Searcharxiv⌕ Search

arXiv subjects

Ziheng Chen

Publications and source records attributed to Ziheng Chen.

At least 19 recordsLinked to original sources

When the Cheapest Feasible Router Fails: Auditing Deployment Commitments in Cost-Aware LLM Routing

Cost-aware LLM routing often selects the cheapest policy meeting empirical risk and coverage constraints. Yet a router that passes an audit can violate its fixed service contract on new requests, even when candidate policies are constructed independently. We study this deployment-commitment decision with disjoint construction, audit, and test samples. Across 19 RouterBench task families, the cheapest audit-feasible router deploys in 91.82% of opportunities but violates the held-out contract in 32.22% of deployments. We introduce Menu-Aware Routing Commitment (MARC), which qualifies policies using audit evidence and a serving-cost tolerance to retain the apparent winner, deploy a supported alternative, or defer. Its practical policywise variant (MARC-P) reduces observed conditional failure to 0.73% at 85.79% deployment, with higher serving cost; fixed-margin and formal risk-control procedures provide competitive alternatives. Nested audits show how added evidence changes deployable policy sets, and learned semantic routing exhibits the same commitment failure. These results establish deployment commitment as a distinct operational decision connecting router evaluation, service reliability, and inference cost.

cs.LG↗

Beyond Marginal Risk: ESG and Multidimensional Fragility in Equity Markets

Do ESG characteristics provide financial resilience beyond conventional risk, and is that inference stable to the financial state and ESG score being measured? We study a point-in-time panel of nonfinancial S&P 500 constituents from January 2019 through November 2025. Once ex ante return, volatility, and liquidity exposures are modeled explicitly, the broad downside-resilience pattern attenuates and none of the three primary marginal ESG $\times$ Down tests survives Holm adjustment under either conventional or wild-cluster-bootstrap inference. A post-primary exact decomposition then shows that the negative Return+RVOL coefficient is almost entirely allocated to observations that also breach the CHL trading-cost tail, while the mutually exclusive Return+RVOL-only component is essentially zero. Finally, a paired standardized comparison on an identical sample shows that MSCI's downstream Industry Adjusted Score and upstream Weighted Average Key Issue Score imply materially different fitted associations in adverse market states: at a 5% market decline, the difference is about one probability percentage point per score standard deviation. Sector-time, leave- one-month-out, and observed-change timing diagnostics preserve most of this contrast, although the result remains exploratory and inference is sensitive to multiplicity and market-state support. The evidence therefore favors a narrower interpretation of ESG resilience: the apparent association weakens once conventional risk is modeled, while the remaining relation depends on both the adverse state and the MSCI score field used to represent ESG.

q-fin.MF↗

TRACER: Token ReAssignment for Concept ERasure in Generative Recommendation

Generative recommendation formulates next-item prediction as autoregressive generation over semantic ID (SID) sequences derived from users' historical interactions, making modern recommender systems structurally similar to large language models (LLMs). As privacy and safety concerns grow, these systems increasingly require concept unlearning to remove sensitive or harmful concepts associated with items. However, existing LLM unlearning methods cannot be directly applied to generative recommendation. Unlike word tokens with explicit semantics, SIDs are abstract identifiers that are often shared by both forget and retain items, leading to severe conflicts between concept removal and recommendation utility preservation. To address this challenge, we propose TRACER, an end-to-end concept unlearning framework based on token reassignment. Rather than directly suppressing shared SIDs, TRACER reassigns concept-related items to alternative tokens that better facilitate forgetting while minimizing side effects on retained items. We further introduce a coherence regularizer to preserve semantic consistency among retain items during unlearning. Experiments on real-world recommendation datasets demonstrate that TRACER effectively removes target concepts while substantially better preserving recommendation utility than existing unlearning baselines.

cs.IR↗

Riemannian Deep Learning: Modules, Networks, and Geometries

Deep neural networks on manifold-valued representations have attracted growing interest, but many basic components remain tied to specific manifolds, rely on Euclidean approximations, or require costly and numerically fragile geometric operations. This thesis develops a unified framework for Riemannian deep learning from three complementary perspectives: reusable neural modules, manifold-specific network architectures, and the design of underlying geometries. It generalizes batch normalization from Euclidean spaces and individual manifolds to broad classes of Lie groups and gyrogroups, and extends multinomial logistic regression from Euclidean space to SPD manifolds and then to general Riemannian manifolds. It further develops neural networks for several important geometric representations, including an unconstrained model of hyperbolic space, Busemann-based hyperbolic learning, and full-rank correlation matrices. Finally, it introduces adaptive and computationally efficient Riemannian metrics on SPD manifolds, including learnable Log-Euclidean geometries and fast, stable Cholesky-based geometries. The proposed methods are supported by theoretical analysis and validated through numerical experiments and applications in vision, signal processing, graph learning, and genomics.

cs.LG↗

A-RSI: Adaptive Task Refresh for Zero-Data Multimodal Self-Improvement

Recursive self-improvement requires models not only to acquire useful training experience, but also to decide when their current experience has ceased to be useful. Existing methods often focus on what experience to generate next, while paying less attention to when fresh experience should be introduced. Many also construct training targets through self-consistency or majority voting, making supervision depend on the model's own uncertain predictions. We address these two questions separately: training dynamics determine when to acquire new experience, while the answers to new tasks remain independently checkable. A-RSI monitors reward variation and accuracy in training logs to refresh a task pool when its learning signal declines, without consulting validation or test scores. A static library of model-authored programs produces multimodal questions and derives their reference answers; we specify the task-generation contract, automatically verify chosen properties of the reward and refresh rules, and separately audit the generated answers. The training loop uses no external training examples or human-written answer labels. On the primary eleven-benchmark suite, A-RSI raises the performance average from 57.77 to 61.05 with 800 fresh tasks.

cs.AI↗

Gestalt: Large Multimodal Interplay Model

In this paper, we propose Gestalt, a new paradigm of large multimodal model built around multimodal interplay. Despite rapid advances, large multimodal models are reaching a bottleneck: existing approaches focus primarily on accommodating additional modalities while overlooking the distinct characteristics of each modality and the relations among them. Motivated by the multistage property of human multisensory perception, we propose a multimodal interplay pyramid that organizes multimodal modeling as a progression from modality-specific processing, through cross-modal alignment, to deeper multimodal integration. Guided by this pyramid, Gestalt adopts a unified discrete diffusion framework and an interplay-partitioned architecture, with learnable interplay tokens mediating cross-modal exchange and integration. The pyramid also structures its data organization and training strategy. Strong performance across image generation, multimodal understanding, and text-only evaluation shows that Gestalt significantly improves cross-modal integration while preserving modality-specific information, effectively harnessing the strengths of diffusion-based multimodal models and offering a promising path toward unified multimodal intelligence. Project page: https://GeWu-Lab.github.io/Gestalt.

cs.CV↗

Constant-Curvature Sliced Gromov-Wasserstein for Heterogeneous Cross-Curvature Alignment

Recent advances in representation learning have highlighted the utility of constant-curvature models, such as hyperbolic and spherical spaces, for modeling complex data. Mixed-curvature models further enhance this by integrating multiple constant-curvature components. However, these models typically learn each component space independently because spaces with different curvatures are inherently heterogeneous and lack a unified metric. Consequently, they lack explicit mechanisms to enforce geometric consistency across various spaces. Moreover, the problem of comparing probability distributions across mixed-curvature spaces remains unexplored. To compare distributions on heterogeneous spaces, Gromov-Wasserstein (GW) distances provide a principled framework by aligning their intra-space geometries. Building on this, we propose constant-curvature sliced Gromov-Wasserstein (CCSGW), a novel divergence for aligning distributions supported on heterogeneous constant-curvature spaces. We first introduce the missing geodesic-based one-dimensional projections for spherical spaces, and then extend sliced GW to constant-curvature spaces, enabling efficient and principled comparison across manifolds with different curvatures. This formulation preserves intrinsic geometric relationships while avoiding the high computational cost. We provide theoretical analysis showing that CCSGW controls intrinsic geometric discrepancy across heterogeneous spaces, promoting distribution-level geometric consistency. By integrating CCSGW into existing mixed-curvature learning tasks, including graph anomaly detection, graph node classification, and multimodal learning, we observe consistent performance gains across diverse settings.

cs.LG↗

When Verifiable Counts Depend on Wording: Auditing Wording Robustness in Instruction Following

Verifiable instruction-following benchmarks often express each constraint through one fixed template. We test whether scores remain stable when the operational requirement is unchanged but its wording varies. We introduce WISE, a matched evaluation suite and reporting protocol instantiated on exact word count, keyword inclusion exactly once, and an inclusive 8--12 word range. Across 100 matched tasks, up to thirteen models from seven providers, and repeated generations scored over the complete visible output, wording alone produces substantial compliance shifts. In an avoidance-family panel, five avoidance and exclusion forms fall below the positive baseline, while constructional controls also shift compliance substantially: in the nine-model control panel, compliance is 54.9% for the original positive form, 48.2% for a longer positive form, 36.7% when the target appears later, and 33.8% for AVOID1. A strict JSON-structure probe shows wording sensitivity beyond counting, with a different direction of effect. Effect sizes, failure directions, weakest forms, and model rankings vary across realizations. Under the most disruptive exclusion form, the top-ranked model changes and 24.1% of strictly ordered model pairs reverse. Human validation further shows that unanimous agreement on an exact-count interpretation can coexist with substantially different model behavior. WISE supplements conventional scores with mean and worst-form compliance, wording gaps, failure profiles, and ranking stability.

cs.CL↗

Building Transformation Layers for Riemannian Neural Networks

Recently, deep neural networks on manifold-valued representations have garnered significant attention across various machine learning applications. One recent focus is the generalization of Euclidean fully connected (FC) and convolutional layers to non-Euclidean geometries. However, previous approaches typically focus on a few selected manifolds and rely on specific properties of the target manifold. In contrast, this work proposes a framework for constructing FC and convolutional layers over computationally tractable Riemannian spaces. This framework incorporates several previous FC layers across different geometries as special cases and is instantiated on ten representative manifolds, including three hyperbolic models, five geometries of the symmetric positive definite (SPD) manifold, and two Grassmannian perspectives. Experiments on different manifolds demonstrate the effectiveness and applicability of our approach. Code can be found at https://github.com/GitZH-Chen/RieTrans.

cs.AI↗

Strong convergence rates of stochastic theta Milstein methods for index-1 stochastic differential algebraic equations under non-globally Lipschitz conditions

This paper studies the strong convergence order of structure-preserving stochastic theta Milstein methods for a class of index-$1$ stochastic differential algebraic equations (SDAEs) with time-dependent singular matrices and non-globally Lipschitz coefficients. The singular matrix is allowed to vary in time while preserving a fixed differential algebraic splitting, and the drift and diffusion coefficients may exhibit superlinear growth. By exploiting the index-$1$ algebraic-differential decomposition of the exact solution, we identify the Milstein coefficient of the reduced stochastic differential equation directly in the original SDAE variables and establish the well-posedness and constraint preserving property of the proposed method for $θ\in[1/2,1]$. Under a coupled monotonicity condition and suitable polynomial regularity assumptions, the method is proved to preserve the algebraic constraints at all time levels and to converge with strong order one in the root mean square norm. Numerical experiments confirm the structure-preserving property and the theoretical convergence order.

math.NA↗

A Mechanistic Perspective and Circuit-Guided Difficulty Metric for Unlearning

Machine unlearning is becoming essential for building trustworthy and compliant language models. Yet unlearning success varies considerably across individual samples: some are reliably erased, while others persist despite the same procedure. We argue that this disparity is not only a data-side phenomenon, but also reflects model-internal mechanisms that encode and protect memorized information. We study this problem from a mechanistic perspective based on model circuits--structured interaction pathways that govern how predictions are formed. We propose Circuit-guided Unlearning Difficulty (CUD), a {\em pre-unlearning} metric that assigns each sample a continuous difficulty score using circuit-level signals. Extensive experiments demonstrate that CUD reliably separates intrinsically easy and hard samples, and remains stable across unlearning methods. We identify key circuit-level patterns that reveal a mechanistic signature of difficulty: easy-to-unlearn samples are associated with shorter, shallower interactions concentrated in earlier-to-intermediate parts of the original model, whereas hard samples rely on longer and deeper pathways closer to late-stage computation. Compared to existing qualitative studies, CUD takes a first step toward a principled, fine-grained, and interpretable analysis of unlearning difficulty; and motivates the development of unlearning methods grounded in model mechanisms.

cs.LG↗

LieBN: Batch Normalization over Lie Groups

Manifold-valued measurements are prevalent in various machine learning tasks. Recent advances have extended Deep Neural Networks (DNNs) to operate on manifolds. These extensions have been accompanied by normalization techniques tailored to different geometries, collectively referred to as Riemannian normalization. However, most existing Riemannian normalization methods are either designed for specific manifolds or fail to effectively normalize manifold-valued sample distributions. To address these limitations, we propose LieBN, a framework for Riemannian Batch Normalization (RBN) over Lie groups. Our approach leverages the theoretically convenient left- and right-invariant metrics, which naturally exist in every Lie group, and provides theoretical guarantees for controlling the Riemannian mean and variance. We instantiate LieBN across nine distinct geometries: four on the Symmetric Positive Definite (SPD) manifold, one on the group of rotation matrices, and four on the manifold of full-rank correlation matrices. Notably, among the SPD metrics, we introduce a novel right-invariant metric and extend three existing Lie group structures via matrix power deformation. Experiments on different manifolds validate the effectiveness of our framework. The code is available at https://github.com/GitZH-Chen/LieBN.git.

cs.LG↗

RELTA-SGLD: Relative-Growth Localized Taming for Nonconvex Stochastic-Gradient Langevin Learning

We introduce RELTA-SGLD, a taming scheme that stabilizes superlinear stochastic-gradient updates while reducing unnecessary suppression of the original learning drift. A threshold determines where the taming turns on, while a relative-growth principle derived from the one-step Lyapunov stability condition determines the required taming strength. Together, they produce a lighter $λ$-scale denominator and preserve a nonvanishing far-tail return. As a consequence, we prove polynomial moment stability and first-order stationary accuracy in both $W_1$ and $W_2$ for nonconvex SGLD with superlinearly growing stochastic-gradient oracles, improving the corresponding half-order and quarter-order bounds for comparable stochastic-gradient tamed schemes. On Fashion-MNIST under active stabilization pressure, RELTA improves the mean learning metrics over both untamed SGLD and TUSLA and remains competitive with a tuned AdamW reference. In an ordinary-training regime, its lighter localized denominator reduces unnecessary perturbation of the original update and maintains nearly untamed learning dynamics.

stat.ML↗

Strong order one-half convergence of a coupled tamed Euler--Peano scheme for reflected stochastic differential equations with super-linearly growing coefficients

We study strong numerical approximations for reflected stochastic differential equations in possibly unbounded convex domains with super-linearly growing drift and diffusion coefficients. Under a coupled monotonicity condition and polynomial local Lipschitz assumptions, we first establish the well-posedness of the reflected SDE and derive uniform moment bounds for its solution. We then introduce a coupled tamed Euler--Peano scheme, in which the drift and the squared diffusion coefficient are tamed by a common factor and the resulting Euler--Peano path is corrected through the Skorokhod problem. This common taming factor preserves the drift--diffusion coercivity structure and yields uniform moment estimates for the numerical solution. We prove strong convergence of order $1/2$ for both the constrained state process and the boundary regulator, thereby recovering the standard Euler-type strong order in this reflected setting. Numerical experiments for a reflected stochastic Ginzburg--Landau type system illustrate the constraint preservation of the scheme and support the theoretical convergence rate.

math.NA↗

Discontinuous Prior-Mode Sections and the Geometry of Ambiguity in Intrinsic Image Decomposition

The viral 2015 photograph known as "The Dress" divides observers into two camps because it is ambiguous: the same image colors can be explained either as a blue-black surface under one illuminant or as a white-gold surface under another. We propose a geometric account in which the ambiguity arises from a singularity in intrinsic image decomposition, the inverse problem of separating an observed image into reflectance and illumination. Our central claim is that the prior-mode section, i.e. the prior-preferred decomposition, switches across an ambiguity boundary in image space, and that any smooth learned model can only approximate this discontinuous switch by forming a thin transition layer. This predicts two observable signatures, where $Δ$ is the jump between branches of the prior-mode section and $λ$ is the regularization strength: for inverse decomposers, an albedo Jacobian scaling as $|Δ|/\sqrtλ$; and for forward encoders, the Fernet curvature that blows up on a scale of $1/\sqrtλ$. On CGIntrinsics ($N=1998$ images, $n=2\times 10^7$ pixels), the color-temperature albedo Jacobian of Careaga DPT has partial Spearman correlation $r=0.41$ with dense ground-truth albedo error, compared with $r=0.087$ and $r=0.021$ for brightness and saturation controls. On "The Dress", CLIP ViT-L/14 exhibits a latent curvature peak of $κ=73.03$ at $6473\,\mathrm{K}$, one sampled step from D65 daylight, while a control dress image peaks at $κ=34.75$ with no comparable feature near D65. The same characteristic appears across architectures (U-Net inverter, diffusion inverter, ViT encoder) and datasets (rendered indoor scenes, web photograph), each measured with the observable appropriate to its model class.

math.DG↗

Deterministic Envelopes for Tamed SGLD: Decoupling Stochastic Gradient Noise and Localizing Taming

Stochastic gradient Langevin algorithms often use tamed denominators to stabilize superlinear drifts. This paper shows that when the denominator depends on the current stochastic gradient, the transformed update can have a biased conditional mean even if the original stochastic gradient is unbiased. This creates a stationary mean-shift channel that is absent for deterministic denominators.We propose a structure-preserving framework for designing tamed denominators. The construction keeps the denominator deterministic given the current state, and uses localized deterministic envelopes to avoid unnecessary taming in typical regions. These kernels retain the stabilizing effect of taming while avoiding the bias introduced by a gradient-dependent denominator. Our theory bounds the stationary bias through Euler, envelope, and stochastic-gradient residuals. The analysis also shows why purely local taming rules can lose control in the far tail and motivates a hybrid construction with additional tail protection. Experiments confirm the stationary distortions of random denominators, the bias reduction of deterministic-envelope designs, and the stabilizing effect of the hybrid construction.

stat.ML↗

Deterministic Denominator Design for Localized Tamed Stochastic Gradient Langevin Dynamics

If the denominator in a tamed stochastic gradient Langevin update uses the current stochastic-gradient draw, the conditional mean can be biased even when the stochastic-gradient oracle is unbiased. A state-dependent denominator fixed before that draw removes this coupling. We build practical deterministic denominators from a short pilot run. A log-scale proxy is fitted to the growth score $G_\star(x)=\|b(x)\|/(1+\|x\|)$, and empirical pilot quantiles set the activation thresholds of a local proxy-quantile envelope. The reported proxy-quantile experiments use this local denominator. Separately, we describe a final denominator with a norm-polynomial tail-floor correction that can be used when one wants to certify the global effective-linearity input required by the companion deterministic-envelope Lyapunov theory. We show how proxy and threshold errors enter denominator errors and the resulting stationary observable errors. In the reported experiments, the local proxy-quantile denominator improves over random-denominator tamed SGLD at comparable production cost. It also gives observable behavior close to the $G_\star$-envelope benchmark, without the full-gradient growth-score evaluations required by that benchmark.

stat.ME↗

No Place to Hide: Benchmarking Video Hallucination with Background-Controlled Pairs

We introduce VidPair-Halluc, a new benchmark for evaluating video hallucination in large video models (LVMs) under rigorous and controlled conditions. Unlike previous benchmarks that primarily rely on text-based perturbations or adversarial questions while neglecting the consistency of visual backgrounds, VidPair-Halluc features video pairs with highly similar backgrounds but distinctly different foreground semantics, enabling precise attribution of model errors to genuine hallucination rather than background variation. The benchmark is constructed through PairFlow, a pipeline that leverages recent advances in text-to-image and video generation to systematically compose stories, generate coherent video clips, and assemble them into adversarial pairs. Covering both spatial and temporal reasoning across ten semantic aspects, VidPair-Halluc comprises 1K high-quality adversarial video pairs and 11K spatio-temporal QA pairs with control over background and foreground variations. Evaluations on mainstream LVMs show persistent difficulty with robust fine-grained video understanding in adversarial settings, and code and data are available at the https://jethrojames.github.io/VidPair-Halluc/.

cs.CV↗