arXiv subjects

Bo Chen

Publications and source records attributed to Bo Chen.

At least 55 records · Page 3Linked to original sources

arXiv2026

Metamorphic Testing for Clinical ML Models: A Framework Proposal and Pilot Study

Machine learning models for clinical prediction tasks, such as in-hospital mortality and sepsis onset, routinely achieve high AUROC scores. However, AUROC measures ranking performance rather than clinical sensibility. A model may rank patients correctly overall while predicting a lower mortality risk when a patient's SOFA score worsens, contradicting established medical knowledge. This paper proposes applying metamorphic testing (MT) to clinical machine learning models to evaluate behavioral correctness without requiring ground-truth labels for individual predictions. We design a catalog of 12 candidate metamorphic relations (MRs) for three ICU prediction tasks using the MIMIC-III and MIMIC-IV datasets, with each MR grounded in an authoritative clinical guideline. We further propose a five-layer validation strategy to ensure that MRs are clinically sound before deployment. As a feasibility study, we evaluate the approach on the UCI Heart Disease dataset. Although the three clinical models achieve strong predictive performance (AUROC = 0.849-0.900), they exhibit MT violation rates ranging from 27% to 87% across five pilot MRs. An injected-fault experiment further shows that a sign-negation error in a blood pressure feature remains undetected by AUROC but increases the MT violation rate by 31-67 percentage points. These findings suggest that metamorphic testing provides a valuable complement to conventional performance metrics for assessing the behavioral correctness of clinical prediction models.

cs.SE↗

arXiv2026

Scheduling Jobs with Multiple Operational Modes and Tail Times

This study explores a scheduling challenge inspired by the production of programmable materials, such as advanced liquid crystal displays. In these systems, the final quality of a product is reached only after a resource-free maturation period, known as a "tail", during which the machine is available for processing other jobs. Each job can be executed in one of several operational modes, with each mode determining a specific combination of machine processing time and subsequent tail duration. The primary task is to simultaneously choose the best mode for every job and determine their processing order. We analyze this model across several key performance goals, including the total time required to finish all jobs, the synchronization of completion times (the gap between the earliest and latest finished products), and the total weighted completion time. Our findings provide a detailed classification of the computational complexity of these problems. We demonstrate that while traditional versions with only one mode per job are simple to solve using standard rules, the introduction of just two modes makes finding optimal solutions for most of these goals computationally difficult. When the number of available modes is large, the complexity increases significantly. However, we also identify specific scenarios that remain efficiently solvable, such as when the processing order is already determined or when the goal is to minimize the average completion time. These results offer theoretical clarity and practical strategies for optimizing complex manufacturing and chemical processes involving forced cooling or maturation stages.

cs.DS↗

arXiv2026

CAPTS: Channel-Aware, Preference-Aligned Trigger Selection for Multi-Channel Item-to-Item Retrieval

Large-scale industrial recommender systems commonly adopt multi-channel retrieval for candidate generation, combining direct user-to-item (U2I) retrieval with two-hop user-to-item-to-item (U2I2I) pipelines. In U2I2I, the system selects a small set of historical interactions as triggers to seed downstream item-to-item (I2I) retrieval across multiple channels. In production, triggers are often selected using rule-based policies or learned scorers and tuned in a channel-by-channel manner. However, these practices face two persistent challenges: biased value attribution that values triggers by on-trigger feedback rather than their downstream utility as retrieval seeds, and uncoordinated multi-channel routing where channels select triggers independently under a shared quota, increasing cross-channel overlap. To address these challenges, we propose Channel-Aware, Preference-Aligned Trigger Selection (CAPTS), a unified and flexible framework that treats multi-channel trigger selection as a learnable routing problem. CAPTS introduces a Value Attribution Module (VAM) that provides look-ahead supervision by crediting each trigger with the subsequent engagement generated by items retrieved from it on each I2I channel, and a Channel-Adaptive Trigger Routing (CATR) module that coordinates trigger-to-channel assignment to maximize the overall value of multi-channel retrieval. Extensive offline experiments and large-scale online A/B tests on Kwai, Kuaishou's international short-video platform, show that CAPTS consistently improves multi-channel recall offline and delivers a +0.351% lift in average time spent per device online.

cs.IR↗

arXiv2026

Meroniums in Spin-Orbit Coupled Bose Gases

In this work we demonstrate the existence of new types of topological states in a two-component Bose gas with Rashba spin-orbit couplings. We construct a wave function by mapping the degenerate ground states to the circular boundary of the system, and further consider its superposition with the time-reversed counterpart. By employing the imaginary-time evolution method, we identify four types of topological phases. The first one is the well-known half vortex(HV). The second one is called radial wave half vortex(RWHV), which is a combination of HV with a radial wave factor $e^{ikr}$. The spin configurations show that both HV and RWHV are merons. The third and fourth types are of particular interest. The third one is a combination of HV and anti-HV. The spin densities of this phase demonstrates peak structures and the peaks of spin up and spin down densities are mismatched, and hence we call it the double peak (DP) phase. The fourth type is a combination of RWHV and anti-RWHV, due to the superposition of the radial wave factor $e^{ikr}$ and $e^{-ikr}$ the spin densities demonstrate a spiral pattern, then it's called spin spiral (SS) phase. They are both combinations of meron and antimeron. Here, we name them as meroniums. The topological charges of the four phases are calculated and the spin distributions are demonstrated.

cond-mat.quant-gas↗

arXiv2026

CaT-GS: Efficient 3DGS Rendering for Large Scale Scenes via Inter-frame Caching and Tile Scheduling

Recent breakthroughs in 3D Gaussian Splatting (3DGS) have advanced neural rendering with high fidelity and speed. However, its performance degrades significantly in large-scale scenes due to the computational burden of tile-based rasterization. Existing optimization efforts either require costly scene re-training or focus on narrow aspects of the pipeline, overlooking critical inefficiencies in real-world deployments. Through a comprehensive analysis, we identify three primary sources of redundancy and low GPU utilization: redundant inter-frame pre-processing, viewpoint-based occlusion redundancy, and severe tile-level load imbalance. To address these issues, we propose CaT-GS, a novel and efficient 3DGS rendering pipeline. CaT-GS introduces a speculative multi-frame preprocessing method to eliminate redundant computations across consecutive frames, and an inter-frame caching mechanism to eliminate viewpoint redundant rendering stages. Furthermore, it refactors rasterization tasks with a dedicated kernel to mitigate tile load imbalance, significantly boosting GPU utilization. Extensive experiments demonstrate that CaT-GS achieves a speedup of up to 10 times over the original 3DGS and up to 70% over previous state-of-the-art methods, establishing a new benchmark for high-fidelity, real-time rendering of large-scale scenes.

cs.CV↗

arXiv2026

Full-Pipeline Inference Optimization for MiMo-V2.5 Series: Pushing Hybrid SWA Efficiency to the Limit

We present a full-pipeline inference optimization for the MiMo-V2.5 model family, which combines Hybrid Sliding Window Attention (Hybrid SWA), sparse Mixture-of-Experts (MoE), and multimodal encoders. While Hybrid SWA can ideally reduce both attention compute and KVCache storage significantly compared to Full Attention, realizing these gains in production requires substantial engineering effort. We systematically optimize the KVCache system with layerwise prefetch, SWA-aware prefix cache trees, and specialized placement strategies, achieving strict $O(W)$ SWA storage and high cache hit rates. We further build GCache, a high-performance distributed cache infrastructure with RDMA-optimized networking, and develop a KVCache-affinity router to reduce computation while preserving load balancing. We also optimize for multimodal inputs, including GPU image preprocessing, parallel video decoding, and multimodal cache sharing. Together, these optimizations constitute the first large-scale LLM serving system in production that efficiently covers the Hybrid SWA + MoE + multimodal composite architecture.

cs.AR↗

arXiv2026

When Screening Misleads: A Robust Mendelian Randomization Test for Reliable Causal Discovery

Mendelian Randomization (MR) has been widely used as a standard approach for identifying causal effects in biomedical research. When the true exposure and outcome variables are unknown among many potential candidates, a common but problematic practice is to first screen for associations and then evaluate causal effects only among exposure-outcome pairs that show significant correlations. We demonstrate that the classical MR ratio estimator suffers from severe type I error inflation under this selection procedure. To address this issue, we propose a novel robust MR test that remains valid regardless of the prior association screening step when there is no true causal effect. We show that the proposed test consistently maintains the correct type I error rate, independent of the association test results. Furthermore, our method can incorporate summary statistics from previous association studies to improve the power of causal effect detection. Extensive simulation studies illustrate the advantages of the proposed method compared with the classical MR approach.

stat.ME↗

arXiv2026

Quantifying the Sources of Instability in LLM-Based Stance Analysis of Public Discourse

Computational social science increasingly relies on automated preprocessing pipelines -- speaker diarization, ASR transcript cleaning, sentence segmentation -- to convert raw media into analyzable text. When these pipelines produce different outputs from the same input, two distinct sources of instability can arise: the preprocessing pipeline itself (diarization method, segmentation rules) and the downstream measurement instrument (LLM annotation vs.\ keyword lexicon). Using 256 YouTube interviews across 41 public figures from five domains, we compare two speaker-diarization pipelines and two measurement methods, all targeting the coupling between affective valence and epistemic modality. We find that (1) preprocessing pipeline sensitivity is concentrated in speakers with limited video samples (N $\leq 5$); for the four best-sampled speakers (N $\geq 16$), the mean absolute pipeline-induced change in $r(\text{neg}, \text{emph})$ is only $0.13$; (2) cross-method disagreement is larger and more systematic -- the LLM and keyword-lexicon methods assign opposite coupling directions to several well-sampled speakers, even within the same preprocessing pipeline; and (3) aggregate valence proportions are highly stable ($|Δp(\text{neg})| < 6$pp) regardless of pipeline or method, masking both sources of instability. The contribution is a diagnostic framework that separates pipeline effects from measurement effects: researchers studying cross-dimensional relationships in interview data should verify that their conclusions are robust to both sources of variation, with particular attention to measurement method choice.

cs.CL↗

arXiv2026

Towards Mitigation of Hallucination for LLM-empowered Agents: Progressive Generalization Bound Exploration and Watchdog Monitor

Empowered by large language models (LLMs), intelligent agents have become a popular paradigm for interacting with open environments to facilitate AI deployment. However, hallucinations generated by LLMs-where outputs are inconsistent with facts-pose a significant challenge, undermining the credibility of intelligent agents. Only if hallucinations can be mitigated, the intelligent agents can be used in real-world without any catastrophic risk. Therefore, effective detection and mitigation of hallucinations are crucial to ensure the dependability of agents. Unfortunately, the related approaches either depend on white-box access to LLMs or fail to accurately identify hallucinations. To address the challenge posed by hallucinations of intelligent agents, we present HalMit, a novel black-box watchdog framework that models the generalization bound of LLM-empowered agents and thus detect hallucinations without requiring internal knowledge of the LLM's architecture. Specifically, a probabilistic fractal sampling technique is proposed to generate a sufficient number of queries to trigger the incredible responses in parallel, efficiently identifying the generalization bound of the target agent. Experimental evaluations demonstrate that HalMit significantly outperforms existing approaches in hallucination monitoring. Its black-box nature and superior performance make HalMit a promising solution for enhancing the dependability of LLM-powered systems.

cs.LG↗

arXiv2026

Overthink-Triggered Slowdown Attacks on LVLM-Based Robotic Systems

Large Vision-Language Models (LVLMs) have been increasingly integrated into robotic systems. However, these models may exhibit overthinking behaviors, where they generate excessively long reasoning traces, incurring an excessive inference time. This overthinking behavior poses a serious risk to robotic systems, as the adversary can deliberately trigger overthinking to slow down the decision making of a victim robotic system, causing a variety of safety issues (i.e., an overthinking-induced slowdown attack). To initiate this attack, an adversary can embed carefully crafted, human-readable scene text into the visual scene observed by a victim robotic agent, causing significant inference delays even under a strict black-box setting. Therefore, the embedded scene text serves as a significant "trigger" for the attack. This work systematically identifies and validates transferable triggers of overthinking in robotic systems by introducing a three-stage framework. First, we construct a diverse corpus of reasoning-intensive scene text and extract overthinking-correlated lexical features from short response prefixes. Second, we perform an efficient black-box search guided by a prefix-based proxy score while selectively confirming a small set of top candidates with full latency measurements. Third, we evaluate black-box transfer using a fixed pool of triggers on unseen images and multiple LVLMs, reporting latency amplification and attack success rates under standard thresholds. Across three representative LVLMs, all triggers yield slowdown ratios greater than 1.0x, with the strongest single-trigger case reaching 6.96x. The physical printing of the text trigger still causes up to 4.74x latency amplification. These results demonstrate that our discovered triggers are transferred between multiple LVLM models and consistently cause significant slowdowns in robotic systems.

cs.CR↗

arXiv2026

Nonthermal melting and density wave instability coupled to the lattice in La$_4$Ni$_3$O$_{10}$

The recent discovery of high-temperature superconductivity in pressurized nickelates has renewed interest in the broken-symmetry states of their ambient-pressure parent phases, where a density-wave (DW) order emerges and competes with superconductivity, but its microscopic origin remains unresolved. Using ultrafast optical spectroscopy, we track quasiparticle relaxation dynamics across the DW transition at $T_{\rm DW} \approx$ 136 K in trilayer nickelate {\LNO} single crystals, revealing the opening of an energy gap of $\sim$52 meV. Multiple coherent phonons, including $A_g$ modes near 3.88, 5.28, and 2.09 THz, display pronounced mode-selective anomalies across the transition, indicating that the DW is strongly coupled to lattice degrees of freedom and suggesting an important role of electron-phonon coupling. At higher excitation densities, the DW is nonthermally suppressed, producing a temperature-fluence phase diagram that parallels pressure-tuned behavior. These results establish the DW in {\LNO} as a lattice-entangled instability involving multiple phonon modes, and highlight ultrafast optical excitation as a nonequilibrium tuning parameter for suppressing density-wave order in nickelates.

cond-mat.str-el↗

arXiv2026

Median-of-Means for Nash Equilibrium Seeking in Heavy-Tailed Games

This paper studies the Nash equilibrium seeking problem for stochastic games under heavy-tailed noise. The gradient noise is considered to have a finite $δ$-th moment ($1<δ\le 2$), which generalizes the Gaussian noise and covers cases with infinite variance. In this work, we employ the classic method Median-of-Means (MoM) in robust estimation. MoM works by dividing samples into blocks, taking the average of each block, and then taking the median of these block averages, achieving a breakdown point of up to $1/2$. This makes the final estimate reliable even when some samples are very noisy or wrong, and thus is effective to handle the heavy-tailed noise. The method also naturally defends against malicious gradient attacks. Compared with gradient clipping, which is the most popular method to deal with the heavy-tailed noise, MoM requires no preset clipping threshold and is insensitive to the tail behavior of the noise. Under standard assumptions, we prove the almost sure convergence of the algorithm and derive its almost sure convergence rate. To address the systematic bias caused by asymmetric noise, we further design an online bias correction strategy. Simulation results show the effectiveness and efficiency of the proposed algorithms.

math.OC↗

arXiv2026

Ultrafast Fluence-Reversal Fingerprint of Fragile Kondo Hybridization in CePt$_2$In$_7$

The emergence of heavy quasiparticles in a Kondo lattice is usually viewed as the formation of a low-energy hybridization gap. Whether this gap represents a rigid electronic structure or a fragile many-body state that can be dynamically reconfigured remains a central question for heavy-fermion systems near magnetic order, quantum criticality, and unconventional superconductivity. Here we use femtosecond pump-probe reflectivity to interrogate this problem in the weakly hybridized Kondo-lattice compound CePt$_2$In$_7$. At low fluence, a slow quasiparticle relaxation channel emerges below $T^* \sim$ 40 K and follows a Rothwarf-Taylor bottleneck response with a low-energy recombination scale 2$Δ\approx$ 7.4 meV. Coherent optical phonons, independently identified by Raman spectroscopy, act as an internal lattice thermometer and rule out large quasi-equilibrium lattice heating as the origin of the nonlinear electronic response. The phonon-free electronic amplitude $A_{\rm elec}$ reveals a fluence-reversal fingerprint: with cooling from the hybridization-crossover regime, the response evolves from weak-linear behavior to Rothwarf-Taylor-like bottleneck suppression and finally to anomalous high-fluence enhancement at the lowest temperatures. This reversal cannot be accounted for by a rigid fixed-gap bottleneck alone and instead identifies an ultrafast optical signature of photoinduced redistribution of a fragile Kondo-hybridized electronic response.

cond-mat.str-el↗

arXiv2026

Probing-Guided Layer Selection from Self-Supervised Speech Models for Generalizable Audio Deepfake Detection

Audio deepfake detection systems often fail to generalize across domains because they rely on features tied to specific attacks or recording conditions. Self-supervised speech models offer rich multi-layer representations, yet existing approaches either use a single layer or fuse all layers indiscriminately, and only reveal layer importance after training. We propose a model-agnostic, two-stage methodology that identifies informative depth zones before any task-specific model is trained. In the first stage, lightweight XGBoost probes evaluate each transformer layer's cross-domain discriminative power, producing a layer ranking. In the second stage, a compact neural classifier fuses only the selected layers through per-layer attention pooling and a shared bottleneck projection, while the backbone remains frozen. Applied across three backbones, the probing reveals two key findings. First, informative layers cluster in depth zones rather than at uniquely optimal positions: within-zone substitutions fall within multi-seed noise, while zone violations degrade performance by up to 5x. Second, the probing produces backbone-specific selections rather than a fixed layer recipe. On XLS-R-300M, four probing-selected layers with 1.34M trainable parameters achieve 4.94 +/- 0.32% equal error rate on In-The-Wild and 5.07% cross-domain average over four shared datasets, a 28% relative improvement over the best prior frozen-backbone result (Xiao and Vu, 2025) using all 25 layers with identical training data.

cs.SD↗

arXiv2026

BaRA: Bayesian Adaptive Rank Allocation for Parameter-Efficient Fine-Tuning

While Low-rank adaptation (LoRA) enables highly efficient fine-tuning by constraining task-specific updates to fixed low-rank subspaces, this rigid design limits representational flexibility and often results in overconfident predictions and miscalibrated uncertainty, especially in low-data regimes. Recent Bayesian LoRA variants improve uncertainty estimation by modeling posterior distributions over adaptation parameters. However, these approaches typically rely on fixed or heuristically determined ranks, overlooking the inherently context-dependent nature of adaptation capacity. In this paper, we propose BaRA, a Bayesian Adaptive Rank Allocation framework for parameter-efficient fine-tuning. Drawing inspiration from probabilistic topic models, BaRA dynamically allocates adaptation capacity by activating a sparse, context-dependent subset of disentangled latent factors, enabling instance-wise variation in effective rank. This Bayesian formulation provides principled, data-driven capacity control, mitigating over-parameterization while preserving expressiveness. Beyond the modeling contribution, we provide a complexity-theoretic generalization analysis showing that the generalization gap of BaRA depends on the learned joint effective rank $\bar{s}_{Φ,θ}$ induced by the global-local gate, rather than the maximum rank $r$. This result explains why sparse adaptive rank allocation can reduce the effective hypothesis complexity while preserving input-dependent expressiveness. Extensive experiments on diverse natural language benchmarks demonstrate that BaRA consistently improves predictive performance, robustness, and uncertainty calibration compared to standard LoRA and existing Bayesian LoRA variants.

cs.LG↗

arXiv2026

Beyond Spectral Decomposition: Bayesian Contrastive Learning and its Non-negative Formulation via Factor Analysis

Factor analysis, often regarded as a Bayesian variant of matrix factorization, offers superior capabilities in capturing uncertainty, modeling complex dependencies, and ensuring robustness. As the deep learning era arrives, factor analysis is receiving less and less attention due to their limited expressive ability. On the contrary, contrastive learning has emerged as a potent technique with demonstrated efficacy in unsupervised representational learning. While the two methods are different paradigms, recent theoretical analysis has revealed the mathematical equivalence between contrastive learning and matrix factorization, providing a potential possibility for factor analysis combined with contrastive learning. Motivated by the interconnectedness of contrastive learning, matrix factorization, and factor analysis, this paper introduces a novel Contrastive Factor Analysis framework, aiming to leverage factor analysis's advantageous properties within the realm of contrastive learning. To further leverage the interpretability properties of non-negative factor analysis, which can learn disentangled representations, contrastive factor analysis is extended to a non-negative version. Finally, extensive experimental validation showcases the efficacy of the proposed contrastive (non-negative) factor analysis methodology across multiple key properties, including expressiveness, robustness, interpretability, and accurate uncertainty estimation.

cs.LG↗

arXiv2026

On the Position Bias of On-Policy Distillation

On-Policy Distillation (OPD) improves the learning efficiency of standard reinforcement learning through dense, token-level supervision from teachers. In the standard KL objective of OPD, token-level losses are uniformly averaged, implying equal weights for all tokens. However, we discover that not all tokens are created equal: as student rollouts grow longer, they deviate further from the teacher's distribution, leading to degraded supervision quality at later positions. As a result, OPD using only the first 30% of tokens can perform comparably to using all tokens, whereas OPD using only the last 30% of tokens barely learns anything. In this work, we provide a principled understanding of this issue through the lens of constrained optimization. Based on these insights, we derive Importance-Weighted On-Policy Distillation (IW-OPD), in which the weight assigned to each token depends on the accumulated discrepancy between the student's and teacher's distributions, naturally upweighting earlier tokens and downweighting later ones with larger deviations. We show that IW-OPD converges significantly faster than OPD, with better learning efficiency, and achieves better final performance than standard OPD in both same-size and cross-scale settings, improving performance up to 6.9 points on AIME-2025.

cs.LG↗

arXiv2026

UniFormer: Efficient and Unified Model-Centric Scaling for Industrial Recommendation

Recently, substantial progress has been made in industrial recommendation through component-centric model scaling, where individual components such as behavior modeling, feature interaction, or task modeling are independently scaled to improve model capacity. Although recent methods such as HyFormer and OneTrans further explore cross-module co-scaling by jointly modeling behavior and interaction, their designs are still confined to the feature space and lack a unified model-centric scaling framework over the overall modeling space. In this paper, we propose UniFormer, an efficient and unified model-centric scaling framework for industrial recommender systems. To improve efficiency, UniFormer decomposes the overall modeling space into feature and task spaces, which are modeled by stacked Feature-space Interaction Modules and Task-space Interaction Modules, respectively. Moreover, UniFormer introduces semantic-based tokenization scheme to enable user-item decoupling, thereby achieving request-level inference acceleration. To prevent preference collapse, UniFormer employs multi-sequence cross-attention to separately capture heterogeneous behavior patterns, followed by the self-attention to enhance interaction modeling. Besides, dedicated multi-view FFNs are introduced to support flexible and scalable parameter scaling across different modeling components. Extensive online A/B testing in two production scenarios, Kuaishou and Kuaishou Lite, shows that UniFormer consistently improves user engagement and interaction metrics, achieving gains of +0.101%/+0.260% in App Stay Time and +0.729%/+1.113% in Watch Time, respectively.

cs.IR↗