arXiv subjects

Yichuan Chen

Publications and source records attributed to Yichuan Chen.

7 recordsLinked to original sources

arXiv2026

Directional Optimal Sub-Gamma Scales for Infinitely Divisible Laws

Fixing the quadratic proxy in a sub-gamma bound at the true variance leaves a scale to optimize, and a two-sided infinitely divisible law generally requires different scales in the two directions. For a centered law with finite nonzero variance, we normalize its Kolmogorov canonical measure and multiply the resulting variable by an independent Beta(1,2) variable. The signed remainder obtained in this way gives exact variational formulas for the right and left scales. We prove that a directional scale vanishes exactly when the Levy measure has no jumps in that direction, establish reflection, scaling, convolution, Levy-time, and opposite-jump perturbation rules, and recover the Levy triplet from the remainder law. The formulas give the two Gamma scales for bilateral Gamma laws and, for centered Skellam laws, the exact transition points p_+=(2+sqrt(3))/4 and p_-=(2-sqrt(3))/4 between local and interior control; when positive jumps are rare, the right scale is asymptotic to 1/log(1/p). The pair therefore records jump direction and the mechanism that controls the variance-exact sub-gamma pole.

math.PR↗

arXiv2026

Minimal Radial Sub-Gamma Envelopes for Infinitely Divisible Random Vectors

Let X be a centered infinitely divisible random vector with finite second moment and covariance matrix Sigma. We define the radial pole C_X(t) as the smallest scale in a right sub-gamma bound for whose quadratic proxy is fixed at the true variance t^T Sigma t. A canonical directional measure and an independent Beta(1,2) multiplier give an exact variational formula for C_X. The resulting extended-valued function is positive homogeneous and is pointwise least among all homogeneous denominators compatible with the covariance quadratic form. We prove linear-map, convolution, and Levy-time rules, and show that the full family of directional remainders determines the law of X. Geometrically, C_X lies between the Minkowski functional of the moment-generating-function domain and one third of the positive support function of the Levy measure; the upper constant is sharp, and the zero set is a polar cone. The pole need not be subadditive. It is continuous on the sphere under global exponential moments and positive-definite covariance, whereas finite variance alone permits a jump from zero to infinity in nearby directions. For additive gamma-ray models, C_X equals the domain gauge and has a finite-polytope formula.

math.PR↗

arXiv2026

One Framework for All: Cross-Modal Membership Inference for Generative Models

Large generative models across text-to-text, text-to-image, and image-to-text modalities have been shown to pose significant privacy risks. One fundamental threat is membership inference attacks (MIA), which aim to determine whether a given data point was used in a model's training set. Although prior work has investigated MIAs against these three classes of generative models, existing approaches treat them in isolation and are not cross-applicable, thereby limiting their real-world utility. To address this limitation, we present the first comprehensive study of a unified membership inference framework that applies across text-to-text, text-to-image, and image-to-text modalities. Our approach is grounded in a key modality-agnostic observation: the output distribution of a generative model can approximate its training data distribution. Leveraging this property, we model the distributions of model-generated outputs and auxiliary non-member samples in a shared embedding space, and perform membership inference via likelihood ratio testing. We conduct extensive experiments in a strict black-box setting under both partial-knowledge and zero-knowledge threat models, and evaluate membership inference against both fine-tuning and pre-training data. Experimental results demonstrate our approach's superior performance in comparison to existing state-of-the-art methods, which are typically optimized for a single model class.

cs.LG↗

arXiv2026

Safety Context Injection: Inference-Time Safety Alignment via Static Filtering and Agentic Analysis

Large Reasoning Models (LRMs) improve performance on complex tasks, but they also make safety control harder at deployment time. In black-box settings, defenders cannot modify model weights and must instead intervene at inference time. This setting creates three practical challenges: harmful intent may be hidden by educational or role-play framing, deep safety analysis can introduce non-trivial latency, and long adversarial contexts can dilute the local cues that simpler filters rely on. These challenges can expose an apparent thinking--output gap, where the model appears cautious during reasoning but still produces an unsafe final answer. To address this problem, we propose Safety Context Injection (SCI), an inference-time framework that separates safety assessment from task generation and prepends a structured external risk report as injected safety context for the protected model. The framework is instantiated in two complementary variants: Static Model Filtering (SMF), a lightweight one-pass guard for fast deployment, and Dynamic Agents Filtering (DAF), an agentic-loop-based analyzer that iteratively gathers and synthesizes evidence for ambiguous or long-context attacks. Across AdvBench and GPTFuzz, spanning base and reasoning models under five jailbreak families, both variants reduce attack success rate and toxicity in the evaluated settings. SMF offers an efficient low-latency option, while DAF is more effective when harmful intent is semantically disguised or dispersed across long contexts.

cs.CR↗

arXiv2025

Unveiling prethermalization and thermal processes through the simplest one-dimensional topological model

Drawing on classical thermodynamic principles-such as the equipartition of energy and entropy maximization-extensive research has shown that the evolution of optical power in multimode optical systems tends toward a Rayleigh-Jeans distribution at thermal equilibrium. Understanding of the processes associated with the thermalization dynamics are of fundamental importance in analyzing and controlling such complex systems. In this work, we utilize a one-dimensional Su-Schrieffer-Heeger lattice as the simplest topological model to investigate the thermalization process of multiband systems in both topologically trivial and nontrivial regimes. Specifically, we identify that thermalization develops in three stages: (i) out-of-equilibrium dynamics, (ii) prethermal stage and (iii) final thermalization. Each individual band constitutes a subsystem that prethermalizes to the Rayleigh-Jeans distribution predicted from its power and internal energy. We find that this leads to a continuously varying prethermalization that eventually relaxes to the final thermal state (a dynamically evolving prethermal state). The presence of topological edge states can accelerate the thermalization process, although prethermal states exist both in the topologically trivial and nontrivial regimes. Factors such as bandgap width, temperature and nonlinearity that can influence the thermalization dynamics are examined in detail. Our work may offer valuable physical insights into understanding and controlling the thermalization process in multiband optical systems, paving the way for more efficient manipulation of light in complex settings.

physics.optics↗

arXiv2025

From Thinking to Output: Chain-of-Thought and Text Generation Characteristics in Reasoning Language Models

Recently, there have been notable advancements in large language models (LLMs), demonstrating their growing abilities in complex reasoning. However, existing research largely overlooks a thorough and systematic comparison of these models' reasoning processes and outputs, particularly regarding their self-reflection pattern (also termed "Aha moment") and the interconnections across diverse domains. This paper proposes a novel framework for analyzing the reasoning characteristics of four cutting-edge large reasoning models (GPT-o1, DeepSeek-R1, Kimi-k1.5, and Grok-3) using keywords statistic and LLM-as-a-judge paradigm. Our approach connects their internal thinking processes with their final outputs. A diverse dataset consists of real-world scenario-based questions covering logical deduction, causal inference, and multi-step problem-solving. Additionally, a set of metrics is put forward to assess both the coherence of reasoning and the accuracy of the outputs. The research results uncover various patterns of how these models balance exploration and exploitation, deal with problems, and reach conclusions during the reasoning process. Through quantitative and qualitative comparisons, disparities among these models are identified in aspects such as the depth of reasoning, the reliance on intermediate steps, and the degree of similarity between their thinking processes and output patterns and those of GPT-o1. This work offers valuable insights into the trade-off between computational efficiency and reasoning robustness and provides practical recommendations for enhancing model design and evaluation in practical applications. We publicly release our project at: https://github.com/ChangWenhan/FromThinking2Output

cs.CL↗

arXiv2021

Reinforcement Learning Based Optimal Camera Placement for Depth Observation of Indoor Scenes

Exploring the most task-friendly camera setting -- optimal camera placement (OCP) problem -- in tasks that use multiple cameras is of great importance. However, few existing OCP solutions specialize in depth observation of indoor scenes, and most versatile solutions work offline. To this problem, an OCP online solution to depth observation of indoor scenes based on reinforcement learning is proposed in this paper. The proposed solution comprises a simulation environment that implements scene observation and reward estimation using shadow maps and an agent network containing a soft actor-critic (SAC)-based reinforcement learning backbone and a feature extractor to extract features from the observed point cloud layer-by-layer. Comparative experiments with two state-of-the-art optimization-based offline methods are conducted. The experimental results indicate that the proposed system outperforms seven out of ten test scenes in obtaining lower depth observation error. The total error in all test scenes is also less than 90% of the baseline ones. Therefore, the proposed system is more competent for depth camera placement in scenarios where there is no prior knowledge of the scenes or where a lower depth observation error is the main objective.

cs.CV↗