arXiv subjects

Yiming Chen

Publications and source records attributed to Yiming Chen.

At least 37 records · Page 2Linked to original sources

arXiv2026

When Modalities Fail to Tango: Conformal Backdoor Detection in Multimodal Contrastive Learning

Backdoor attacks in multimodal contrastive learning (MCL) have garnered growing attention in recent years, as many downstream tasks critically depend on pre-trained MCL models. Existing detection-based defenses predominantly rely on the CLIPScore metric, under the assumption that poisoned pairs exhibit lower semantic similarity between the image and the caption. However, we identify two critical flaws remaining in existing methods: (1) the substantial overlap between CLIPScore distributions of benign and poisoned pairs undermines the reliability of this metric, and (2) fixed-threshold detection cannot provide statistical guarantees for ambiguous samples within overlapping regions. To overcome these limitations, we propose integrating conformal prediction (CP), a statistical framework that quantifies uncertainty through nonconformity scores (NCSs), to establish provable confidence bounds for detecting poisoned image-caption pairs. Building on CP, we introduce CASCADE, a novel two-stage Coarse-to-Fine Conformal Backdoor Detection framework. The coarse-grained stage uses cross-modality consistency to identify high-confidence benign and poisoned pairs. In the fine-grained stage, a reference set is constructed from high-confidence poisoned pairs, and instance-level NCSs based on text-space similarity are computed for each sample in the unidentified subset. These NCSs measure conformity to the poisoning distribution and enable precise identification of latent poisoned pairs within the unidentified subset. Extensive experiments on the large-scale CC3M dataset demonstrate that CASCADE achieves an average FPR of 5.79% at 100% TPR and an average AUROC of 0.9867 across diverse attacks, while remaining effective against adaptive attacks.

cs.CR↗

arXiv2026

Test-time Scaling over Perception: Resolving the Grounding Paradox in Thinking with Images

Recent multimodal large language models (MLLMs) support Thinking with Images, invoking visual tools such as zooming and cropping to inspect image regions during inference. Yet these systems remain brittle in fine-grained reasoning: to acquire a decisive detail, a model must ground its attention on the correct region, but knowing which region is correct presupposes having already observed that detail. We identify this circular dependency as the grounding paradox, show that grounding errors are rarely self-corrected within a single trajectory---once a misleading region is inspected, all subsequent reasoning conditions on that observation and the error propagates to the final answer---and observe that because each trajectory constructs its own evidence, answer-level aggregation discards the very information that distinguishes trajectories. We propose Test-Time Scaling over Perception (TTSP), a closed-loop framework that treats perception as the unit of scalable inference and allocates compute along two axes: Entropy-Gated Perceptual Exploration samples diverse trajectories and uses critical-token entropy to withhold evidence the model cannot commit to, while Evidence-Guided Iterative Refinement distills validated observations into a correctable Evidence Ledger that steers later rounds to re-inspect unresolved regions. Across high-resolution and general multimodal benchmarks, TTSP consistently outperforms strong test-time scaling baselines, while improving grounding quality with favorable token efficiency.

cs.CV↗

arXiv2026

CADIR: A Cross-Backend Editable Intermediate Representation for Agentic CAD Generation

Large language models have made it possible to generate executable computer-aided design (CAD) programs from natural-language descriptions or images. However, existing methods represent modeling processes as backend-specific sequential scripts with implicit dependencies or as static geometry, making it difficult to simultaneously preserve construction history, stable topological references, and feature-level editability across different CAD systems. We present CADIR, an agent-friendly executable intermediate representation for CAD generation and cross-backend editing. Built on the OCCT geometry kernel via OCP, CADIR provides explicit, compositional modeling operations and fine-grained execution diagnostics. During program execution, CADIR records modeling operations, parameter dependencies, constraints, and topology selections in a construction graph. To enable reliable cross-backend reconstruction, we introduce Geometric Signature Matching, which identifies corresponding edges and faces despite parameter changes and backend differences, allowing adapters to reconstruct native editable feature histories in FreeCAD, SolidWorks, and Fusion 360. Building on this representation, we further propose a construction-graph retrieval method for text and image queries that supports both full-graph and subgraph retrieval, enabling agents to leverage complete models and modeling substructures. Extensive experiments demonstrate that CADIR achieves higher geometric fidelity and execution reliability than existing CAD representations, that construction-graph retrieval further improves model generation quality, and that cross-backend editing enables reliable model reconstruction and post-reconstruction editing across multiple CAD environments.

cs.AI↗

arXiv2026

Gaussian Convexity Principles for Sharp Moderate Deviations of Gaussian Maxima and Critical SK Free Energy Variance

In this paper, we establish a moderate deviation bound for Gaussian maxima and the variance asymptotics of the Sherrington-Kirkpatrick free energy at criticality based on Gaussian convexity. First, let $(X_1,\ldots,X_N)$ be centered Gaussian vector with $\operatorname{Var}(X_i)\leq 1$. Suppose that, for fixed $α\in(0,\sqrt 2)$ and $κ>0$, $\mathbb{E}\max_iX_i\geqα\sqrt{\log N}$ and $\mathbb{E}\max_iX_i+κ\sqrt{\log N}\leq\sqrt{2\log N}$. We prove that $$ \mathbb{P}\left(\max_iX_i\geq \mathbb{E}\max_iX_i+κ\sqrt{\log N}\right) \leq N^{-κ^2/(2-α^2)+o(1)}. $$ This answers a question of Ding, Eldan and Zhai. The exponent is sharp, as witnessed by an equicorrelated Gaussian field. Second, for the Sherrington--Kirkpatrick model at the critical inverse temperature $β_c=1/\sqrt2$, we prove $$ \operatorname{Var}\bigl(F_N(β_c)\bigr)=\frac16\log N+O(1). $$ Our argument provides the variance asymptotics at the critical temperature from an entropy perspective, via a route distinct from that of Du and Huang. For the upper bound, we express the variance as an entropy under exponential tilting and identify this entropy with the Kullback--Leibler divergence of a Gaussian synchronization model. Its derivative is then bounded using the I-MMSE formula, information percolation, and estimates for the susceptibility of the critical Erdős--R'enyi random graph. For the lower bound, we combine Gaussian convexity applied at the replica parameter with an estimate for inverse moments on the sphere and an identity relating GOE eigenvalue densities in consecutive dimensions.

math.PR↗

arXiv2026

Harnessing X-ray Absorption Spectroscopy Data through Multimodal Mining of Battery Literature

X-ray absorption spectroscopy (XAS) is central to understanding the local electronic and atomic structure of materials, yet most published spectra remain inaccessible to data-driven analysis because they are embedded in figures and described through fragmented textual context in the literature. Here, we use multimodal (image and text) literature mining to transform this dispersed knowledge into an AI-ready experimental data resource. We developed a scalable spectroscopy data digitization pipeline that identifies XAS figures in full-text articles, digitizes spectral curves, and links each spectrum to accompanying metadata on the measured edge and material. Applying this pipeline to the battery literature produced an open dataset of 13,740 XAS spectra, spanning 66 absorbing elements and diverse battery chemistries, with expert validation confirming accurate extraction of spectral and metadata information. By converting literature-embedded spectra into structured numerical data, this dataset provides a foundation for large-scale XAS analysis, cross-laboratory comparison, high-throughput characterization, and autonomous discovery of advanced materials.

cond-mat.mtrl-sci↗

arXiv2026

Sharp Asymptotics for the Largest Component in the Subcritical Regime of Preferential Attachment Without Vertex Growth

We study the size of the largest component in Pittel's preferential attachment process without vertex growth. Starting from the empty graph on a fixed vertex set $[n]$, edges are added one by one with probabilities proportional to $(d_u+α)(d_v+α)$, where $d_u$ and $d_v$ are the current degrees of $u$ and $v$, and $α>0$. Let $L_1$ denote the size of the largest component, and set $m_c:=\frac{αn}{2(α+1)}.$ We prove that if $m=m_c(1-\varepsilon), \varepsilon=\varepsilon(n)\to0, \varepsilon^3 n\to\infty,$ then \[ L_1=(1+o_p(1))\frac{2(α+2)}{α+1}\varepsilon^{-2}\log(\varepsilon^3 n) \] for every fixed $α>0$. Moreover, the same asymptotic holds whenever $α=α(n)\to a\in(0,\infty]$. In particular, the constant $2(α+2)/(α+1)$ converges to the Erdős--Rényi value $2$ as $α\to\infty$. If $m=\left\lfloor \frac n2(1-\varepsilon)\right\rfloor$ and $α\varepsilon\to\infty$, then \[ L_1=(2+o_p(1))\varepsilon^{-2}\log(\varepsilon^3 n). \] The subcritical asymptotics for $L_1$ resolve the problem left open by Janson and Warnke. The upper bound argument relies on the fact that, after conditioning on the degree sequence, the graph can be treated through the corresponding configuration model, the lower bound follows from tree component asymptotics and a second moment argument.

math.PR↗

arXiv2026

Equivariant Schubert Calculus for Inverse Grassmannian Permutations

We give a Graham-positive expansion for the product of two double Schubert polynomials indexed by two inverse Grassmannian permutations. Surprisingly, the nonzero structure constants are double Schubert polynomials in two disjoint sets of equivariant variables. We also give a positive expansion for the product of two single Schubert polynomials indexed by a $321$-avoiding permutation (e.g., a Grassmannian permutation) and an inverse Grassmannian permutation. Unexpectedly, the nonzero structure constants are Edelman--Greene coefficients.

math.CO↗

arXiv2026

Negative shocks versus static patch holography

We study a version of de Sitter static patch holography in which the Euclidean gravitational path integral, with an observer worldline included, is conjectured to compute a trace. Motivated by recent evidence for this conjecture from the sphere path integral, we test it further by inserting operators along the observer worldline and computing two-point functions and out-of-time-ordered four-point correlators (OTOCs). We extend an earlier OTOC calculation by Kolchmeyer and Liu using the shockwave formalism, incorporating both observer recoil and gravitational backreaction. We find that the OTOC conflicts with two basic properties of a trace in a Hilbert space: cyclicity and positivity. The signaling feature of the shockwave geometry gives rise to two distinct resummations of the perturbative eikonal expansion, which are interchanged by cyclicity. Positivity is violated by the fact that the leading perturbative contribution causes the (regularized) OTOC to increase.

hep-th↗

arXiv2026

Escaping the Procrustean Bed: Groupwise Orthogonal Connectors for Audio-Language Models

Audio-language models compress a speech encoder's output through a Querying Transformer (Q-Former) connector before feeding it to a large language model. We identify two failures in this compression. The connector's output vectors collapse to a single direction, and different speakers produce nearly indistinguishable outputs, with paralinguistic cues such as speaker identity, gender, and prosody lost along the way. Our method, ORCA, reverses this collapse by splitting the queries into groups whose outputs are constrained to point in different directions. On SAKURA multi-hop reasoning, ORCA gains 26.4 points over an identically trained 4B baseline, reaching 75.2% (vs. 49.0% for the 8B Audio Flamingo-3). At the connector level, the same change cuts query redundancy by 12x and raises cross-speaker variance by 75x.

cs.SD↗

arXiv2026

Listen, Think, Transcribe: Continuous Latent Test-Time Scaling for ASR

End-to-end ASR models transcribe in a single pass, leaving no room for the decoder to revisit hard inputs. We propose LatentASR, a parameter-efficient method that adds continuous latent test-time scaling to a frozen ASR backbone. Two small trainable modules drive it: a Latent Adapter that iteratively refines a few latent prefix positions through bounded, stabilized updates, and a Value Head that predicts whether extra computation will help and halts the loop early. The Qwen3-ASR-0.6B backbone stays fully frozen, and we train only ~4M extra parameters. We activate this loop with a deliberately small, diverse 500-utterance training set. Under this minimal-data regime, standard adaptation methods all regress: full fine-tuning, LoRA, and prompt tuning each increase WER. LatentASR is the only tested method that reduces WER on both clean benchmarks (FLEURS -2.54% and VoxPopuli -0.47% relative). The reductions are concentrated on intrinsically hard inputs. On accented and code-switched speech (ASCEND), LatentASR achieves a 16.0% relative CER reduction. Across 30 FLEURS languages (23,049 utterances), the multilingual WER decreases uniformly across resource tiers, confirming that the adapter generalizes without overfitting. Dynamic halting preserves most of the clean-set reduction at a fraction of the compute, skipping roughly half of all utterances at the entry gate. Our results show that a small, carefully chosen activation set can switch on test-time scaling inside a frozen ASR model without corrupting the model itself, converting fixed per-utterance compute into input-dependent compute where it is most needed.

cs.SD↗

arXiv2026

Context-Aware ASR for Mandarin Technical Lectures

Technical lectures mix Mandarin speech with English technical terms. These terms carry the core meaning of the lecture, yet they occupy few characters. Character error rate (CER) therefore hides their recognition failures. We study whether lecture context helps recognize these terms. We build a term-rich Mandarin AI/ML lecture benchmark, and we define term-centric metrics that measure technical-term recognition directly. We then propose a two-pass, reference-free decoding method. The first pass runs segment-only ASR. We extract the most frequent technical terms from the first-pass hypotheses, and we prompt the recognizer with this self-built glossary in the second pass. Across five ASR backbones, the first-pass glossary raises term recall for every model and holds or lowers CER on all five. On Breeze-ASR-25 it lifts term recall from 52.50% to 60.13% while lowering CER, and a hybrid that adds a small external term list reaches 62.05% recall and 82.73% term precision. Lecture context, recovered from the model's own output, is a practical signal for technical-term recognition. Term-centric evaluation exposes errors that CER misses.

cs.SD↗

arXiv2026

Randomized second order Riesz projections on the Hamming cube

In this paper, we improve the arbitrary Banach space $n \log n$ bound of Ivanisvili--Volberg \cite{IvanisviliVolberg2022} for the second order projection bound to the order $\sqrt{n}$ bound. Moreover, we study the lower Riesz estimate with the pointwise square gradient, and prove a fixed chaos characterization: on every fixed homogeneous Walsh chaos $H_k$, the dimension free estimate \[ \|Δ^{1/2}f\|_{L^p(Ω_n;X)} \lesssim_{p,k,X} \||\nabla f|_X\|_{L^p(Ω_n)} \] holds for all $n$ if and only if $X$ has Rademacher type $2$. We also consider an exact tail space norm of the analytic paraproduct $T_φg(z)=\int_0^z g(ζ)φ'(ζ)\,dζ$ on Banach valued $H^\infty$ spaces. A matching lower bound of Volberg \cite{Volberg2024} \[ \|T_φ:H_d^\infty(\mathbb D;Y)\to H^\infty(\mathbb D;Y)\| \asymp_{α,φ} d^{-α} \] under a nondegenerate boundary singularity assumption is established.

math.PR↗

arXiv2026

AgentX: Towards Agent-Driven Self-Iteration of Industrial Recommender Systems

Recommendation algorithm iteration is moving from an artisanal, engineer-bound process toward an industrialized research loop, but this transition remains blocked by a structural execution bottleneck: the idea-to-launch cycle still depends on human engineers to generate hypotheses, modify production code, launch A/B experiments, and attribute online results. Innovation therefore scales linearly with headcount rather than compounding with evidence, compute, and accumulated experimental knowledge. We present AgentX, a production-deployed multi-agent system that fundamentally restructures this production function. AgentX operates as a self-evolving development engine: it autonomously generates, implements, evaluates, and learns from recommendation experiments at a scale and pace that no manual workflow can sustain. The system orchestrates four tightly coupled stages in a closed loop. A Brainstorm Agent synthesizes evidence from historical experiments, system architecture, data analysis, and external research into ranked, executable proposals. A Developing Agent translates each proposal into production-ready code through repository-grounded generation and multi-dimensional reliability verification. An Evaluation Agent conducts safe online rollout with guardrail-vetoed A/B judgment, converting both successes and failures into structured knowledge assets. A Harness Evolution layer (SGPO) then distills execution trajectories into semantic-gradient updates that continuously sharpen the agents themselves -- making the system not merely automated, but self-improving.

cs.AI↗

arXiv2026

PhysiFormer: Learning to Simulate Mechanics in World Space

We present PhysiFormer, a diffusion transformer for physically-plausible 3D object motion. Unlike video world models that operate in view-dependent pixel space, PhysiFormer represents objects as 3D meshes expressed in world coordinates. Given the initial vertex positions and velocities, as well as object material type, rigid or elastic, the model samples future vertex trajectories. While related neural physics approaches build on ad-hoc latent spaces or explicitly enforce rigidity and causality, PhysiFormer shows that excellent results can be obtained without any such inductive biases, by casting vertex trajectory prediction as a single denoising diffusion process directly in world coordinates. The probabilistic formulation captures uncertainty in the learned dynamics, enabling diverse plausible futures from initial conditions, making this framework potentially useful for applications with unobserved uncertainty. The model features attention factorised over time, space, and objects for efficiency, enabling permutation-invariant multi-object reasoning without needing explicit object encoding. Trained on over 100k simulated trajectories, PhysiFormer generates rigid and elastic mechanics, and generalises to mixed-material settings, unseen real-world geometries, and larger object counts. It substantially outperforms autoregressive baselines in trajectory accuracy, rigidity preservation, and momentum-based physical consistency. Our results position coordinate-space diffusion as a promising step toward view-invariant, geometry-aware world modelling for robotics, graphics, and physical design. Visualisations, code, and models are available at https://yimingc9.github.io/physiformer.

cs.CV↗

arXiv2026

Availability-Aware and Efficiency-Driven AI Service Chain Provisioning in Multi-Domain Edge Intelligence Cloud

In a multi-domain edge intelligence cloud (MDEIC) managed by multiple network operators, AI services are delivered by chains of virtual network functions (VNFs) executed in sequence, called AI service chains (AISCs). Therefore, achieving an efficient and economical AISC provisioning approach is essential. However, the interaction between the environmental characteristics (heterogeneity, resource constraints and limited information visibility) of MDEIC and the time-dependence of AISCs, introduces various challenges to AISC provisioning in MDEIC. In this paper, we first formulate the AISC provisioning problem as a partially observable stochastic game (POSG). Then, we propose a graph-and-time-based multi-agent AISC provisioning (GT-MAAISCP) approach to achieve the collaborative optimization of AISC provisioning cost, delay and availability. Specifically, each agent uses the graph-time dueling network (GTDN) architecture to extract network topology information and temporal relationships. Finally, the experimental results demonstrate that the proposed approach outperforms benchmark approaches in MDEIC and also illustrate its performance under varying network topologies and different numbers of local EICs (LEICs).

cs.NI↗

arXiv2026

BPS Chaos

Black holes are chaotic quantum systems that are expected to exhibit random matrix statistics in their finite energy spectrum. Lin, Maldacena, Rozenberg and Shan (LMRS) have proposed a related characterization of chaos for the ground states of BPS black holes with finite area horizons. On a separate front, the "fuzzball program" has uncovered large families of horizon-free geometries that account for the entropy of holographic BPS systems, but only in situations with sufficient supersymmetry to exclude finite area horizons. The highly structured, non-random nature of these solutions seems in tension with strong chaos. We verify this intuition by performing analytic and numerical calculations of the LMRS diagnostic in the corresponding boundary quantum system. In particular we examine the 1/2 and 1/4-BPS sectors of $\mathcal{N}=4$ SYM, and the two charge sector of the D1-D5 CFT. We find evidence that these systems are only weakly chaotic, with a Thouless time determining the onset of chaos that grows as a power of $N$. In contrast, finite horizon area BPS black holes should be strongly chaotic, with a Thouless time of order one. In this case, finite energy chaotic states become BPS as $N$ is decreased through the recently discovered "fortuity" mechanism. Hence they can plausibly retain their strongly chaotic character.

hep-th↗

arXiv2026

Error bounds of Median-of-means estimators with VC-dimension

We obtain the upper error bounds of robust estimators for mean vector, using the median-of-means (MOM) method. The method is designed to handle data with heavy tails and contamination, with only a finite second moment, which is weaker than many others, relying on the VC dimension rather than the Rademacher complexity to measure statistical complexity. This allows us to implement MOM in covariance estimation, without imposing conditions such as $L$-sub-Gaussian or $L_{4}-L_{2}$ norm equivalence. In particular, we derive a new robust estimator, the MOM version of the halfspace depth, along with error bounds for mean estimation in any norm.

math.ST↗

arXiv2026

CLR-voyance: Reinforcing Open-Ended Reasoning for Inpatient Clinical Decision Support with Outcome-Aware Rubrics

Inpatient clinical reasoning is a sequential decision under partial observability: the clinician sees the admission so far and must choose the next action whose downstream consequences are not yet visible. Existing clinical-LLM evaluations and RL rewards signals collapse this into closed-form retrieval, clinical journey leakage, or unanchored LLM-as-judge scoring. We introduce CLR-voyance, a framework that reformulates inpatient reasoning as a Partially Observable Markov Decision Process (POMDP) and supervises it with rewards that are simultaneously outcome-grounded and clinician-validated. We instantiate the formulation as CLR-POMDP, which partitions successful patient journeys into a policy-visible past and an oracle-only future. Using the past information, an oracle LLM generates a case-specific query-answer pair, and the first adaptive rubric for clinical reasoning which is verifiable in the future of the patient journey. These rubrics are used for both post-training and evaluation of models for inpatient clinical reasoning. We post-train Qwen3-8B and MedGemma-4B with GRPO followed by model merging, yielding state-of-the-art inpatient clinical reasoning while retaining generalist capabilities. CLR-voyance-8B achieves 84.91% on CLR-POMDP, ahead of frontier medical reasoning models like GPT-5 (77.83%) and MedGemma-27B (66.66%) and has comparable or better performance on existing medical benchmarks. To ensure a clinically meaningful setting, we conduct a large-scale clinician alignment study, where physicians curate per-case rubrics, grade candidate responses, and provide blinded pairwise preferences of model reasoning. This study provides insights on clinical LLM-as-a-judge and clinical preference-model selection, which can inform the community at large. CLR-voyance has been deployed for 6+ months at a partner public hospital, drafting thousands of reasoning-heavy inpatient notes.

cs.CL↗