arXiv subjects

Suqi Liu

Publications and source records attributed to Suqi Liu.

9 recordsLinked to original sources

arXiv2026

Toward a Unified Statistical Theory of Unsupervised Pretraining and Supervised Neural Knowledge Graph Learning

Knowledge graph learning provides a powerful framework for representing and inferring structured knowledge, with broad practical applications. However, the scarcity of relation-specific labeled triples per entity hinders the training of expressive models, and the ad hoc design of scoring functions limits generalizability and lacks theoretical grounding. We address both issues with a theoretically grounded, end-to-end training framework that extends and subsumes existing methods. Our framework is a two-stage procedure: unsupervised pretraining over heterogeneous corpora followed by supervised learning with multiple relation types. We establish a nonasymptotic risk bound that disentangles pretraining representation error from labeled-sample complexity, formally quantifying the benefit of large-scale unlabeled data for downstream knowledge prediction. Synthetic experiments validate each theoretical component, and real-world experiments confirm the effectiveness of our approach on large-scale knowledge graph benchmarks.

math.ST↗

arXiv2026

Graph Attention Network for Node Regression on Random Geometric Graphs with Erd\H{o}s--R\'enyi contamination

Graph attention networks (GATs) are widely used and often appear robust to noise in node covariates and edges, yet rigorous statistical guarantees demonstrating a provable advantage of GATs over non-attention graph neural networks~(GNNs) are scarce. We partially address this gap for node regression with graph-based errors-in-variables models under simultaneous covariate and edge corruption: responses are generated from latent node-level covariates, but only noise-perturbed versions of the latent covariates are observed; and the sample graph is a random geometric graph created from the node covariates but contaminated by independent Erd\H{o}s--R\'enyi edges. We propose and analyze a carefully designed, task-specific GAT that constructs denoised proxy features for regression. We prove that regressing the response variables on the proxies achieves lower error asymptotically in (a) estimating the regression coefficient compared to the ordinary least squares (OLS) estimator on the noisy node covariates, and (b) predicting the response for an unlabelled node compared to a vanilla graph convolutional network~(GCN) -- under mild growth conditions. Our analysis leverages high-dimensional geometric tail bounds and concentration for neighbourhood counts and sample covariances. We verify our theoretical findings through experiments on synthetically generated data. We also perform experiments on real-world graphs and demonstrate the effectiveness of the attention mechanism in several node regression tasks.

stat.ML↗

arXiv2025

Crystal structures and electronic and magnetic properties of Janus bilayer Cl3Cr2I3

Two-dimensional (2D) magnetic material CrI3 has aroused extensive attention, because it could provide a new platform for investigating the relations between crystal structures and electronic and magnetic properties. Here, we study crystal structures and electronic and magnetic properties of three configurations (Cl-Cl I-I and Cl-I) of Janus bilayer Cl3Cr2I3 with two stacking orders (AB and AA1=3) by using the first principles approach incorporating the spin-orbit coupling (SOC) effect and the dipole correction. It is found that the spin polarization and the SOC effect can expand the lattice constant and the interlayer distance (ID) of the three configurations. Especially, the ID of the I-I configuration is 1 {\AA} larger than that of the Cl-Cl configuration. The total energy calculation results show that the atomic configuration, the SOC effect and the stacking order play an important role in determining the magnetic ground states of the Janus layers. Our results indicate the atomic configurations of the Janus bilayers are not conducive to the increase of critical temperature. Interestingly, the Cl-I configuration has a vertical dipole moment, and its AFM state has large spin splitting. It is revealed that the non-periodic structure, the symmetry breaking of the average potential and the weak interlayer interaction lead to the vertical dipole moment and the abnormal AFM state that is not the most stable state.

cond-mat.mtrl-sci↗

arXiv2025

Representation learning to advance multi-institutional studies with electronic health record data from US and France

The widespread adoption of electronic health records has created new opportunities for translational clinical research, yet this promise remains constrained by fragmented data across privacy-siloed institutions and substantial heterogeneity in local coding practices. While privacy-preserving collaborative learning allows institutions to work together without sharing patient-level data, it does not address inconsistencies in how clinical concepts are represented across sites. We introduce a graph-based framework that addresses this gap by treating data harmonization as a scalable representation learning problem. Rather than relying on fixed standards or manual mappings, the framework integrates institution-specific summary statistics from health records, curated biomedical knowledge graphs, and semantic information derived from large language models to learn a shared semantic space. This joint learning approach aligns diverse, site-specific vocabularies while preserving patient privacy. Evaluated across seven institutions and two languages, the framework provides a robust, data-centric foundation for training and deploying clinical models across heterogeneous healthcare systems.

cs.AI↗

arXiv2024

Unified Representation of Genomic and Biomedical Concepts through Multi-Task, Multi-Source Contrastive Learning

We introduce GENomic Encoding REpresentation with Language Model (GENEREL), a framework designed to bridge genetic and biomedical knowledge bases. What sets GENEREL apart is its ability to fine-tune language models to infuse biological knowledge behind clinical concepts such as diseases and medications. This fine-tuning enables the model to capture complex biomedical relationships more effectively, enriching the understanding of how genomic data connects to clinical outcomes. By constructing a unified embedding space for biomedical concepts and a wide range of common SNPs from sources such as patient-level data, biomedical knowledge graphs, and GWAS summaries, GENEREL aligns the embeddings of SNPs and clinical concepts through multi-task contrastive learning. This allows the model to adapt to diverse natural language representations of biomedical concepts while bypassing the limitations of traditional code mapping systems across different data sources. Our experiments demonstrate GENEREL's ability to effectively capture the nuanced relationships between SNPs and clinical concepts. GENEREL also emerges to discern the degree of relatedness, potentially allowing for a more refined identification of concepts. This pioneering approach in constructing a unified embedding system for both SNPs and biomedical concepts enhances the potential for data integration and discovery in biomedical research.

cs.LG↗

arXiv2024

Representation-Enhanced Neural Knowledge Integration with Application to Large-Scale Medical Ontology Learning

A large-scale knowledge graph enhances reproducibility in biomedical data discovery by providing a standardized, integrated framework that ensures consistent interpretation across diverse datasets. It improves generalizability by connecting data from various sources, enabling broader applicability of findings across different populations and conditions. Generating reliable knowledge graph, leveraging multi-source information from existing literature, however, is challenging especially with a large number of node sizes and heterogeneous relations. In this paper, we propose a general theoretically guaranteed statistical framework, called RENKI, to enable simultaneous learning of multiple relation types. RENKI generalizes various network models widely used in statistics and computer science. The proposed framework incorporates representation learning output into initial entity embedding of a neural network that approximates the score function for the knowledge graph and continuously trains the model to fit observed facts. We prove nonasymptotic bounds for in-sample and out-of-sample weighted MSEs in relation to the pseudo-dimension of the knowledge graph function class. Additionally, we provide pseudo-dimensions for score functions based on multilayer neural networks with ReLU activation function, in the scenarios when the embedding parameters either fixed or trainable. Finally, we complement our theoretical results with numerical studies and apply the method to learn a comprehensive medical knowledge graph combining a pretrained language model representation with knowledge graph links observed in several medical ontologies. The experiments justify our theoretical findings and demonstrate the effect of weighting in the presence of heterogeneous relations and the benefit of incorporating representation learning in nonparametric models.

stat.ME↗

arXiv2024

Perfect Recovery for Random Geometric Graph Matching with Shallow Graph Neural Networks

We study the graph matching problem in the presence of vertex feature information using shallow graph neural networks. Specifically, given two graphs that are independent perturbations of a single random geometric graph with sparse binary features, the task is to recover an unknown one-to-one mapping between the vertices of the two graphs. We show under certain conditions on the sparsity and noise level of the feature vectors, a carefully designed two-layer graph neural network can, with high probability, recover the correct mapping between the vertices with the help of the graph structure. Additionally, we prove that our condition on the noise parameter is tight up to logarithmic factors. Finally, we compare the performance of the graph neural network to directly solving an assignment problem using the noisy vertex features and demonstrate that when the noise level is at least constant, this direct matching fails to achieve perfect recovery, whereas the graph neural network can tolerate noise levels growing as fast as a power of the size of the graph. Our theoretical findings are further supported by numerical studies as well as real-world data experiments.

cs.LG↗

arXiv2021

A probabilistic view of latent space graphs and phase transitions

We study random graphs with latent geometric structure, where the probability of each edge depends on the underlying random positions corresponding to the two endpoints. We focus on the setting where this conditional probability is a general monotone increasing function of the inner product of two vectors; such a function can naturally be viewed as the cumulative distribution function of some independent random variable. We consider a one-parameter family of random graphs, characterized by the variance of this random variable, that smoothly interpolates between a random dot product graph and an Erdős--Rényi random graph. We prove phase transitions of detecting geometry in these graphs, in terms of the dimension of the underlying geometric space and the variance parameter of the conditional probability. When the dimension is high or the variance is large, the graph is similar to an Erdős--Rényi graph with the same edge density that does not possess geometry; in other parameter regimes, there is a computationally efficient signed triangle statistic that distinguishes them. The proofs make use of information-theoretic inequalities and concentration of measure phenomena.

math.PR↗

arXiv2021

Phase transition in noisy high-dimensional random geometric graphs

We study the problem of detecting latent geometric structure in random graphs. To this end, we consider the soft high-dimensional random geometric graph $\mathcal{G}(n,p,d,q)$, where each of the $n$ vertices corresponds to an independent random point distributed uniformly on the sphere $\mathbb{S}^{d-1}$, and the probability that two vertices are connected by an edge is a decreasing function of the Euclidean distance between the points. The probability of connection is parametrized by $q \in [0,1]$, with smaller $q$ corresponding to weaker dependence on the geometry; this can also be interpreted as the level of noise in the geometric graph. In particular, the model smoothly interpolates between the spherical hard random geometric graph $\mathcal{G}(n,p,d)$ (corresponding to $q = 1$) and the Erdős-Rényi model $\mathcal{G}(n,p)$ (corresponding to $q = 0$). We focus on the dense regime (i.e., $p$ is a constant). We show that if $nq \to 0$ or $d \gg n^{3} q^{2}$, then geometry is lost: $\mathcal{G}(n,p,d,q)$ is asymptotically indistinguishable from $\mathcal{G}(n,p)$. On the other hand, if $d \ll n^{3} q^{6}$, then the signed triangle statistic provides an asymptotically powerful test for detecting geometry. These results generalize those of Bubeck, Ding, Eldan, and Rácz (2016) for $\mathcal{G}(n,p,d)$, and give quantitative bounds on how the noise level affects the dimension threshold for losing geometry. We also prove analogous results under a related but different distributional assumption, and we further explore generalizations of signed triangles in order to understand the intermediate regime left open by our results.

math.PR↗