCodec-24k bit accuracy from 78.8% to 95.6% and 97.1%, while PESQ decreases from 3.727 to 3.514 and 3.427.

cs.SD↗