arXiv subjects

Jiale Deng

Publications and source records attributed to Jiale Deng.

5 recordsLinked to original sources

arXiv2026

DeMix: Debugging Training Data with Mixed Data Error Types by Investigating Influence Vectors

High-quality training data is essential for the success of machine learning models. However, real-world datasets often contain mixed types of errors arising from systematic flaws in data preparation pipelines, including label errors, feature errors, and spurious correlations. Effective debugging of training data requires both detecting erroneous samples and identifying their specific error types to enable targeted repair, yet existing data cleaning and attribution methods fail to adequately address this dual requirement. In this paper, we propose DeMix, a novel framework that simultaneously diagnoses erroneous samples and their error types. Our key insight is that different error types produce distinct patterns on model behavior. DeMix captures such error-specific patterns by influence vectors that characterize how each training sample affects model predictions across all validation samples. We formulate training data debugging as a multi-label classification problem where a classifier is developed to predict error types directly from influence vectors. We further introduce an intervention-based learning strategy that guides the classifier to capture invariant rationales specific to each error type, ensuring the learned classifier generalizes effectively. Empirical evaluations on 11 tasks across tabular data prediction, recommendation systems, and LLM alignment demonstrate that DeMix significantly outperforms state-of-the-art approaches, achieving a 22.61% improvement in data debugging F1-score and a 9.32% gain in task model performance after data repair. Code is available at: https://github.com/SJTU-DMTai/DeMix.

cs.LG↗

arXiv2025

Influence Guided Context Selection for Effective Retrieval-Augmented Generation

Retrieval-Augmented Generation (RAG) addresses large language model (LLM) hallucinations by grounding responses in external knowledge, but its effectiveness is compromised by poor-quality retrieved contexts containing irrelevant or noisy information. While existing approaches attempt to improve performance through context selection based on predefined context quality assessment metrics, they show limited gains over standard RAG. We attribute this limitation to their failure in holistically utilizing available information (query, context list, and generator) for comprehensive quality assessment. Inspired by recent advances in data selection, we reconceptualize context quality assessment as an inference-time data valuation problem and introduce the Contextual Influence Value (CI value). This novel metric quantifies context quality by measuring the performance degradation when removing each context from the list, effectively integrating query-aware relevance, list-aware uniqueness, and generator-aware alignment. Moreover, CI value eliminates complex selection hyperparameter tuning by simply retaining contexts with positive CI values. To address practical challenges of label dependency and computational overhead, we develop a parameterized surrogate model for CI value prediction during inference. The model employs a hierarchical architecture that captures both local query-context relevance and global inter-context interactions, trained through oracle CI value supervision and end-to-end generator feedback. Extensive experiments across 8 NLP tasks and multiple LLMs demonstrate that our context selection method significantly outperforms state-of-the-art baselines, effectively filtering poor-quality contexts while preserving critical information. Code is available at https://github.com/SJTU-DMTai/RAG-CSM.

cs.CL↗

arXiv2024

On-chip electro-optically tunable narrow linewidth Brillouin microlasers implemented in thin film lithium niobate

On-chip narrow linewidth microlasers with real-time wavelength tunability are highly desirable for various applications including precision metrology, quantum technology, and coherent information processing. Realizing such laser remains a challenge despite significant advances made by various groups in recent years [Nat. Commun. 13, 5344 (2022); Nature 615, 411 (2023); Appl. Phys. Lett. 124, 131101 (2024); Nat. Photonics 13, 60 (2019)]. In this work, we overcome these hurdles and demonstrate on-chip electro-optically tunable Brillouin microlasers in compact lithium niobate on insulator (LNOI) microdisks with diameters of 31.5 um and 117.0 um by using cross-polarized SBS arrangement. A quasi-continuum band of bound shear mechanical modes inside the suspended microdisk are revealed for the first time, allowing feasible phase matching of stimulated Brillouin lasing (SBL). We achieve efficient cross-polarized optomechanical coupling and SBL via the significant photoelastic tensors of lithium niobate (e.g., p41=-1.51). This approach yields a 118 Hz intrinsic linewidth and a comparatively low threshold power of 3.15 mW. A real-time electro-optic tuning of the cross-polarized Brillouin scheme with a tuning efficiency of ~93.1 kHz/V is also achieved, further showcasing potential of LNOI platform for next-generation tunable photonic systems.

physics.optics↗

arXiv2024

Efficient transverse multi-wave interactions up to six-wave mixing in a high-Q lithium niobate microresonator

High-order nonlinear optical processes beyond four-wave mixing serve as fundamental tools for advancing photonic technologies, yet their practical implementation remains challenging due to stringent phase-matching requirements and inherently weak high-order nonlinear susceptibilities - limitations that persist even in state-of-the-art high-Q microresonators. In this work, we demonstrate a breakthrough in synthesizing transverse nonlinear processes up to six-wave mixing in an integrated lithium niobate microresonator, under single continuous-wave (CW) telecom-band laser pump. Our approach leverages self-organized subwavelength photorefractive gratings (SPGs) generated through bidirectional stimulated Raman scattering (SRS) process in the microresonator, without using two external counterpropagating lasers. Under 1546 nm pumping, bidirectional SRS at 1713 nm creates two counterpropagating light waves that spontaneously form SPGs. These SPGs critically enable broadband phase-matching compensation across 500 nm spectral range by providing additional momentum matching for transverse nonlinear processes while maintaining ultrahigh-Q factor. Moreover, cascaded SRS process is simultaneously activated to generate light signal for subsequent nonlinear interactions. This novel approach enables, to our knowledge, the first demonstration of single-pump phase-matched transverse sum-frequency generation (SFG) with record conversion efficiency (590%/W). Furthermore, transverse multi-wave mixing processes from four-wave to six-wave mixing processes are achieved with high conversion efficiencies for the first time using only the single CW pump, representing a notable advance in nonlinear integration.

physics.optics↗

arXiv2023

Towards Fine-Grained Explainability for Heterogeneous Graph Neural Network

Heterogeneous graph neural networks (HGNs) are prominent approaches to node classification tasks on heterogeneous graphs. Despite the superior performance, insights about the predictions made from HGNs are obscure to humans. Existing explainability techniques are mainly proposed for GNNs on homogeneous graphs. They focus on highlighting salient graph objects to the predictions whereas the problem of how these objects affect the predictions remains unsolved. Given heterogeneous graphs with complex structures and rich semantics, it is imperative that salient objects can be accompanied with their influence paths to the predictions, unveiling the reasoning process of HGNs. In this paper, we develop xPath, a new framework that provides fine-grained explanations for black-box HGNs specifying a cause node with its influence path to the target node. In xPath, we differentiate the influence of a node on the prediction w.r.t. every individual influence path, and measure the influence by perturbing graph structure via a novel graph rewiring algorithm. Furthermore, we introduce a greedy search algorithm to find the most influential fine-grained explanations efficiently. Empirical results on various HGNs and heterogeneous graphs show that xPath yields faithful explanations efficiently, outperforming the adaptations of advanced GNN explanation approaches.

cs.LG↗