arXiv subjects

Haonan Zhang

Publications and source records attributed to Haonan Zhang.

At least 37 records · Page 2Linked to original sources

arXiv2026

Single-laser stimulated Brillouin scattering microscopy

Stimulated Brillouin scattering (SBS) microscopy enables label-free mapping of local viscoelastic properties, but frequency-domain implementations are often limited by uncertainty in the pump-probe frequency-difference axis. We demonstrate an RF-defined single-laser electro-optic-modulation SBS microscope in which the pump and probe are derived from the same optical carrier and their frequency difference is set by an electro-optically generated sideband. This architecture makes laser-frequency noise largely common mode and eliminates optical wavelength tuning during spectral scanning. It achieves Brillouin frequency shift and linewidth precisions of 0.07 MHz and 0.30 MHz, respectively. Comparison with a low-NA reference linewidth indicates a system-level spectral broadening of approximately 3.1 MHz, corresponding to an effective spectral resolution of approximately 3 MHz. Imaging of femtosecond-laser-modified chalcogenide glass resolves MHz-level Brillouin contrasts corresponding to 10^-4-level apparent longitudinal-modulus contrast. This work demonstrates the feasibility of transferring the frequency definition of SBS spectral scanning from optical wavelength tuning to RF-domain control, providing a new conceptual and technical basis for high-precision, high-spectral-fidelity Brillouin imaging.

physics.optics↗

arXiv2026

Channel Adoption Pathways and Post-Adoption Behavior

The rapid growth of digital shopping channels has led many traditional retailers to invest in e-commerce websites and mobile apps. While prior research shows that multichannel customers are more valuable, it overlooks how the motive for adopting a new channel shapes post-adoption behavior. Using transaction-level data from a major Brazilian pet supplies retailer, we study offline-only consumers who adopt online shopping through four pathways: organic adoption, the COVID-19 pandemic, Black Friday promotions, and a loyalty program. Using consumer-level panel data and difference-in-differences estimates, we examine how these pathways are associated with post-adoption spending, profitability, and channel usage. We find that all adopters spend more than comparable offline-only consumers, but their post-adoption behavior differs systematically by adoption pathway. Promotion-driven adopters exhibit patterns consistent with forward buying and lower subsequent profitability, whereas COVID adopters display stronger offline persistence consistent with consumer inertia and habit persistence. These findings suggest that managers may benefit from accounting for adoption-pathway heterogeneity when forecasting customer lifetime value and assessing the breakeven and ROI of promotions designed to induce online adoption.

econ.GN↗

arXiv2026

A Beckmann boundary form of Talagrand's conjecture on the discrete cube

We introduce the Beckmann boundary of a Boolean function \[ \mathsf{B}(f)=\inf_{\operatorname{div} V=Lf}\mathbb E\|V(x)\|_2. \] Here \[ L=\sum_iD_i,\qquad D_i f(x)=\frac{f(x)-f(x^{\oplus i})}{2}, \] and $\operatorname{div} V(x)=\sum_i (V_{i}(x)-V_{i}(x^{\oplus i}))$. This nonlocal quantity is no larger than the usual two-sided, one-sided, colored, optimized colored, or optimized fractional colored boundaries. Nevertheless, every nonconstant Boolean $f$ satisfies \[ \mathsf{B}(f)\gtrsim \operatorname{Var}(f) \sqrt{\log\!\left(1+\frac{1}{\sum_i\operatorname{Inf}_i(f)^2}\right)}. \] We also prove strong one-sided fractional spectral estimates. If $A\subset\{-1,1\}^n$ and \[ h_{A}(x)=\#\{i:x\in A,\ x^{\oplus i}\notin A\}, \] then, for $0<α<1$, \[ \sum_{S\ne\varnothing}|S|^α\widehat{\mathbf 1_{A}}(S)^2 \lesssim_α\mathbb Eω_α(h_{A}), \] where $ω_α(m)=\sqrt m$ for $α<1/2$, $ω_{1/2}(m)=\sqrt m\log(e+m)$, and $ω_α(m)=m^α$ for $α>1/2$. These profiles are sharp, up to $α$-dependent constants, for majority. We also show that the comparison is genuinely nonreversible: an explicit quotient-cube family makes the optimized fractional, and hence optimized colored, boundary exceed $\mathsf{B}$ by a factor $\gtrsim\sqrt{\log n}$. We further obtain a driftless Bernstein-multiplier inequality.

math.CA↗

arXiv2026

Decoding Hidden Deception in Reasoning LLMs: Activation Explainers for Deception Auditing

As LLMs acquire stronger reasoning capabilities, deceptive behavior becomes an increasingly serious safety concern. Existing deception monitors either score visible transcripts or derive scalar probe scores from representation vectors, leaving little inspectable evidence about why a response is suspicious. We introduce STATEWITNESS, an activation explainer for deception auditing. A separate decoder reads a target model's hidden states, then answers natural-language queries or emits structured reports about them. We evaluate STATEWITNESS on two target reasoning LLMs across seven deception datasets. STATEWITNESS reaches 0.916 mean AUROC, a relative gain of 11.6% over the best black-box text monitor and 25.0% over the best activation-probe baseline under the same evaluation protocol. When combined with existing monitors, STATEWITNESS reduces missed deceptive examples in simple threshold ensembles. Beyond scalar detection, the decoder returns query-level answers, schema reports, and token- or sentence-level evidence traces for human inspection. We view this interface as a potential building block for broader interpretability and alignment tools.

cs.CL↗

arXiv2026

DDOR: Delta Debugging for Explainable Overrefusal Testing and Repair

While safety alignment and guardrails help large language models (LLMs) avoid harmful outputs, they can also induce overrefusal, i.e., unwarranted rejection of benign queries that merely appear risky. We present DDOR (Delta Debugging for OverRefusal), a fully automated and explainable framework for overrefusal testing and repair in a black-box setting, where only model inputs and outputs are accessible and internal safety mechanisms remain opaque. DDOR applies delta debugging to localize minimal refusal-triggering fragments (mRTFs) that provide phrase-level, explainable evidence for why a refusal occurs. Conditioned on these mRTFs, DDOR generates diverse, context-rich prompts and performs multi-oracle validation to filter intrinsically unsafe or ambiguous cases, producing scalable and model-specific overrefusal test suites (approximately 1K cases per model). Beyond evaluation, we further leverage localized mRTFs to perform targeted prompt repair, substantially reducing overrefusal while preserving the original intent and maintaining safety on genuinely harmful inputs. Overall, DDOR offers a practical end-to-end solution to both evaluate and mitigate overrefusal, improving LLM usability without sacrificing safety.

cs.SE↗

arXiv2026

Proof of the Holevo--Utkin conjecture on sharp $\ell_p$ norms for zero-sum vectors

Let $d\ge 3$ and $p>0$. Let $\|x\|_p$ denote the $\ell_p$ (quasi-)norm of a $d$-dimensional vector $x$. Holevo and Utkin \cite{HU26} conjectured that for $0<p\le 1$, \[ \min \left\{\frac{\|x\|_p}{\|x\|_2}:\vec{0}\neq x\in\mathbb R^d,\ \sum_{i=1}^d x_i=0\right\} =2^{1/p-1/2}; \] for $1<p<2$, \[ \min \left\{\frac{\|x\|_p}{\|x\|_2}:\vec{0}\neq x\in\mathbb R^d,\ \sum_{i=1}^d x_i=0\right\} = \min\left\{2^{1/p-1/2},\left(\frac{(d-1)^{p/2}+(d-1)^{1-p/2}}{d^{p/2}}\right)^{1/p}\right\}; \] and for $2<q<\infty$ \[ \max\left\{\frac{\|x\|_q}{\|x\|_2}:\vec{0}\neq x\in\mathbb R^d,\ \sum_{i=1}^d x_i=0\right\} = \max\left\{2^{1/q-1/2},\left(\frac{(d-1)^{q/2}+(d-1)^{1-q/2}}{d^{q/2}}\right)^{1/q}\right\}. \] They proved the $d=3$ case in \cite{HU26}. In this paper, we confirm the conjecture of the remaining cases $d\ge 4$.

math.CA↗

arXiv2026

Disentangled Learning Improves Implicit Neural Representations for Medical Reconstruction

Implicit neural representations (INRs) have emerged as a powerful paradigm for medical imaging via physics-informed unsupervised learning. Classical INRs optimize an entire network from scratch for each subject, leading to inefficient training and suboptimal imaging quality. Recent initialization-based approaches attempt to inject population priors into pre-trained networks, yet they rely on high-quality images and often suffer from catastrophic forgetting during fine-tuning. We present DisINR, a novel INR framework that explicitly disentangles shared and subject-specific representations. DisINR introduces a shared encoder-decoder pair and subject-specific encoders, whose features are jointly decoded for image reconstruction. By integrating differentiable forward models, it pre-trains the shared modules directly from limited raw measurements, removing the need for pre-acquired high-quality images. During test-time adaptation, only the subject-specific encoder is optimized, while the shared pair remains frozen, effectively preserving learned priors. Extensive evaluations on three representative medical imaging tasks show that DisINR significantly outperforms state-of-the-art INRs in both reconstruction accuracy and efficiency.

cs.CV↗

arXiv2026

LLM-VA: Resolving the Jailbreak-Overrefusal Trade-off via Vector Alignment

Safety-aligned LLMs suffer from two failure modes: jailbreak (answering harmful inputs) and over-refusal (declining benign queries). Existing vector steering methods adjust the magnitude of answer vectors, but this creates a fundamental trade-off -- reducing jailbreak increases over-refusal and vice versa. We identify the root cause: LLMs encode the decision to answer (answer vector $v_a$) and the judgment of input safety (benign vector $v_b$) as nearly orthogonal directions, treating them as independent processes. We propose LLM-VA, which aligns $v_a$ with $v_b$ through closed-form weight updates, making the model's willingness to answer causally dependent on its safety assessment -- without fine-tuning or architectural changes. Our method identifies vectors at each layer using SVMs, selects safety-relevant layers, and iteratively aligns vectors via minimum-norm weight modifications. Experiments on 12 LLMs demonstrate that LLM-VA achieves 11.45% higher F1 than the best baseline while preserving 95.92% utility, and automatically adapts to each model's safety bias without manual tuning. Code and models are available at https://hotbento.github.io/LLM-VA-Web/.

cs.LG↗

arXiv2026

The Boolean surface area of polynomial threshold functions

Polynomial threshold functions (PTFs) are an important low-complexity class of Boolean functions, with strong connections to learning theory and approximation theory. Recent work on learning and testing PTFs has exploited structural and isoperimetric properties of the class, especially bounds on average sensitivity, one of the central themes in the study of PTFs since the Gotsman--Linial conjecture. In this work we study PTFs through the lens of the Boolean surface area (or Talagrand boundary) \[ \mathbf{BSA}[f]=\mathbb{E}|\nabla f|=\mathbb{E}\sqrt{s_{f}(x)}, \] a natural measure of vertex-boundary complexity on the discrete cube. Our main result is that every degree-$d$ PTF has polylogarithmic Boolean surface area: \[ \mathbf{BSA}[f]\le C_d(\log(en))^{C_d}. \] The proof is based on the PTF Restriction Lemma of Kabanets, Kane, and Lu \cite{KKL2017} and proceeds through a tail bound for the pointwise sensitivity. In particular, it controls all subcritical fractional moments of the sensitivity. We also record a random block partition principle for Boolean surface area and an alternative recursive argument following Kane's work \cite{DK} on average sensitivity, which independently yields the weaker bound \[ \mathbf{BSA}[f]\le \exp(C_d\sqrt{\log n}). \]

cs.CC↗

arXiv2026

Periodic Steady-State Control of a Handkerchief-Spinning Task Using a Parallel Anti-Parallelogram Tendon-driven Wrist

Spinning flexible objects, exemplified by traditional Chinese handkerchief performances, demands periodic steady-state motions under nonlinear dynamics with frictional contacts and boundary constraints. To address these challenges, we first design an intuitive dexterous wrist based on a parallel anti-parallelogram tendon-driven structure, which achieves 90 degrees omnidirectional rotation with low inertia and decoupled roll-pitch sensing, and implement a high-low level hierarchical control scheme. We then develop a particle-spring model of the handkerchief for control-oriented abstraction and strategy evaluation. Hardware experiments validate this framework, achieving an unfolding ratio of approximately 99% and fingertip tracking error of RMSE = 2.88 mm in high-dynamic spinning. These results demonstrate that integrating control-oriented modeling with a task-tailored dexterous wrist enables robust rest-to-steady-state transitions and precise periodic manipulation of highly flexible objects. More visualizations: https://slowly1113.github.io/icra2026-handkerchief/

cs.RO↗

arXiv2026

Token Reduction via Local and Global Contexts Optimization for Efficient Video Large Language Models

Video Large Language Models (VLLMs) demonstrate strong video understanding but suffer from inefficiency due to redundant visual tokens. Existing pruning primary targets intra-frame spatial redundancy or prunes inside the LLM with shallow-layer overhead, yielding suboptimal spatiotemporal reduction and underutilizing long-context compressibility. All of them often discard subtle yet informative context from merged or pruned tokens. In this paper, we propose a new perspective that elaborates token \textbf{A}nchors within intra-frame and inter-frame to comprehensively aggregate the informative contexts via local-global \textbf{O}ptimal \textbf{T}ransport (\textbf{AOT}). Specifically, we first establish local- and global-aware token anchors within each frame under the attention guidance, which then optimal transport aggregates the informative contexts from pruned tokens, constructing intra-frame token anchors. Then, building on the temporal frame clips, the first frame within each clip will be considered as the keyframe anchors to ensemble similar information from consecutive frames through optimal transport, while keeping distinct tokens to represent temporal dynamics, leading to efficient token reduction in a training-free manner. Extensive evaluations show that our proposed AOT obtains competitive performances across various short- and long-video benchmarks on leading video LLMs, obtaining substantial computational efficiency while preserving temporal and visual fidelity. Project webpage: https://tyroneli.github.io/AOT.

cs.CV↗

arXiv2026

Zero-shot Low-Field MRI Enhancement via Diffusion-Based Adaptive Contrast Transport

Low-field (LF) magnetic resonance imaging (MRI) democratizes access to diagnostic imaging but is fundamentally limited by low signal-to-noise ratio and significant tissue contrast distortion due to field-dependent relaxation dynamics. Reconstructing high-field (HF) quality images from LF data is a blind inverse problem, severely challenged by the scarcity of paired training data and the unknown, non-linear contrast transformation operator. Existing zero-shot methods, which assume simplified linear degradation, often fail to recover authentic tissue contrast. In this paper, we propose DACT(Diffusion-Based Adaptive Contrast Transport), a novel zero-shot framework that restores HF-quality images without paired supervision. DACT synergizes a pre-trained HF diffusion prior to ensure anatomical fidelity with a physically-informed adaptive forward model. Specifically, we introduce a differentiable Sinkhorn optimal transport module that explicitly models and corrects the intensity distribution shift between LF and HF domains during the reverse diffusion process. This allows the framework to dynamically learn the intractable contrast mapping while preserving topological consistency. Extensive experiments on simulated and real clinical LF datasets demonstrate that DACT achieves state-of-the-art performance, yielding reconstructions with superior structural detail and correct tissue contrast.

cs.CV↗

arXiv2026

Fourier growth of degree $2$ polynomials

We prove bounds for the absolute sum of all level-$k$ Fourier coefficients for $(-1)^{p(x)}$, where polynomial $p:\mathbf{F}_2^n \to \mathbf{F}_2$ is of degree $1$ or degree $2$.

math.NT↗

arXiv2026

Approximating the operator norm of local Hamiltonians via few quantum states

Consider a Hermitian operator $A$ acting on a complex Hilbert space of dimension $2^n$. We show that when $A$ has small degree in the Pauli expansion, or in other words, $A$ is a local $n$-qubit Hamiltonian, its operator norm can be approximated independently of $n$ by maximizing $|\braket{ψ|A|ψ}|$ over a small collection $\mathbf{X}_n$ of product states $\ketψ\in (\mathbf{C}^{2})^{\otimes n}$. More precisely, we show that whenever $A$ is $d$-local, \textit{i.e.,} $°(A)\le d$, we have the following discretization-type inequality: \[ \|A\|\le C(d)\max_{ψ\in \mathbf{X}_n}|\braket{ψ|A|ψ}|. \] The constant $C(d)$ depends only on $d$. This collection $\mathbf{X}_n$ of $ψ$'s, termed a \emph{quantum norm design}, is independent of $A$, and consists of product states, and can have cardinality as small as $(1+\eps)^n$, which is essentially tight. Previously, norm designs were known only for homogeneous $d$-localHamiltonians $A$ \cite{L,BGKT,ACKK}, and for non-homogeneous $2$-local traceless $A$ \cite{BGKT}. Several other results, such as boundedness of Rademacher projections for all levels and estimates of operator norms of random Hamiltonians, are also given.

quant-ph↗

arXiv2026

Act-Adaptive Margin: Dynamically Calibrating Reward Models for Subjective Ambiguity

Currently, most reinforcement learning tasks focus on domains like mathematics and programming, where verification is relatively straightforward. However, in subjective tasks such as role-playing, alignment techniques struggle to make progress, primarily because subjective reward modeling using the Bradley-Terry model faces significant challenges when dealing with ambiguous preferences. To improve reward modeling in subjective tasks, this paper proposes AAM (\textbf{\underline{A}}ct-\textbf{\underline{A}}daptive \textbf{\underline{M}}argin), which enhances reward modeling by dynamically calibrating preference margins using the model's internal parameter knowledge. We design two versions of AAM that efficiently generate contextually-appropriate preference gaps without additional human annotation. This approach fundamentally improves how reward models handle subjective rewards by better integrating generative understanding with preference scoring. To validate AAM's effectiveness in subjective reward modeling, we conduct evaluations on RewardBench, JudgeBench, and challenging role-playing tasks. Results show that AAM significantly improves subjective reward modeling performance, enhancing Bradley-Terry reward models by 2.95\% in general tasks and 4.85\% in subjective role-playing tasks. Furthermore, reward models trained with AAM can help downstream alignment tasks achieve better results. Our test results show that applying rewards generated by AAM-Augmented RM to preference learning techniques (e.g., GRPO) achieves state-of-the-art results on CharacterEval and Charm. Code and dataset are available at https://github.com/calubkk/AAM.

cs.CL↗

arXiv2025

ORFuzz: Fuzzing the "Other Side" of LLM Safety -- Testing Over-Refusal

Large Language Models (LLMs) increasingly exhibit over-refusal - erroneously rejecting benign queries due to overly conservative safety measures - a critical functional flaw that undermines their reliability and usability. Current methods for testing this behavior are demonstrably inadequate, suffering from flawed benchmarks and limited test generation capabilities, as highlighted by our empirical user study. To the best of our knowledge, this paper introduces the first evolutionary testing framework, ORFuzz, for the systematic detection and analysis of LLM over-refusals. ORFuzz uniquely integrates three core components: (1) safety category-aware seed selection for comprehensive test coverage, (2) adaptive mutator optimization using reasoning LLMs to generate effective test cases, and (3) OR-Judge, a human-aligned judge model validated to accurately reflect user perception of toxicity and refusal. Our extensive evaluations demonstrate that ORFuzz generates diverse, validated over-refusal instances at a rate (6.98% average) more than double that of leading baselines, effectively uncovering vulnerabilities. Furthermore, ORFuzz's outputs form the basis of ORFuzzSet, a new benchmark of 1,855 highly transferable test cases that achieves a superior 63.56% average over-refusal rate across 10 diverse LLMs, significantly outperforming existing datasets. ORFuzz and ORFuzzSet provide a robust automated testing framework and a valuable community resource, paving the way for developing more reliable and trustworthy LLM-based software systems.

cs.SE↗

arXiv2025

Delving into Dynamic Scene Cue-Consistency for Robust 3D Multi-Object Tracking

3D multi-object tracking is a critical and challenging task in the field of autonomous driving. A common paradigm relies on modeling individual object motion, e.g., Kalman filters, to predict trajectories. While effective in simple scenarios, this approach often struggles in crowded environments or with inaccurate detections, as it overlooks the rich geometric relationships between objects. This highlights the need to leverage spatial cues. However, existing geometry-aware methods can be susceptible to interference from irrelevant objects, leading to ambiguous features and incorrect associations. To address this, we propose focusing on cue-consistency: identifying and matching stable spatial patterns over time. We introduce the Dynamic Scene Cue-Consistency Tracker (DSC-Track) to implement this principle. Firstly, we design a unified spatiotemporal encoder using Point Pair Features (PPF) to learn discriminative trajectory embeddings while suppressing interference. Secondly, our cue-consistency transformer module explicitly aligns consistent feature representations between historical tracks and current detections. Finally, a dynamic update mechanism preserves salient spatiotemporal information for stable online tracking. Extensive experiments on the nuScenes and Waymo Open Datasets validate the effectiveness and robustness of our approach. On the nuScenes benchmark, for instance, our method achieves state-of-the-art performance, reaching 73.2% and 70.3% AMOTA on the validation and test sets, respectively.

cs.CV↗

arXiv2025

On the Eldan-Gross inequality

A recent discovery of Eldan and Gross states that there exists a universal $C>0$ such that for all Boolean functions $f:\{-1,1\}^n\to \{-1,1\}$, $$ \int_{\{-1,1\}^n}\sqrt{s_f(x)}dμ(x) \ge C\text{Var}(f)\sqrt{\log \left(1+\frac{1}{\sum_{j=1}^{n}\text{Inf}_j(f)^2}\right)} $$ where $s_f(x)$ is the sensitivity of $f$ at $x$, $\text{Var}(f)$ is the variance of $f$, $\text{Inf}_j(f)$ is the influence of $f$ along the $j$-th variable, and $μ$ is the uniform probability measure. In this note, we give an alternative proof that applies to biased discrete hypercube, and spaces having positive Ricci curvature lower bounds in the sense of Bakry and Émery.

math.FA↗