Searcharxiv⌕ Search

arXiv subjects

Yichi Zhang

Publications and source records attributed to Yichi Zhang.

At least 55 records · Page 3Linked to original sources
arXiv2026

Hyperparameter Transfer Laws for Non-Recurrent Multi-Path Neural Networks

Deeper modern architectures are costly to train, making hyperparameter transfer preferable to expensive repeated tuning. Maximal Update Parametrization ($μ$P) helps explain why many hyperparameters transfer across width. Yet depth scaling is less understood for modern architectures, whose computation graphs contain multiple parallel paths and residual aggregation. To unify various non-recurrent multi-path neural networks such as CNNs, ResNets, and Transformers, we introduce a graph-based notion of effective depth. Under stabilizing initializations and a maximal-update criterion, we show that the optimal learning rate decays with effective depth following a universal -3/2 power law. Here, the maximal-update criterion maximizes the typical one-step representation change at initialization without causing instability, and effective depth is the minimal path length from input to output, counting layers and residual additions. Experiments across diverse architectures confirm the predicted slope and enable reliable zero-shot transfer of learning rates across depths and widths, turning depth scaling into a predictable hyperparameter-transfer problem.

cs.LG↗
arXiv2026

HijackKV: New Threat in Position-Independent KV Cache Reuse

Key-Value (KV) cache reduces inference latency in large language models (LLMs). Traditional prefix-based reuse has low cache hit rates across inference requests because it requires exact token and position matches. To improve efficiency, recent system optimizations introduce position-independent KV reuse, allowing KV cache to be reused whenever identical text chunks appear, regardless of their position in the sequence. We show this design introduces a new threat, KV Cache Hijacking. Since KV caches are retrieved by token match but encode the context in which they were originally computed, the KV tied to a benign-looking token chunk may encode an attacker-controlled prefix. When later reused in a victim query, this contaminated KV silently hijacks the model's behavior, even if no attacker-controlled text appears in the input. We introduce HIJACKKV, the first attack framework that systematically exploits this vulnerability, demonstrating its severity and practicality. HIJACKKV optimizes an attacker-controlled prefix, so that the KV computed for a subsequent common benign text encodes the attacker's goal, while the text remains unchanged for future cache hits. HIJACKKV achieves an average 94% success rate in a single attempt, remains effective under realistic constraints including low hit rates (10%) and frequent recomputation (50%), persists over multi-turn interactions, and transfers across models in black-box settings. We further provide design insights for building secure KV reuse systems.

cs.CR↗
arXiv2026

Knowledge before Reasoning: EC-Reason-Bench, a Training-Free Diagnostic Benchmark for LLM Enzyme Classification

Enzyme function prediction is a hierarchical, knowledge-intensive form of protein function classification. Existing benchmarks expose an anomaly: general LLMs often get the coarse first level right, yet once asked for a complete EC number their accuracy at levels two through four drops to almost zero, while specialized models and tools stay usable. We propose EC-Reason-Bench, a training-free, diagnostic evaluation protocol built to answer two questions: why general LLMs score close to nothing on EC number prediction, and how much of that loss can be recovered without updating a single weight. We break enzyme classification ability into four orthogonal levers that can each be measured on their own: output structure, external knowledge, reasoning structure, and reasoning robustness. We test each lever with an inference-time method against a shared zero-shot baseline reproducing previously reported near-zero performance. Experiments with several strong reasoning LLMs yield four main findings. First, external knowledge is decisive and must precede reasoning: uniformly low closed-book performance rises sharply with open-book access, narrowing model gaps. Second, in closed-book settings, whether cascading and chain-of-thought help or hurt depends on a model's tendency to abstain. Third, once evidence is available the aggregate score of the best LLM setting is indistinguishable from simply voting the EC numbers of the nearest retrieved neighbors; that tie is an artifact of averaging, and it hides a large gain on adversarial evidence set against an equally large loss on multi-functional enzymes. Reasoning over evidence therefore acts as an arbiter of conflicting neighbors rather than as a source of knowledge, and no single-number leaderboard can see it. Fourth, accuracy obeys a law of homology availability.

cs.CL↗
arXiv2026

Decoder-Guided Lossy Contour Coding Via Anchor Refinement

Object contours serve as compact structural priors for many receiver-side vision tasks such as image super-resolution, edge-conditioned generation, and machine vision. When such tasks are deployed over a bandwidth-limited channel, the sender transmits the high-quality object contour as structural side information to guide reconstruction at the receiver, while-to save bandwidth-only a low-quality reference such as a downsampled image or base-layer reconstruction is delivered. As a result, the decoder can already extract a coarse contour from this reference at no transmission cost, creating an encoder-decoder asymmetry: the fine contour must be coded and sent, yet a free coarse version is available at the decoder. This asymmetry is ignored by existing contour codecs such as JBIG2 and chain coding, which are lossless, symmetric, and offer no rate-distortion control, leading to high bitrates. In this paper, we propose a coarse-to-fine contour coding framework that models a high-quality contour as a structured geometric refinement of the decoder-available coarse contour. The encoder extracts ordered anchors along the fine contour and performs adaptive anchor skipping under a distortion constraint. The decoder then reconstructs the contour by using the coarse prior to guide anchor connectivity. This formulation enables lossy contour compression with an explicit rate-distortion trade-off. Experiments show 54.5%-66.9% bitrate reduction over methods without decoder-side guidance, and up to 5 times savings over JBIG2, while preserving high geometric accuracy.

eess.IV↗
arXiv2026

High-fidelity multiqubit gates with Rydberg atoms via level-crossing-free Rapid adiabatic passage

We propose a rapid adiabatic passage (RAP) scheme based on level-crossing-free pulses for deterministic generation of multiqubit entangled states in Rydberg atom systems. Unlike conventional RAP protocols that rely on level crossings, our approach uses an antisymmetric Rabi frequency and an even-symmetric detuning, enabling robust population transfer without passing through any level crossing. By exploiting the Rydberg blockade effect, the protocol prepares entangled states directly from an initial product state. Specifically, two sequential RAP pulses separated by a pi_g pulse generate two-qubit Bell states, three-qubit W states, four-qubit GHZ states, and six-qubit honeycomb W states. Numerical simulations show that the fidelities exceed 0.9997 for the Bell and three-qubit W states, reach 0.997 for the four-qubit GHZ state, and surpass 0.9995 for the six-qubit honeycomb W state. The scheme demonstrates excellent robustness against pulse parameter fluctuations, with fidelities remaining above 0.99 under +/-5% parameter variations. This work provides a simple, efficient, and robust method for entangled-state preparation in neutral-atom quantum information processing.

quant-ph↗
arXiv2026

OmniPhys: Knowledge-Graph-Driven Benchmarking and Collective Optimization for Physical Commonsense in Text-to-Image Generation

While text-to-image models exhibit remarkable visual fidelity, they frequently violate fundamental physical commonsense. Existing benchmarks often rely on coarse-grained descriptions, failing to diagnose the mastery of specific physical principles. Moreover, the high stochasticity of generative processes causes current prompt optimization methods to suffer from gradient hallucinations, where optimizers are misled by transient visual artifacts rather than systemic flaws. To address these challenges, we introduce OmniPhys, a rigorous benchmark of 1,551 samples grounded in a Physical Knowledge Graph. By aligning PhET simulations with standard curricula, OmniPhys operationalizes a knowledge-to-scenario pipeline that performs diagnostic stress tests via a dual-path verification protocol. We further propose OmniPrompt, an iterative framework that treats physical alignment as a discrete optimization problem. For each query, OmniPrompt aggregates K stochastic images into a per-query feedback buffer. Across training, it further merges feedback from batches of B queries before each meta-policy update, filtering seed and query-local noise. Evaluations across 12 representative text-to-image models reveal universal physical bottlenecks. Results demonstrate that OmniPrompt significantly enhances physical consistency across diverse backbones, proving the transferability and efficacy of our evolved meta-policies. The code and data are available at https://github.com/zjukg/OmniPhys

cs.CV↗
arXiv2026

Universality Reconsidered: Rethinking the Validation of Foundation Models for General-Purpose 3D Medical Segmentation

Foundation models have emerged as a transformative paradigm in 3D medical imaging, with the promise of unified quantitative analysis across diverse targets and imaging modalities. Yet the prevailing conception of universality remains incomplete. Current models are predominantly developed and evaluated on datasets largely concentrated around a limited set of imaging modalities and anatomical regions. In this Perspective, we evaluate representative 3D segmentation foundation models using paired whole-body structural and functional imaging data. Our analysis reveals a substantial gap between benchmark-reported performance and real-world generalization, with marked degradation on previously unseen data and particularly severe failures on functional imaging modalities. These findings suggest that current foundation models remain far from achieving true universality. We argue that progress requires not only scaling models and datasets, but also a reconsideration of how universality is defined and validated, extending evaluation beyond regional structural benchmarks toward whole-body structural and functional imaging. Our observations highlight the need to distinguish benchmark success from genuine clinical generalization. Bridging this gap will be essential for translating foundation models from controlled evaluation settings to real-world medical practice.

cs.CV↗
arXiv2026

X-Translator: A Real-Time Multilingual Speaker-Aware Speech-to-Speech Translation System

Real-time speech-to-speech translation (S2ST) systems must balance translation quality, latency, speech naturalness, and speaker consistency. Publicly documented S2ST systems have advanced direct, multilingual, streaming, and expressive modeling, while proprietary products and APIs increasingly expose real-time translation capabilities to users. However, practical deployment remains challenging for open and reproducible systems, especially in long-form and multi-speaker conversations where partial ASR hypotheses are unstable, turn boundaries are ambiguous, and target speech must be generated with an appropriate speaker prompt. We present X-Translator, a low-cost modular cascaded S2ST system that combines streaming ASR, machine translation, and prompt-conditioned TTS through a session-level runtime controller. The system uses incremental segment commitment to convert unstable ASR streams into translation-ready units, and an online speaker prompt manager to bind source speech spans to speaker-specific voice prompts for synthesis. We evaluate translation, speech quality, and latency with OpenSTBench, compare against proprietary speech translation APIs as behavioral baselines, measure long-form voice stability, evaluate speaker preservation in multi-speaker conversations, and assess multilingual translation quality. X-Translator provides an open platform for understanding the practical trade-offs of deployment-oriented S2ST. Code and demo are available at https://github.com/zhaoyx239/X-Translator.

eess.AS↗
arXiv2026

Orca: The World is in Your Mind

We introduce Orca, an initial instantiation of a general world foundation model. Orca learns a unified world latent space from multimodal world signals and exposes it through multimodal readout interfaces. Rather than optimizing isolated next-token, next-frame, or next-action prediction, we are centered on Next-State-Prediction modeling, offering a unified state-transition modeling route toward understanding, predicting, and acting upon the world. Orca learns through two complementary paradigms: unconscious learning captures dense natural state transitions from continuous videos, and conscious learning models sparse meaningful state transitions by language-described events and VQA supervision. For pre-training, we construct a large-scale world-learning inventory data, including 125K hours of video data and 160M event annotations. After pre-training, Orca learns a unified world latent space. To examine whether the learned latent supports downstream, we evaluate it by three representative downstream readouts: text generation, image prediction, and embodied action generation. Orca's backbone is frozen, and only the lightweight modality-specific decoders are trainable. Experiments show the scalability of the proposed paradigm and verify that stronger world latent enables stronger downstream readouts. Orca outperforms similar-sized specialized baselines. These results show that Orca, as a general world foundation model, presents a promising approach to understanding, predicting, and acting upon the world. Finally, we discuss the current limitations, aiming to provide useful insights and inspiration for the community.

cs.CV↗
arXiv2026

Empirical stratification for predictive treatment effect heterogeneity with post-treatment variables

Post-treatment variables (PVs), such as intercurrent events, treatment noncompliance, and behavioral responses to treatment, provide information about individuals' post-treatment responses and may help characterize heterogeneity in treatment effects on the primary outcome. This paper develops an empirical stratification framework to study the treatment effect heterogeneity across baseline-predicted PV response profiles. Specifically, we construct empirical scores from baseline-covariate predictions of potential PV responses and use these scores to define empirically accessible subgroups for treatment effect evaluation. The resulting empirical-stratum treatment effects (ETEs) quantify treatment effects across these baseline-predicted PV response profiles, and are identifiable under a standard set of assumptions in causal inference. We further introduce projected ETE curves and develop efficient-influence-function-based estimators that allow flexible nuisance estimation. We clarify the distinction and connection of our framework to principal stratification analysis, which address two inferential questions. We evaluate the proposed framework through simulation studies and illustrate its use in two real-world applications.

stat.ME↗
arXiv2026

Auditing Belief-Conditioned LLM Agents in Hidden-Information Social Deduction Games

Evaluating LLM agents in hidden-information multi-agent settings is hard: final outcomes are high-variance and rarely reveal why an agent decided as it did. We study this in a 9-player Werewolf environment where agents act under strict, code-level information isolation, and we build an auditable framework that maintains an external belief state over hidden roles, logs belief updates and belief-action deviations as structured evidence, and supports a defensive offline improvement loop that reviews bad cases before any strategy change. Across 1,080 frozen games spanning belief-disabled, active-belief, kernel-ablation, camp-restricted, consumption-policy, and high-load arms, and including a seed-paired A0/A1 comparison, the active-belief condition is associated with substantially better good-side outcomes: in the 200-seed A0/A1 comparison the good-side win rate rises from 0.205 to 0.390 (paired McNemar $χ^2 = 16.4$, $p < 0.001$), with fewer irreversible witch-poison errors. We do not, however, attribute this shift to belief content. Direct action-belief consistency is low ($\approx 0.21$), and giving belief only to the werewolves helps the good side more than giving it only to the good side, which argues against a simple holder-benefit account; we therefore report the effect as an association and treat its mechanism as unresolved. The contribution is the audit framework itself: it makes the effect measurable, exposes low direct action-belief consistency, rejects an unreliable forced-consumption intervention with evidence, and separates strategy effects from load confounds. We accordingly position external belief in high-noise hidden-information games primarily as an auditable cognitive baseline that also carries decision-relevant signal, turning opaque agent behavior into replayable evidence for safer, controlled iteration.

cs.MA↗
arXiv2026

Design of an Electrically Tunable Microtoroid for Frequency Selection of Polarization-Entangled Photons

Encoding quantum information into discrete optical frequencies, or "frequency bins," uses different colors of light as additional information channels, allowing each photon to carry more information than polarization alone. We present a computational design for an electrically tunable silica microtoroid that selects desired frequency channels after a polarization-entangled photon pair has been generated without disturbing the photons' polarization entanglement. In the proposed architecture, the 750 nm signal photon passes through the microtoroid, while its entangled 880 nm partner bypasses the resonator and serves as a reference for the selected frequency channel. The principal challenge is resonator birefringence: because horizontally and vertically polarized light resonate at slightly different frequencies, the selected frequency can reveal the photon's polarization state and weaken the quantum correlation between the photon pair. We solve this problem by adding a small lithium-niobate tuning element controlled with a single applied voltage. The voltage shifts the resonator so that it responds almost identically to horizontally and vertically polarized light, reducing the remaining mismatch to only 0.286 optical linewidths across nine frequency channels. The photons remain strongly entangled after passing through the device, with a concurrence of C = 0.969, a Bell-state fidelity of F = 0.981, and a Bell parameter of S_max = 2.785. If the relative timing between the frequency channels is also controlled, the same device can generate a nine-channel polarization-frequency hyperentangled state with an effective dimension of K = 8.97. This computational design provides a compact, electrically tunable bridge between polarization-entangled photon sources and future high-capacity quantum photonic systems.

quant-ph↗
arXiv2026

UniTacVLA: Unified Tactile Understanding and Prediction in Vision Language Action Models

Vision-language-action (VLA) models have achieved strong performance in many robotic manipulation tasks, yet remain limited in contact-rich dexterous manipulation. To overcome this limitation, recent vision-tactile-language-action (VTLA) methods incorporate tactile sensing into VLA models to provide direct contact information. However, they typically treat tactile signals as passive auxiliary inputs, making it difficult to model tactile semantics and future physical interactions. To this end, we propose a unified tactile learning framework for contact-rich manipulation that models tactile signals as dynamic interaction cues for both contact understanding and prediction. Specifically, we construct a unified tactile latent space and jointly model current tactile states and future contact changes through tactile chain-of-thought reasoning and coarse-to-fine future tactile prediction, thereby forming a state-aware and dynamics-aware tactile prior. Based on this prior, we introduce a tactile-action mixed controller that combines real-time and predicted tactile feedback to refine low-frequency action chunks with high-frequency corrections. Real-world experiments on four categories of contact-rich tasks, including adjustment, insertion, wiping, and assembly, under both clean and externally perturbed settings, show that our method improves success rate, manipulation accuracy, and contact robustness over existing methods, demonstrating its effectiveness in dexterous physical interaction.

cs.RO↗
arXiv2026

Rethinking Forgery Attacks on Semantic Watermarks in Black-Box Settings: A Geometric Distortion Perspective

Recent studies have shown that semantic watermarks, which embed information into the initial noise of latent diffusion models (LDMs), are vulnerable to black-box forgery attacks. However, existing methods primarily rely on empirical evidence and lack a rigorous theoretical understanding of the conditions under which such attacks succeed or fail. To bridge this gap, we rethink the nature of such attacks through the lens of rate-distortion in the latent space. Our analysis identifies an irreducible distortion floor due to structural mismatches between proxy and target models, which fundamentally limits the fidelity of forged watermarks. We further characterize this distortion as structured geometric deviations on the latent manifold, in the form of global drift and local deformation rather than stochastic noise. Leveraging these insights, we propose a scheme-agnostic detection method that distinguishes forged samples before watermark verification. Extensive experiments demonstrate the effectiveness of our method across diverse black-box scenarios, while preserving robustness to common distortions.

cs.CR↗
arXiv2026

EgoSAT: A Comprehensive Benchmark of Egocentric Streaming Interaction Understanding

We introduce EgoSAT, the first comprehensive benchmark for egocentric video reasoning in streaming settings, designed to evaluate the capabilities of modern vision-language models (VLMs). The benchmark targets streaming interaction understanding, where video frames arrive sequentially and models must continuously interpret evolving visual context. EgoSAT unifies several previously distinct tasks within a single streaming framework. In this formulation, queries about completed events correspond to retrospective reasoning, queries about ongoing activities require online understanding, and queries about future actions involve prospective anticipation. This unified setting requires models to reason about the past, present, and future while operating under the constraint that only previously observed frames are available. EgoSAT contains 1,997 unique videos spanning 165 hours of egocentric footage and around 4,800 high-quality question-answer pairs, carefully designed to probe reasoning across varying temporal contexts. Using this benchmark, we evaluate a diverse set of both open-weight and closed-weight VLMs, providing a systematic assessment of their ability for streaming interaction understanding. By distinguishing answerability and conducting diagnostics on confidence of models, we find existing models not only struggle with prospective and retrospective modeling, but also exhibit severe mis-calibration: confidence often fails to track inherent answerability, leading to dangerous "confidently wrong" behaviors. Project page: https://leiyj23.github.io/EgoSAT/

cs.CV↗
arXiv2026

Mind the Intention: Task-Aware Backdoor Attacks for Forecast-Driven Distribution Network Operations

Accurate distributed energy resources (DERs) forecasting is critical for downstream optimal operations. However, such forecast-based operation can be highly vulnerable to cyberattacks. While existing research mainly focuses on adversarial attacks, we pivot to a more controllable and persistent threat: backdoor attacks. In time series forecasting, a backdoored model generates an attacker-specified target pattern whenever a trigger is embedded in historical inputs. This paradigm naturally fits the entire DER forecast-optimization-operation chain. In this paper, we investigate whether and how backdoor attacks can compromise distribution network operations and propose GridTroj, a unified backdoor framework tailored for this scenario. Unlike standard time series backdoor approaches that train a poisoned model to match a predefined target only in terms of forecasting error, GridTroj explicitly incorporates the attacker's intention and optimizes the attack toward operational disruption. Specifically, GridTroj coordinates two key modules. The Intention Planner designs operation-damaging targets and poisoning strategies, while the Backdoor Realizer constructs the corresponding network architecture and training strategy to learn the trigger-target association. Experiments on three downstream optimization tasks demonstrate that GridTroj can effectively compromise grid operations and outperforms existing baselines. Our code is available at https://github.com/YuxuanCEE/GridTroj.

cs.CR↗
arXiv2026

A Pilot Study of Mildly Recycled Pulsars: A Case Study of PSR J2338+4818

Mildly recycled pulsars are neutron stars partially spun up through relatively short mass-transfer phases, typically with massive carbon-oxygen (CO) or oxygen-neon-magnesium (ONeMg) white dwarf companions. PSR J2338+4818, a mildly recycled pulsar, was discovered with the Five-hundred-meter Aperture Spherical Telescope (FAST). As a pilot study on the formation and evolutionary pathways of mildly recycled pulsars, we present the updated timing solution for PSR J2338+4818 and examine its single pulses and scintillation properties. Aided by the sensitivity of FAST, the single pulses of PSR J2338+4818 were systematically studied. 27,228 single pulses with S/N > 7 have been detected in our observations. For the FAST ultra-wideband observation on MJD 61045, the receiver was still in the technical commissioning phase, and then only a preliminary single-pulse search was performed. Pulse nulling was examined using a Markov Chain Monte Carlo (MCMC) method, but no evidence for nulling was found. The possible long-term nulling reported by previous studies did not occur in any of our observations in either the 1.0 to 1.5 GHz band or the 300 to 600 MHz band. Interstellar scintillation is evident in our observations. The measured scintillation timescales and bandwidths range from 2.93 to 25.26 minutes and 1.68 to 27.41 MHz, respectively. In all observations, no clear scintillation arc was found in the secondary spectra of PSR J2338+4818.

astro-ph.HE↗
arXiv2026

PRISMR: Overcoming Parse Collapse in Multimodal Listwise Ranking via Parameterized Representation Internalization

Generative listwise ranking with Large Multimodal Models (LMMs) aims to capture global list context in a single forward pass, but its effectiveness degrades in long-context multimodal scenarios. We identify a recurring failure mode, parse collapse, where the autoregressive decoder produces fluent yet incomplete rankings by silently omitting candidates and terminating early. This failure stems from limited context utilization rather than simple formatting mistakes, making prompt engineering and constrained decoding insufficient. We propose PRISMR (Parameterized Representation Internalization for Semantic Multimodal Ranking), a framework that replaces transient in-context list processing with parametric structural conditioning. PRISMR uses a lightweight hypernetwork to encode multimodal candidates in parallel and generate item-specific LoRA weights, which are synthesized into an instance-specific adapter for a LMM. This paradigm enables more robust internalization of list structure while preserving the base model. We further introduce a large-scale multimodal review-ranking benchmark for evaluation. Experiments demonstrate that PRISMR substantially reduces parse collapse, improves listwise ranking performance, and transfers effectively across domains and instruction-tuned backbones.

cs.AI↗