arXiv subjects

Jian Kang

Publications and source records attributed to Jian Kang.

At least 73 records · Page 4Linked to original sources

arXiv2025

Leveraging LLM and Self-Supervised Training Models for Speech Recognition in Chinese Dialects: A Comparative Analysis

Large-scale training corpora have significantly improved the performance of ASR models. Unfortunately, due to the relative scarcity of data, Chinese accents and dialects remain a challenge for most ASR models. Recent advancements in self-supervised learning have shown that self-supervised pre-training, combined with large language models (LLM), can effectively enhance ASR performance in low-resource scenarios. We aim to investigate the effectiveness of this paradigm for Chinese dialects. Specifically, we pre-train a Data2vec2 model on 300,000 hours of unlabeled dialect and accented speech data and do alignment training on a supervised dataset of 40,000 hours. Then, we systematically examine the impact of various projectors and LLMs on Mandarin, dialect, and accented speech recognition performance under this paradigm. Our method achieved SOTA results on multiple dialect datasets, including Kespeech. We will open-source our work to promote reproducible research

cs.CL↗

arXiv2025

Ensuring User-side Fairness in Dynamic Recommender Systems

User-side group fairness is crucial for modern recommender systems, aiming to alleviate performance disparities among user groups defined by sensitive attributes like gender, race, or age. In the ever-evolving landscape of user-item interactions, continual adaptation to newly collected data is crucial for recommender systems to stay aligned with the latest user preferences. However, we observe that such continual adaptation often exacerbates performance disparities. This necessitates a thorough investigation into user-side fairness in dynamic recommender systems, an area that has been unexplored in the literature. This problem is challenging due to distribution shifts, frequent model updates, and non-differentiability of ranking metrics. To our knowledge, this paper presents the first principled study on ensuring user-side fairness in dynamic recommender systems. We start with theoretical analyses on fine-tuning v.s. retraining, showing that the best practice is incremental fine-tuning with restart. Guided by our theoretical analyses, we propose FAir Dynamic rEcommender (FADE), an end-to-end fine-tuning framework to dynamically ensure user-side fairness over time. To overcome the non-differentiability of recommendation metrics in the fairness loss, we further introduce Differentiable Hit (DH) as an improvement over the recent NeuralNDCG method, not only alleviating its gradient vanishing issue but also achieving higher efficiency. Besides that, we also address the instability issue of the fairness loss by leveraging the competing nature between the recommendation loss and the fairness loss. Through extensive experiments on real-world datasets, we demonstrate that FADE effectively and efficiently reduces performance disparities with little sacrifice in the overall recommendation performance.

cs.IR↗

arXiv2025

GOAT-TTS: Expressive and Realistic Speech Generation via A Dual-Branch LLM

While large language models (LLMs) have revolutionized text-to-speech (TTS) synthesis through discrete tokenization paradigms, current architectures exhibit fundamental tensions between three critical dimensions: 1) irreversible loss of acoustic characteristics caused by quantization of speech prompts; 2) stringent dependence on precisely aligned prompt speech-text pairs that limit real-world deployment; and 3) catastrophic forgetting of the LLM's native text comprehension during optimization for speech token generation. To address these challenges, we propose an LLM-based text-to-speech Generation approach Optimized via a novel dual-branch ArchiTecture (GOAT-TTS). Our framework introduces two key innovations: (1) The modality-alignment branch combines a speech encoder and projector to capture continuous acoustic embeddings, enabling bidirectional correlation between paralinguistic features (language, timbre, emotion) and semantic text representations without transcript dependency; (2) The speech-generation branch employs modular fine-tuning on top-k layers of an LLM for speech token prediction while freezing the bottom-n layers to preserve foundational linguistic knowledge. Moreover, multi-token prediction is introduced to support real-time streaming TTS synthesis. Experimental results demonstrate that our GOAT-TTS achieves performance comparable to state-of-the-art TTS models while validating the efficacy of synthesized dialect speech data.

cs.CL↗

arXiv2025

Characterizing Bias: Benchmarking Large Language Models in Simplified versus Traditional Chinese

While the capabilities of Large Language Models (LLMs) have been studied in both Simplified and Traditional Chinese, it is yet unclear whether LLMs exhibit differential performance when prompted in these two variants of written Chinese. This understanding is critical, as disparities in the quality of LLM responses can perpetuate representational harms by ignoring the different cultural contexts underlying Simplified versus Traditional Chinese, and can exacerbate downstream harms in LLM-facilitated decision-making in domains such as education or hiring. To investigate potential LLM performance disparities, we design two benchmark tasks that reflect real-world scenarios: regional term choice (prompting the LLM to name a described item which is referred to differently in Mainland China and Taiwan), and regional name choice (prompting the LLM to choose who to hire from a list of names in both Simplified and Traditional Chinese). For both tasks, we audit the performance of 11 leading commercial LLM services and open-sourced models -- spanning those primarily trained on English, Simplified Chinese, or Traditional Chinese. Our analyses indicate that biases in LLM responses are dependent on both the task and prompting language: while most LLMs disproportionately favored Simplified Chinese responses in the regional term choice task, they surprisingly favored Traditional Chinese names in the regional name choice task. We find that these disparities may arise from differences in training data representation, written character preferences, and tokenization of Simplified and Traditional Chinese. These findings highlight the need for further analysis of LLM biases; as such, we provide an open-sourced benchmark dataset to foster reproducible evaluations of future LLM behavior across Chinese language variants (https://github.com/brucelyu17/SC-TC-Bench).

cs.CL↗

arXiv2025

TopSpace: spatial topic modeling for unsupervised discovery of multicellular spatial tissue structures in multiplex imaging

Motivation: Understanding the spatial architecture of tissues is essential for decoding the complex interactions within cellular ecosystems and their implications for disease pathology and clinical outcomes. Recent advances in multiplex imaging technologies have enabled high-resolution profiling of cellular phenotypes and their spatial distributions, revealing critical roles of tissue structures such as tertiary lymphoid structures (TLSs) in shaping immune responses and influencing disease progression. However, existing methods for analyzing spatial tissue structures often rely on hard clustering or adjacency-based spatial models, which are limited in capturing the nuanced and overlapping nature of cellular communities. To address these challenges, we develop a novel spatial topic modeling framework for the unsupervised discovery of spatial tissue structures in multiplex imaging data. Results: We propose TopSpace, a novel Bayesian spatial topic model that integrates Gaussian processes into latent Dirichlet allocation to flexibly model spatial dependencies in tissue microenvironments. By leveraging the Bayesian framework, TopSpace supports multicellular mixed-membership clustering and offers key inferential advantages, including robust uncertainty quantification and data-driven determination of the number of multicellular microenvironments. We demonstrate the utility of TopSpace through simulations and a case study on non-small cell lung cancer (NSCLC) data. Simulations show that TopSpace accurately recovers latent tissue microenvironments and spatial clustering patterns, outperforming existing methods in scenarios with varying spatial dependencies. Applied to NSCLC data, TopSpace successfully identifies TLS and captures their spatial probability distribution, which strongly correlates with patient survival outcomes.

q-bio.QM↗

arXiv2025

Discovering Knowledge Deficiencies of Language Models on Massive Knowledge Base

Large language models (LLMs) possess impressive linguistic capabilities but often fail to faithfully retain factual knowledge, leading to hallucinations and unreliable outputs. Understanding LLMs' knowledge deficiencies by exhaustively evaluating against full-scale knowledge bases is computationally prohibitive, especially for closed-weight models. We propose stochastic error ascent (SEA), a scalable and efficient framework for discovering knowledge deficiencies (errors) in closed-weight LLMs under a strict query budget. Rather than naively probing all knowledge candidates, SEA formulates error discovery as a stochastic optimization process: it iteratively retrieves new high-error candidates by leveraging the semantic similarity to previously observed failures. To further enhance search efficiency and coverage, SEA employs hierarchical retrieval across document and paragraph levels, and constructs a relation directed acyclic graph to model error propagation and identify systematic failure modes. Empirically, SEA uncovers 40.7x more knowledge errors than Automated Capability Discovery and 26.7% more than AutoBencher, while reducing the cost-per-error by 599x and 9x, respectively. Human evaluation confirms the high quality of generated questions, while ablation and convergence analyses validate the contribution of each component in SEA. Further analysis on the discovered errors reveals correlated failure patterns across LLM families and recurring deficits, highlighting the need for better data coverage and targeted fine-tuning in future LLM development.

cs.CL↗

arXiv2025

Spectral signature of periodic modulation and sliding of pseudogap state in moire system

The nature of the pseudogap state is widely believed as a key to understanding the pairing mechanism underlying unconventional superconductivity. Over the past two decades, significant efforts have been devoted to searching for spontaneous symmetry breaking or potential order parameters associated with these pseudogap states, aiming to better characterize their properties. Recently, pseudogap states have also been realized in moire systems with extensive gate tunability, yet their local electronic structure remains largely unexplored8. In this study, we report the observation of gate-tunable spontaneous symmetry breaking and sliding behavior of the pseudogap state in magic-angle twisted bilayer graphene (MAtBG) using spectroscopic imaging scanning tunneling microscopy. Our spectroscopy reveals a distinct pseudogap at 4.4 K within the doping range -3 < v < -2. Spectroscopic imaging highlights a gap size modulation at moire scale that is sensitive to the filling, indicative of a wave-like fluctuating pseudogap feature. Specifically, the positions of gap size minima (GSM) coincide with regions of the highest local density of states (LDOS) at the filling v = -2.63, but a unidirectional sliding behavior of GSM is observed for other fillings. In addition, the pseudogap size distribution at certain doping levels also causes a clear nematic order, or an anisotropic gap distribution. Our results have shed light on the complex nature of this pseudogap state, revealing critical insights into the phase diagram of correlated electron systems.

cond-mat.mes-hall↗

arXiv2025

What Influences the Field Goal Attempts of Professional Players? Analysis of Basketball Shot Charts via Log Gaussian Cox Processes with Spatially Varying Coefficients

Basketball shot charts provide valuable information regarding local patterns of in-game performance to coaches, players, sports analysts, and statisticians. The spatial patterns of where shots were attempted and whether the shots were successful suggest options for offensive and defensive strategies as well as historical summaries of performance against particular teams and players. The data represent a marked spatio-temporal point process with locations representing locations of attempted shots and an associated mark representing the shot's outcome (made/missed). Here, we develop a Bayesian log Gaussian Cox process model allowing joint analysis of the spatial pattern of locations and outcomes of shots across multiple games. We build a hierarchical model for the log intensity function using Gaussian processes, and allow spatially varying effects for various game-specific covariates. We aim to model the spatial relative risk under different covariate values. For inference via posterior simulation, we design a Markov chain Monte Carlo (MCMC) algorithm based on a kernel convolution approach. We illustrate the proposed method using extensive simulation studies. A case study analyzing the shot data of NBA legends Stephen Curry, LeBron James, and Michael Jordan highlights the effectiveness of our approach in real-world scenarios and provides practical insights into optimizing shooting strategies by examining how different playing conditions, game locations, and opposing team strengths impact shooting efficiency.

stat.ME↗

arXiv2025

A general, flexible and harmonious framework to construct interpretable functions in regression analysis

An interpretable model or method has several appealing features, such as reliability to adversarial examples, transparency of decision-making, and communication facilitator. However, interpretability is a subjective concept, and even its definition can be diverse. The same model may be deemed as interpretable by a study team, but regarded as a black-box algorithm by another squad. Simplicity, accuracy and generalizability are some additional important aspects of evaluating interpretability. In this work, we present a general, flexible and harmonious framework to construct interpretable functions in regression analysis with a focus on continuous outcomes. We formulate a functional skeleton in light of users' expectations of interpretability. A new measure based on Mallows's $C_p$-statistic is proposed for model selection to balance approximation, generalizability, and interpretability. We apply this approach to derive a sample size formula in adaptive clinical trial designs to demonstrate the general workflow, and to explain operating characteristics in a Bayesian Go/No-Go paradigm to show the potential advantages of using meaningful intermediate variables. Generalization to categorical outcomes is illustrated in an example of hypothesis testing based on Fisher's exact test. A real data analysis of NHANES (National Health and Nutrition Examination Survey) is conducted to investigate relationships between some important laboratory measurements. We also discuss some extensions of this method.

stat.ME↗

arXiv2025

Heavy Fermions as an Efficient Representation of Atomistic Strain and Relaxation in Twisted Bilayer Graphene

Although the strongly interacting flat bands in twisted bilayer graphene (TBG) have been approached using the minimal Bistritzer-MacDonald (BM) Hamiltonian, there is mounting evidence that strain and lattice relaxation are essential in correctly determining the order of the correlated insulator groundstates. These effects can be incorporated in an enhanced continuum model by introducing additional terms computed from the relaxation profile. To develop an analytical and physical understanding of these effects, we include strain and relaxation in the topological heavy fermion (HF) model of TBG. We find that strain and relaxation are very well captured in first order perturbation theory by projection onto the fully symmetric HF Hilbert space, and remarkably do not alter the interacting terms in the periodic Anderson model. Their effects are fully incorporated in the single-particle HF Hamiltonian, and can be reproduced in a minimal model with only 4 symmetry-breaking terms. Our results demonstrate that the heavy fermion framework of TBG is an efficient and robust representation of the perturbations encountered in experiment.

cond-mat.mes-hall↗

arXiv2025

Kekulé Spiral Order from Strained Topological Heavy Fermions

The topological heavy fermion (THF) model of twisted bilayer graphene is a framework for treating its strongly interacting topological flat bands. In this work, we employ the THF model with heterostrain and particle-hole symmetry breaking corrections to study its symmetry-broken ground states. We find that the heterostrain correction motivates a specific parent-state wavefunction which dictates the presence or absence of an incommensurate Kekulé spiral (IKS) at each integer filling by invoking Dirac node braiding and annihilation as a mechanism to achieve low energy gapped states. We then show that one-shot Hartree-Fock faithfully replicates the numerical results of fully self-consistent states and motivates an analytical approximation for the IKS wavevector. We can also account for the particle-hole asymmetry in the correlated insulator gaps. In particular, the THF model predicts stronger correlated states on the electron side rather than hole side in agreement with magic angle experiments, despite the electron side being more dispersive in the single-particle band structure. This work demonstrates that we can analytically explain even the more subtle symmetry breaking order properties observed in experiments where heterostrain, relaxation, and interactions together determine the ground state.

cond-mat.str-el↗

arXiv2025

Understanding and Rectifying Safety Perception Distortion in VLMs

Recent studies reveal that vision-language models (VLMs) become more susceptible to harmful requests and jailbreak attacks after integrating the vision modality, exhibiting greater vulnerability than their text-only LLM backbones. To uncover the root cause of this phenomenon, we conduct an in-depth analysis and identify a key issue: multimodal inputs introduce an modality-induced activation shift toward a "safer" direction compared to their text-only counterparts, leading VLMs to systematically overestimate the safety of harmful inputs. We refer to this issue as safety perception distortion. To mitigate such distortion, we propose Activation Shift Disentanglement and Calibration (ShiftDC), a training-free method that decomposes and calibrates the modality-induced activation shift to reduce the impact of modality on safety. By isolating and removing the safety-relevant component, ShiftDC restores the inherent safety alignment of the LLM backbone while preserving the vision-language capabilities of VLMs. Empirical results demonstrate that ShiftDC significantly enhances alignment performance on safety benchmarks without impairing model utility.

cs.CV↗

arXiv2024

Bayesian Inference of Spatially Varying Correlations via the Thresholded Correlation Gaussian Process

A central question in multimodal neuroimaging analysis is to understand the association between two imaging modalities and to identify brain regions where such an association is statistically significant. In this article, we propose a Bayesian nonparametric spatially varying correlation model to make inference of such regions. We build our model based on the thresholded correlation Gaussian process (TCGP). It ensures piecewise smoothness, sparsity, and jump discontinuity of spatially varying correlations, and is well applicable even when the number of subjects is limited or the signal-to-noise ratio is low. We study the identifiability of our model, establish the large support property, and derive the posterior consistency and selection consistency. We also develop a highly efficient Gibbs sampler and its variant to compute the posterior distribution. We illustrate the method with both simulations and an analysis of functional magnetic resonance imaging data from the Human Connectome Project.

stat.ME↗

arXiv2024

Contribution of soundscape appropriateness to soundscape quality assessment in space: a mediating variable affecting acoustic comfort

Soundscape appropriateness (SA) provides supplemental information on the matching degree between auditory information and the surrounding scene in soundscape perception. This indicator has been integrated into the standard ISO process for collecting soundscape data, forming a component of the sound quality assessment questionnaire. However, its role in soundscape quality assessment has not been fully understood. Herein, we present the findings from soundscape data collected from Beiling Park in Shenyang, China. A method was developed that integrates mediation effect models with multiscale geographically weighted regression models to explore the mediating role of SA in the impact of sound source types on soundscape quality, as well as the spatial heterogeneity of this mediation effect. The results confirm that SA does mediates the influence of sound source types on acoustics comfort (AC). Specifically, natural sounds (indirect effect/total effect = .19/.19), traffic sounds (indirect effect/total effect = -.46/-.65), and commercial sounds (indirect effect/total effect = -.25/-.12) impact the perception of AC by either enhancing or reducing SA. Moreover, the relationships among variables depicted in this model demonstrate spatial heterogeneity, demonstrating that in urban open spaces with complex constructures, local spatial models may be needed for soundscape assessment. The research reaffirms the significance of SA in urban open spaces. In terms of practical implications for urban and landscape planners, when sound sources cannot be controlled or altered, coordinating between the sound and the surrounding environment through landscape optimisation could also improve the quality of the soundscape through enhancing SA and help achieve the goal of creating healthy urban open spaces.

cs.SD↗

arXiv2024

PageRank Bandits for Link Prediction

Link prediction is a critical problem in graph learning with broad applications such as recommender systems and knowledge graph completion. Numerous research efforts have been directed at solving this problem, including approaches based on similarity metrics and Graph Neural Networks (GNN). However, most existing solutions are still rooted in conventional supervised learning, which makes it challenging to adapt over time to changing customer interests and to address the inherent dilemma of exploitation versus exploration in link prediction. To tackle these challenges, this paper reformulates link prediction as a sequential decision-making process, where each link prediction interaction occurs sequentially. We propose a novel fusion algorithm, PRB (PageRank Bandits), which is the first to combine contextual bandits with PageRank for collaborative exploitation and exploration. We also introduce a new reward formulation and provide a theoretical performance guarantee for PRB. Finally, we extensively evaluate PRB in both online and offline settings, comparing it with bandit-based and graph-based methods. The empirical success of PRB demonstrates the value of the proposed fusion approach. Our code is released at https://github.com/jiaruzouu/PRB.

cs.LG↗

arXiv2024

Theoretical and Empirical Insights into the Origins of Degree Bias in Graph Neural Networks

Graph Neural Networks (GNNs) often perform better for high-degree nodes than low-degree nodes on node classification tasks. This degree bias can reinforce social marginalization by, e.g., privileging celebrities and other high-degree actors in social networks during social and content recommendation. While researchers have proposed numerous hypotheses for why GNN degree bias occurs, we find via a survey of 38 degree bias papers that these hypotheses are often not rigorously validated, and can even be contradictory. Thus, we provide an analysis of the origins of degree bias in message-passing GNNs with different graph filters. We prove that high-degree test nodes tend to have a lower probability of misclassification regardless of how GNNs are trained. Moreover, we show that degree bias arises from a variety of factors that are associated with a node's degree (e.g., homophily of neighbors, diversity of neighbors). Furthermore, we show that during training, some GNNs may adjust their loss on low-degree nodes more slowly than on high-degree nodes; however, with sufficiently many epochs of training, message-passing GNNs can achieve their maximum possible training accuracy, which is not significantly limited by their expressive power. Throughout our analysis, we connect our findings to previously-proposed hypotheses for the origins of degree bias, supporting and unifying some while drawing doubt to others. We validate our theoretical findings on 8 common real-world networks, and based on our theoretical and empirical insights, describe a roadmap to alleviate degree bias.

cs.LG↗

arXiv2024

Automating Urban Soundscape Enhancements with AI: In-situ Assessment of Quality and Restorativeness in Traffic-Exposed Residential Areas

Formalized in ISO 12913, the "soundscape" approach is a paradigmatic shift towards perception-based urban sound management, aiming to alleviate the substantial socioeconomic costs of noise pollution to advance the United Nations Sustainable Development Goals. Focusing on traffic-exposed outdoor residential sites, we implemented an automatic masker selection system (AMSS) utilizing natural sounds to mask (or augment) traffic soundscapes. We employed a pre-trained AI model to automatically select the optimal masker and adjust its playback level, adapting to changes over time in the ambient environment to maximize "Pleasantness", a perceptual dimension of soundscape quality in ISO 12913. Our validation study involving ($N=68$) residents revealed a significant 14.6 % enhancement in "Pleasantness" after intervention, correlating with increased restorativeness and positive affect. Perceptual enhancements at the traffic-exposed site matched those at a quieter control site with 6 dB(A) lower $L_\text{A,eq}$ and road traffic noise dominance, affirming the efficacy of AMSS as a soundscape intervention, while streamlining the labour-intensive assessment of "Pleasantness" with probabilistic AI prediction.

eess.AS↗

arXiv2024

BSNMani: Bayesian Scalar-on-network Regression with Manifold Learning

Brain connectivity analysis is crucial for understanding brain structure and neurological function, shedding light on the mechanisms of mental illness. To study the association between individual brain connectivity networks and the clinical characteristics, we develop BSNMani: a Bayesian scalar-on-network regression model with manifold learning. BSNMani comprises two components: the network manifold learning model for brain connectivity networks, which extracts shared connectivity structures and subject-specific network features, and the joint predictive model for clinical outcomes, which studies the association between clinical phenotypes and subject-specific network features while adjusting for potential confounding covariates. For posterior computation, we develop a novel two-stage hybrid algorithm combining Metropolis-Adjusted Langevin Algorithm (MALA) and Gibbs sampling. Our method is not only able to extract meaningful subnetwork features that reveal shared connectivity patterns, but can also reveal their association with clinical phenotypes, further enabling clinical outcome prediction. We demonstrate our method through simulations and through its application to real resting-state fMRI data from a study focusing on Major Depressive Disorder (MDD). Our approach sheds light on the intricate interplay between brain connectivity and clinical features, offering insights that can contribute to our understanding of psychiatric and neurological disorders, as well as mental health.

stat.ME↗