arXiv subjects

Ming Cai

Publications and source records attributed to Ming Cai.

At least 19 recordsLinked to original sources

arXiv2026

Causal DAG Identification for Count Data via Poisson Thinning Structural Equation Models

Count-valued variables arise in many scientific and applied settings, yet explicit structural models that allow full identification of causal DAGs from observational data remain limited. The Poisson branching structural causal model (PB-SCM) provides a count-valued analogue of linear structural equation models using binomial thinning and independent Poisson exogenous variables, but its causal DAG is generally only partially identifiable. Building on this framework, we propose the Poisson thinning structural equation model (PT-SEM), which replaces binomial thinning in PB-SCM with Poisson thinning and allows node-wise exogenous distributions from diverse count-distribution families. Under node-wise regularity conditions, we establish identifiability of the causal DAG, the thinning coefficients, and the node-wise exogenous distributions. The same identification analysis extends to binomial thinning, yielding full identifiability whenever every nonsink has non-Poisson exogenous noise. We further develop a structure learning algorithm that optimizes, via dynamic programming, a BIC score based on local likelihoods evaluated at plug-in moment estimates, and establish its consistency for DAG selection. Simulations demonstrate favorable performance in DAG recovery and thinning-coefficient estimation, and a real-data application illustrates the practical utility of PT-SEM.

stat.ME↗

arXiv2026

When Less is More: Understanding When Token Filtering Helps and Fails in AI-generated Text Detection

The rapid advancement of large language models (LLMs) has made AI-generated text detection increasingly critical. Existing zero-shot detectors assume that more token-level evidence leads to more reliable detection. However, our empirical study challenges this consensus: fewer tokens sometimes work better, retaining only 40% can yield optimal performance, yet this benefit is not universal. Using the Entropy Gap Score (EGS), we introduce top-$k$ cumulative probability filtering as a diagnostic probe. Across three representative settings, filtering exhibits strikingly different behaviors. We analyze EGS via typical set theory and quantify its dynamics through entropy calibration and distribution analysis. We find that filtering helps for weak source LMs, where low-entropy tokens are harmful, but fails for strong source LMs, where they are not notably harmful. Our work provides the first systematic analysis showing that some tokens are not merely uninformative but systematically harmful due to entropy miscalibration, revealing a two-sided trade-off in token-level detection.

cs.CL↗

arXiv2026

Learning Sparsest Linear Causal DAGs with Latent Confounders via Higher-Order Cumulants

Recovering the exact directed acyclic graph (DAG) in linear non-Gaussian acyclic models with latent confounders (LvLiNGAM) remains a challenging problem. Although LvLiNGAM is identifiable only up to an observational equivalence class, each equivalence class is characterized by a unique sparsest DAG. Recovering the sparsest DAG from finite samples, however, remains difficult. Although existing methods are asymptotically consistent, they do not provide an explicit finite-sample procedure for recovering the unique sparsest DAG, nor do they handle models with an arbitrary number of latent confounders. In this paper, we propose a finite-sample method for recovering the sparsest DAG without imposing any restriction on the number of latent confounders. Simulation studies and real-data analyses demonstrate that the proposed method achieves superior finite-sample performance compared with existing approaches.

cs.LG↗

arXiv2026

Conservation-Based Feedback-Circuit Decomposition for Linear Forced Systems

We present a conservation-based feedback-circuit decomposition specifically for general linear forced systems. In a role parallel to that of eigenvalues and eigenvectors for initial-value problems, the complete set of independent intrinsic circuit gains and their associated forcing-transformation vectors provide a complete analytical representation of both transient and equilibrium forced solutions. The sign of intrinsic circuit gains determines whether successive feedback cycles exhibit monotonic or oscillatory convergence to transformed forcing, while the forcing-transformation vectors determine the structure of transformed forcing. The exact transient and equilibrium solutions are represented analytically through the convergence of the finite-cycle forcing-transformation kernel to the equilibrium forcing-transformation kernel, which is guaranteed regardless of whether the magnitudes of circuit gains exceed one or unstable modes exist in the system. The feedback-circuit decomposition provides a new generic foundational mathematical tool for understanding, predicting, and controlling forced responses in a broad range of coupled linear systems across science and engineering.

physics.ao-ph↗

arXiv2026

Whisper-MLA: Reducing GPU Memory Consumption of ASR Models based on MHA2MLA Conversion

The Transformer-based Whisper model has achieved state-of-the-art performance in Automatic Speech Recognition (ASR). However, its Multi-Head Attention (MHA) mechanism results in significant GPU memory consumption due to the linearly growing Key-Value (KV) cache usage, which is problematic for many applications especially with long-form audio. To address this, we introduce Whisper-MLA, a novel architecture that incorporates Multi-Head Latent Attention (MLA) into the Whisper model. Specifically, we adapt MLA for Whisper's absolute positional embeddings and systematically investigate its application across encoder self-attention, decoder self-attention, and cross-attention modules. Empirical results indicate that applying MLA exclusively to decoder self-attention yields the desired balance between performance and memory efficiency. Our proposed approach allows conversion of a pretrained Whisper model to Whisper-MLA with minimal fine-tuning. Extensive experiments on the LibriSpeech benchmark validate the effectiveness of this conversion, demonstrating that Whisper-MLA reduces the KV cache size by up to 87.5% while maintaining competitive accuracy.

cs.SD↗

arXiv2026

Talk Less, Verify More: Improving LLM Assistants with Semantic Checks and Execution Feedback

As large language model (LLM) assistants become increasingly integrated into enterprise workflows, their ability to generate accurate, semantically aligned, and executable outputs is critical. However, current conversational business analytics (CBA) systems often lack built-in verification mechanisms, leaving users to manually validate potentially flawed results. This paper introduces two complementary verification techniques: Q*, which performs reverse translation and semantic matching between code and user intent, and Feedback+, which incorporates execution feedback to guide code refinement. Embedded within a generator-discriminator framework, these mechanisms shift validation responsibilities from users to the system. Evaluations on three benchmark datasets, Spider, Bird, and GSM8K, demonstrate that both Q* and Feedback+ reduce error rates and task completion time. The study also identifies reverse translation as a key bottleneck, highlighting opportunities for future improvement. Overall, this work contributes a design-oriented framework for building more reliable, enterprise-grade GenAI systems capable of trustworthy decision support.

cs.CL↗

arXiv2025

LOOPRAG: Enhancing Loop Transformation Optimization with Retrieval-Augmented Large Language Models

Loop transformations are semantics-preserving optimization techniques, widely used to maximize objectives such as parallelism. Despite decades of research, applying the optimal composition of loop transformations remains challenging due to inherent complexities, including cost modeling for optimization objectives. Recent studies have explored the potential of Large Language Models (LLMs) for code optimization. However, our key observation is that LLMs often struggle with effective loop transformation optimization, frequently leading to errors or suboptimal optimization, thereby missing opportunities for performance improvements. To bridge this gap, we propose LOOPRAG, a novel retrieval-augmented generation framework designed to guide LLMs in performing effective loop optimization on Static Control Part. We introduce a parameter-driven method to harness loop properties, which trigger various loop transformations, and generate diverse yet legal example codes serving as a demonstration source. To effectively obtain the most informative demonstrations, we propose a loop-aware algorithm based on loop features, which balances similarity and diversity for code retrieval. To enhance correct and efficient code generation, we introduce a feedback-based iterative mechanism that incorporates compilation, testing and performance results as feedback to guide LLMs. Each optimized code undergoes mutation, coverage and differential testing for equivalence checking. We evaluate LOOPRAG on PolyBench, TSVC and LORE benchmark suites, and compare it against compilers (GCC-Graphite, Clang-Polly, Perspective and ICX) and representative LLMs (DeepSeek and GPT-4). The results demonstrate average speedups over base compilers of up to 11.20$\times$, 14.34$\times$, and 9.29$\times$ for PolyBench, TSVC, and LORE, respectively, and speedups over base LLMs of up to 11.97$\times$, 5.61$\times$, and 11.59$\times$.

cs.PL↗

arXiv2025

Causal Discovery for Linear DAGs with Dependent Latent Variables via Higher-order Cumulants

This paper addresses the problem of estimating causal directed acyclic graphs in linear non-Gaussian acyclic models with latent confounders (LvLiNGAM). Existing methods assume mutually independent latent confounders or cannot properly handle models with causal relationships among observed variables. We propose a novel algorithm that identifies causal DAGs in LvLiNGAM, allowing causal structures among latent variables, among observed variables, and between the two. The proposed method leverages higher-order cumulants of observed data to identify the causal structure. Extensive simulations and experiments with real-world data demonstrate the validity and practical utility of the proposed algorithm.

cs.LG↗

arXiv2025

DKGCM: A Spatio-Temporal Prediction Model for Traffic Flow by Fusing Spatial Node Clustering Method and Fourier Bidirectional Mamba Mechanism

Accurate traffic demand forecasting enables transportation management departments to allocate resources more effectively, thereby improving their utilization efficiency. However, complex spatiotemporal relationships in traffic systems continue to limit the performance of demand forecasting models. To improve the accuracy of spatiotemporal traffic demand prediction, we propose a new graph convolutional network structure called DKGCM. Specifically, we first consider the spatial flow distribution of different traffic nodes and propose a novel temporal similarity-based clustering graph convolution method, DK-GCN. This method utilizes Dynamic Time Warping (DTW) and K-means clustering to group traffic nodes and more effectively capture spatial dependencies. On the temporal scale, we integrate the Fast Fourier Transform (FFT) within the bidirectional Mamba deep learning framework to capture temporal dependencies in traffic demand. To further optimize model training, we incorporate the GRPO reinforcement learning strategy to enhance the loss function feedback mechanism. Extensive experiments demonstrate that our model outperforms several advanced methods and achieves strong results on three public datasets.

cs.LG↗

arXiv2025

Wave Energy Is Conserved in a Spatially Varying and Inhomogeneously Moving Medium

Waves are propagating disturbances that redistribute energy across space. Previous studies have shown that for waves propagating through an inhomogeneously moving mean flow, the conserved quantity is wave action rather than wave energy, raising questions about the validity of energy conservation, which is one of the foundational principles of physics. In this study, we prove that wave action conservation is, in fact, an apparent form of wave energy conservation in spatially varying and inhomogeneously moving media, where waves undergo deformation during propagation. We further show that wave action conservation can be derived directly from the law of energy conservation. This result holds universally across all isolated wave systems in varying media, including hydrodynamic and non-hydrodynamic waves.

physics.flu-dyn↗

arXiv2025

The Ceiling Height of Wildland Fire Plumes in Sheared Boundary Layer Flow

Radar observations from a prescribed fire experiment reveal a large-scale, billow-like vorticity pair associated with the plume head at the onset of plume bending. The bending confines the ceiling height of the plume, delaying its smoke dispersion and increasing fire spotting risks. This study aims to investigate the onset of plume bending in a sheared crossflow and stratified atmospheric conditions, providing insights into smoke dispersion and fire behavior. Large Eddy Simulations (LES) using the Cloud Model 1 (CM1) are conducted to simulate the observed development of plume structure and its associated dynamical fields, with particular focus on the plume head and its evolution from initial plume development under different fire intensities and atmospheric boundary layer (ABL) conditions. A scaling analysis of plume ceiling height is proposed based on a modified Byram's convective number that accounts for sheared crossflow. The proposed scaling agrees well with the LES results, highlighting the critical roles of shear and stratification in controlling plume dynamics.

physics.flu-dyn↗

arXiv2024

Learning linear acyclic causal model including Gaussian noise using ancestral relationships

This paper discusses algorithms for learning causal DAGs. The PC algorithm makes no assumptions other than the faithfulness to the causal model and can identify only up to the Markov equivalence class. LiNGAM assumes linearity and continuous non-Gaussian disturbances for the causal model, and the causal DAG defining LiNGAM is shown to be fully identifiable. The PC-LiNGAM, a hybrid of the PC algorithm and LiNGAM, can identify up to the distribution-equivalence pattern of a linear causal model, even in the presence of Gaussian disturbances. However, in the worst case, the PC-LiNGAM has factorial time complexity for the number of variables. In this paper, we propose an algorithm for learning the distribution-equivalence patterns of a linear causal model with a lower time complexity than PC-LiNGAM, using the causal ancestor finding algorithm in Maeda and Shimizu, which is generalized to account for Gaussian disturbances.

cs.LG↗

arXiv2024

From Graph to Word Bag: Introducing Domain Knowledge to Confusing Charge Prediction

Confusing charge prediction is a challenging task in legal AI, which involves predicting confusing charges based on fact descriptions. While existing charge prediction methods have shown impressive performance, they face significant challenges when dealing with confusing charges, such as Snatch and Robbery. In the legal domain, constituent elements play a pivotal role in distinguishing confusing charges. Constituent elements are fundamental behaviors underlying criminal punishment and have subtle distinctions among charges. In this paper, we introduce a novel From Graph to Word Bag (FWGB) approach, which introduces domain knowledge regarding constituent elements to guide the model in making judgments on confusing charges, much like a judge's reasoning process. Specifically, we first construct a legal knowledge graph containing constituent elements to help select keywords for each charge, forming a word bag. Subsequently, to guide the model's attention towards the differentiating information for each charge within the context, we expand the attention mechanism and introduce a new loss function with attention supervision through words in the word bag. We construct the confusing charges dataset from real-world judicial documents. Experiments demonstrate the effectiveness of our method, especially in maintaining exceptional performance in imbalanced label distributions.

cs.AI↗

arXiv2024

Learning causal graphs using variable grouping according to ancestral relationship

Several causal discovery algorithms have been proposed. However, when the sample size is small relative to the number of variables, the accuracy of estimating causal graphs using existing methods decreases. And some methods are not feasible when the sample size is smaller than the number of variables. To circumvent these problems, some researchers proposed causal structure learning algorithms using divide-and-conquer approaches. For learning the entire causal graph, the approaches first split variables into several subsets according to the conditional independence relationships among the variables, then apply a conventional causal discovery algorithm to each subset and merge the estimated results. Since the divide-and-conquer approach reduces the number of variables to which a causal structure learning algorithm is applied, it is expected to improve the estimation accuracy of causal graphs, especially when the sample size is small relative to the number of variables and the model is sparse. However, existing methods are either computationally expensive or do not provide sufficient accuracy when the sample size is small. This paper proposes a new algorithm for grouping variables based the ancestral relationships among the variables, under the LiNGAM assumption, where the causal relationships are linear, and the mutually independent noise are distributed as continuous non-Gaussian distributions. We call the proposed algorithm CAG. The time complexity of the ancestor finding in CAG is shown to be cubic to the number of variables. Extensive computer experiments confirm that the proposed method outperforms the original DirectLiNGAM without grouping variables and other divide-and-conquer approaches not only in estimation accuracy but also in computation time when the sample size is small relative to the number of variables and the model is sparse.

stat.ML↗

arXiv2024

Enhancing Court View Generation with Knowledge Injection and Guidance

Court View Generation (CVG) is a challenging task in the field of Legal Artificial Intelligence (LegalAI), which aims to generate court views based on the plaintiff claims and the fact descriptions. While Pretrained Language Models (PLMs) have showcased their prowess in natural language generation, their application to the complex, knowledge-intensive domain of CVG often reveals inherent limitations. In this paper, we present a novel approach, named Knowledge Injection and Guidance (KIG), designed to bolster CVG using PLMs. To efficiently incorporate domain knowledge during the training stage, we introduce a knowledge-injected prompt encoder for prompt tuning, thereby reducing computational overhead. Moreover, to further enhance the model's ability to utilize domain knowledge, we employ a generating navigator, which dynamically guides the text generation process in the inference stage without altering the model's architecture, making it readily transferable. Comprehensive experiments on real-world data demonstrate the effectiveness of our approach compared to several established baselines, especially in the responsivity of claims, where it outperforms the best baseline by 11.87%.

cs.AI↗

arXiv2023

Impact of Indoor Mobility Behavior on the Respiratory Infectious Diseases Transmission Trends

The importance of indoor human mobility in the transmission dynamics of respiratory infectious diseases has been acknowledged. Previous studies have predominantly addressed a single type of mobility behavior such as queueing and a series of behaviors under specific scenarios. However, these studies ignore the abstraction of mobility behavior in various scenes and the critical examination of how these abstracted behaviors impact disease propagation. To address these problems, this study considers people's mobility behaviors in a general scenario, abstracting them into two main categories: crowding behavior, related to the spatial aspect, and stopping behavior, related to the temporal aspect. Accordingly, this study investigates their impacts on disease spreading and the impact of individual spatio-temporal distribution resulting from these mobility behaviors on epidemic transmission. First, a point of interest (POI) method is introduced to quantify the crowding-related spatial POI factors (i.e., the number of crowdings and the distance between crowdings) and stopping-related temporal POI factors (i.e., the number of stoppings and the duration of each stopping). Besides, a personal space determined with Voronoi diagrams is used to construct the individual spatio-temporal distribution factor. Second, two indicators (i.e., the daily number of new cases and the average exposure risk of people) are applied to quantify epidemic transmission. These indicators are derived from a fundamental model which accurately predicts disease transmission between moving individuals. Third, a set of 200 indoor scenarios is constructed and simulated to help determine variable values. Concurrently, the influences and underlying mechanisms of these behavioral factors on disease transmission are examined using structural equation modeling and causal inference modeling......

cs.CY↗

arXiv2023

Enhancing Language Representation with Constructional Information for Natural Language Understanding

Natural language understanding (NLU) is an essential branch of natural language processing, which relies on representations generated by pre-trained language models (PLMs). However, PLMs primarily focus on acquiring lexico-semantic information, while they may be unable to adequately handle the meaning of constructions. To address this issue, we introduce construction grammar (CxG), which highlights the pairings of form and meaning, to enrich language representation. We adopt usage-based construction grammar as the basis of our work, which is highly compatible with statistical models such as PLMs. Then a HyCxG framework is proposed to enhance language representation through a three-stage solution. First, all constructions are extracted from sentences via a slot-constraints approach. As constructions can overlap with each other, bringing redundancy and imbalance, we formulate the conditional max coverage problem for selecting the discriminative constructions. Finally, we propose a relational hypergraph attention network to acquire representation from constructional information by capturing high-order word interactions among constructions. Extensive experiments demonstrate the superiority of the proposed model on a variety of NLU tasks.

cs.CL↗

arXiv2023

Exploring assessment method of technological advancement based on literature cross-citation

Assessing advancements of technology is essential for creating science and technology policies and making informed investments in the technology market. However, current methods primarily focus on the characteristics of the technologies themselves, making it difficult to accurately assess technologies across various fields and generations. To address this challenge, we propose a novel approach that uses bibliometrics, specifically literature citation networks, to measure changes in knowledge flow throughout the evolution of technology. This method can identify diverse trends in technology development and is an effective tool for evaluating technological advancements. We demonstrate its accuracy and applicability by applying it to mobile communication technology and comparing its quantitative results with other assessment methods. Our work provides critical support for assessing different technical routes and formulating technology policy.

stat.AP↗