arXiv subjects

Xiaolei Xu

Publications and source records attributed to Xiaolei Xu.

8 recordsLinked to original sources

arXiv2026

SafePyramid: A Hierarchical Benchmark for In-context Policy Guardrailing

In real-world applications, guardrails are often expected to identify unsafe user-model interactions according to application-specific safety policies, rather than relying on predefined risk taxonomies. In this work, we study this setting under the paradigm of in-context policy guardrailing, where guardrails predict safety violations based on policy specifications provided in context. To systematically evaluate this capability, we introduce SafePyramid, a safety benchmark comprising 1,000 multi-turn conversations across 10 domains and 3,000 corresponding application-specific policies, which together contain 61,699 distinct natural-language rules. SafePyramid organizes the evaluation into three difficulty levels: L0 evaluates individual-rule understanding, L1 evaluates reasoning over rule dependencies, and L2 evaluates adaptation of full novel policy frameworks defined in context. To ensure benchmark quality, we employ a rigorous multi-stage pipeline to construct and validate the benchmark. Using SafePyramid, we evaluate 10 frontier LLMs and 5 policy-configurable guardrails and find that in-context policy guardrailing remains highly challenging: even the best-performing model, GPT-5.5, exactly identifies the full set of violated rules in only 54.0%, 35.3%, and 12.9% cases on L0, L1, and L2, respectively. These results highlight the limitations of current guardrails and call for stronger in-context policy guardrails that can reliably execute policies, resolve rule dependencies, and adapt to novel policy frameworks.

cs.AI↗

arXiv2026

From Failure Taxonomy to Intervention: A Diagnostic Methodology for Industry-Scale AVLM in Video and Live-Streaming Platform Moderation

Industry-scale video and live-streaming moderation imposes requirements that are difficult to satisfy with generic pretrained public models or external APIs, including adaptation to platform-specific data distributions, policy-specific objectives, and product-level safety constraints. As a result, platforms must undertake internal model development, naturally turning to shared public research for guidance. However, existing multimodal foundation-model studies primarily report architectures, training recipes, data scaling strategies, and benchmark results, but provide less systematic guidance on how failures should be localized and translated into targeted model-development interventions. Interventions are essential because deployment failures are rarely self-explanatory. Similar failures can originate from different causes. Without targeted interventions, improvement reduces to heuristic trial-and-error, where benchmark improvements are weakly attributable, and failures are difficult to trace to their underlying causes. To address this gap, we present a diagnostic methodology for industry-scale Audio-Visual-Language Models AVLM development. The methodology maps model failures into a taxonomy of observable failure signatures and links each class of failure to an intervention space. We instantiate this methodology across the development and alignment lifecycle of an AVLM foundation model for a large-scale video and live-streaming platform. The resulting system supports over 100 regions and is designed for noisy, ambiguous, and highly diverse content drawn from global platform traffic.

cs.LG↗

arXiv2026

Estimating Respiratory Effort from Nocturnal Breathing Sounds for Obstructive Sleep Apnoea Screening

Obstructive sleep apnoea (OSA) is a prevalent condition with significant health consequences, yet many patients remain undiagnosed due to the complexity and cost of over-night polysomnography. Acoustic-based screening provides a scalable alternative, yet performance is limited by environmental noise and the lack of physiological context. Respiratory effort is a key signal used in clinical scoring of OSA events, but current approaches require additional contact sensors that reduce scalability and patient comfort. This paper presents the first study to estimate respiratory effort directly from nocturnal audio, enabling physiological context to be recovered from sound alone. We propose a latent-space fusion framework that integrates the estimated effort embeddings with acoustic features for OSA detection. Using a dataset of 157 nights from 103 participants recorded in home environments, our respiratory effort estimator achieves a concordance correlation coefficient of 0.48, capturing meaningful respiratory dynamics. Fusing effort and audio improves sensitivity and AUC over audio-only baselines, especially at low apnoea-hypopnoea index thresholds. The proposed approach requires only smartphone audio at test time, which enables sensor-free, scalable, and longitudinal OSA monitoring.

cs.SD↗

arXiv2025

Frame-Voyager: Learning to Query Frames for Video Large Language Models

Video Large Language Models (Video-LLMs) have made remarkable progress in video understanding tasks. However, they are constrained by the maximum length of input tokens, making it impractical to input entire videos. Existing frame selection approaches, such as uniform frame sampling and text-frame retrieval, fail to account for the information density variations in the videos or the complex instructions in the tasks, leading to sub-optimal performance. In this paper, we propose Frame-Voyager that learns to query informative frame combinations, based on the given textual queries in the task. To train Frame-Voyager, we introduce a new data collection and labeling pipeline, by ranking frame combinations using a pre-trained Video-LLM. Given a video of M frames, we traverse its T-frame combinations, feed them into a Video-LLM, and rank them based on Video-LLM's prediction losses. Using this ranking as supervision, we train Frame-Voyager to query the frame combinations with lower losses. In experiments, we evaluate Frame-Voyager on four Video Question Answering benchmarks by plugging it into two different Video-LLMs. The experimental results demonstrate that Frame-Voyager achieves impressive results in all settings, highlighting its potential as a plug-and-play solution for Video-LLMs.

cs.CV↗

arXiv2024

Generalized Entropy Theory Investigation of the Relatively High Segmental Fragility of Many Glass-Forming Polymers

We utilize the generalized entropy theory (GET) of glass formation to address one of the most singular and least understood properties of polymer glass-forming liquids in comparison to atomic and small molecule liquids -- the often relatively high fragility of the polymer dynamics on a segmental scale, $m_s$. We first analyze the relation between $m_s$ and the ratio, $S_c^*/ S_c(T_{\mathrm{g}})$. We find that an apparently general nonlinear relation between $m_s$ and $S_c^*/ S_c(T_{\mathrm{g}})$ holds to a good approximation for a large class of polymer models, $m_s \approx 7.9 \exp [0.6S_c^*/ S_c(T_{\mathrm{g}})]$. The predicted ranges of $m_s$ and $S_c^*/ S_c(T_{\mathrm{g}})$ are consistent with experimental estimates for high molecular-mass polymer, oligomeric, small molecule, and atomic glass-forming liquids. In particular, relatively high values of $m_s$ are found for polymers having complex monomer structures and significant chain stiffness. The variation of $m_s$ with molecular mass, chain stiffness, and intermolecular interaction strength can be traced to the variation of $S_c^*$, which is shown to provide a measure of packing frustration defined in terms of the dimensionless thermal expansion coefficient and isothermal compressibility. The often relatively high fragility and large extent of cooperative motion are found in the GET to derive from the often relatively large packing frustration in this class of polymer glass-forming liquids. Finally, we also develop a tentative model of the ``dynamical segmental relaxation time'' based on the GET, in which the polymers on a coarse-grained scale are modeled as strings of structureless ``beads'', as assumed in the Rouse and reptation models of polymer dynamics.

cond-mat.soft↗

arXiv2023

Parallel Emergence of Rigidity and Collective Motion in a Family of Simulated Glass-Forming Polymer Fluids

The emergence of the solid state in glass-forming materials upon cooling is accompanied by changes in both thermodynamic and viscoelastic properties and by a precipitous drop in fluidity. Here, we investigate changes in basic elastic properties upon cooling in a family of simulated polymer fluids, as characterized by a number of stiffness measures. We show that $τ_α$ can be expressed quantitatively both in terms of measures of the material ``stiffness'', $G_p$ and $\langle u^2 \rangle$, and the extent $L$ of cooperative particle exchange motion in the form of strings, establishing a direct relation between the growth of emergent elasticity and collective motion. Moreover, the macroscopic stiffness parameters, $G_p$, $B$, and $f_{s, q^*}$, can all be expressed quantitatively in terms of the molecular scale stiffness parameter, $k_{\mathrm{B}}T / \langle u^2 \rangle$ with $k_{\mathrm{B}}$ being Boltzmann's constant, and we discuss the thermodynamic scaling of these properties. We also find that $G_p$ is related to the cohesive energy density $Π_{\mathrm{CED}}$, pointing to the critical importance of attractive interactions in the elasticity and dynamics of glass-forming liquids. Finally, we discuss fluctuations in the local stiffness parameter as a quantitative measure of elastic heterogeneity and their significance for understanding both the linear and nonlinear elastic properties of glassy materials.

cond-mat.soft↗

arXiv2022

Thermodynamic-Dynamic Interrelations in Glass-Forming Polymer Fluids

We utilize the lattice cluster theory (LCT) of polymer fluids to show that the configurational entropy, enthalpy, and internal energy are all closely interrelated, as suggested by recent measurements by Caruthers and Medvedev, so that the generalized entropy theory (GET) of glass formation, a combination of the LCT and Adam-Gibbs model, can be recast in terms of any of these thermodynamic properties as a matter of convenience. Thermodynamic scaling, a form of density-temperature scaling exhibited by dynamic and some thermodynamic properties, is used to assess which thermodynamic properties are most naturally linked to dynamics, and we explore the origin of this scaling by both direct calculations based on the GET and molecular dynamics simulations of a coarse-grained polymer model. Through a combination of our comprehensive modeling of thermodynamic properties using the LCT and the highly predictive GET model for how the fluid thermodynamics relate to its dynamics, along with simulation results confirming these theoretical frameworks, we obtain insights into thermodynamic aspects of collective motion and the slow $β$-relaxation processes of glass-forming liquids.

cond-mat.soft↗

arXiv2021

Deep learning-based framework for cardiac function assessment in embryonic zebrafish from heart beating videos

Zebrafish is a powerful and widely-used model system for a host of biological investigations including cardiovascular studies and genetic screening. Zebrafish are readily assessable during developmental stages; however, the current methods for quantification and monitoring of cardiac functions mostly involve tedious manual work and inconsistent estimations. In this paper, we developed and validated a Zebrafish Automatic Cardiovascular Assessment Framework (ZACAF) based on a U-net deep learning model for automated assessment of cardiovascular indices, such as ejection fraction (EF) and fractional shortening (FS) from microscopic videos of wildtype and cardiomyopathy mutant zebrafish embryos. Our approach yielded favorable performance with accuracy above 90% compared with manual processing. We used only black and white regular microscopic recordings with frame rates of 5-20 frames per second (fps); thus, the framework could be widely applicable with any laboratory resources and infrastructure. Most importantly, the automatic feature holds promise to enable efficient, consistent and reliable processing and analysis capacity for large amounts of videos, which can be generated by diverse collaborating teams.

eess.IV↗