arXiv subjects

Lihui Liu

Publications and source records attributed to Lihui Liu.

35 records · Page 2Linked to original sources

arXiv2025

THeGCN: Temporal Heterophilic Graph Convolutional Network

Graph Neural Networks (GNNs) have exhibited remarkable efficacy in diverse graph learning tasks, particularly on static homophilic graphs. Recent attention has pivoted towards more intricate structures, encompassing (1) static heterophilic graphs encountering the edge heterophily issue in the spatial domain and (2) event-based continuous graphs in the temporal domain. State-of-the-art (SOTA) has been concurrently addressing these two lines of work but tends to overlook the presence of heterophily in the temporal domain, constituting the temporal heterophily issue. Furthermore, we highlight that the edge heterophily issue and the temporal heterophily issue often co-exist in event-based continuous graphs, giving rise to the temporal edge heterophily challenge. To tackle this challenge, this paper first introduces the temporal edge heterophily measurement. Subsequently, we propose the Temporal Heterophilic Graph Convolutional Network (THeGCN), an innovative model that incorporates the low/high-pass graph signal filtering technique to accurately capture both edge (spatial) heterophily and temporal heterophily. Specifically, the THeGCN model consists of two key components: a sampler and an aggregator. The sampler selects events relevant to a node at a given moment. Then, the aggregator executes message-passing, encoding temporal information, node attributes, and edge attributes into node embeddings. Extensive experiments conducted on 5 real-world datasets validate the efficacy of THeGCN.

cs.LG↗

arXiv2024

PyG-SSL: A Graph Self-Supervised Learning Toolkit

Graph Self-Supervised Learning (SSL) has emerged as a pivotal area of research in recent years. By engaging in pretext tasks to learn the intricate topological structures and properties of graphs using unlabeled data, these graph SSL models achieve enhanced performance, improved generalization, and heightened robustness. Despite the remarkable achievements of these graph SSL methods, their current implementation poses significant challenges for beginners and practitioners due to the complex nature of graph structures, inconsistent evaluation metrics, and concerns regarding reproducibility hinder further progress in this field. Recognizing the growing interest within the research community, there is an urgent need for a comprehensive, beginner-friendly, and accessible toolkit consisting of the most representative graph SSL algorithms. To address these challenges, we present a Graph SSL toolkit named PyG-SSL, which is built upon PyTorch and is compatible with various deep learning and scientific computing backends. Within the toolkit, we offer a unified framework encompassing dataset loading, hyper-parameter configuration, model training, and comprehensive performance evaluation for diverse downstream tasks. Moreover, we provide beginner-friendly tutorials and the best hyper-parameters of each graph SSL algorithm on different graph datasets, facilitating the reproduction of results. The GitHub repository of the library is https://github.com/iDEA-iSAIL-Lab-UIUC/pyg-ssl.

cs.LG↗

arXiv2024

Logic Query of Thoughts: Guiding Large Language Models to Answer Complex Logic Queries with Knowledge Graphs

Despite the superb performance in many tasks, large language models (LLMs) bear the risk of generating hallucination or even wrong answers when confronted with tasks that demand the accuracy of knowledge. The issue becomes even more noticeable when addressing logic queries that require multiple logic reasoning steps. On the other hand, knowledge graph (KG) based question answering methods are capable of accurately identifying the correct answers with the help of knowledge graph, yet its accuracy could quickly deteriorate when the knowledge graph itself is sparse and incomplete. It remains a critical challenge on how to integrate knowledge graph reasoning with LLMs in a mutually beneficial way so as to mitigate both the hallucination problem of LLMs as well as the incompleteness issue of knowledge graphs. In this paper, we propose 'Logic-Query-of-Thoughts' (LGOT) which is the first of its kind to combine LLMs with knowledge graph based logic query reasoning. LGOT seamlessly combines knowledge graph reasoning and LLMs, effectively breaking down complex logic queries into easy to answer subquestions. Through the utilization of both knowledge graph reasoning and LLMs, it successfully derives answers for each subquestion. By aggregating these results and selecting the highest quality candidate answers for each step, LGOT achieves accurate results to complex questions. Our experimental findings demonstrate substantial performance enhancements, with up to 20% improvement over ChatGPT.

cs.IR↗

arXiv2024

Meta Clustering of Neural Bandits

The contextual bandit has been identified as a powerful framework to formulate the recommendation process as a sequential decision-making process, where each item is regarded as an arm and the objective is to minimize the regret of $T$ rounds. In this paper, we study a new problem, Clustering of Neural Bandits, by extending previous work to the arbitrary reward function, to strike a balance between user heterogeneity and user correlations in the recommender system. To solve this problem, we propose a novel algorithm called M-CNB, which utilizes a meta-learner to represent and rapidly adapt to dynamic clusters, along with an informative Upper Confidence Bound (UCB)-based exploration strategy. We provide an instance-dependent performance guarantee for the proposed algorithm that withstands the adversarial context, and we further prove the guarantee is at least as good as state-of-the-art (SOTA) approaches under the same assumptions. In extensive experiments conducted in both recommendation and online classification scenarios, M-CNB outperforms SOTA baselines. This shows the effectiveness of the proposed approach in improving online recommendation and online classification performance.

cs.LG↗

arXiv2024

Can Contrastive Learning Refine Embeddings

Recent advancements in contrastive learning have revolutionized self-supervised representation learning and achieved state-of-the-art performance on benchmark tasks. While most existing methods focus on applying contrastive learning to input data modalities such as images, natural language sentences, or networks, they overlook the potential of utilizing outputs from previously trained encoders. In this paper, we introduce SIMSKIP, a novel contrastive learning framework that specifically refines input embeddings for downstream tasks. Unlike traditional unsupervised learning approaches, SIMSKIP takes advantage of the output embeddings of encoder models as its input. Through theoretical analysis, we provide evidence that applying SIMSKIP does not result in larger upper bounds on downstream task errors than those of the original embeddings, which serve as SIMSKIP's input. Experimental results on various open datasets demonstrate that the embeddings produced by SIMSKIP improve performance on downstream tasks.

cs.LG↗

arXiv2024

Conversational Question Answering with Reformulations over Knowledge Graph

Conversational question answering (convQA) over knowledge graphs (KGs) involves answering multi-turn natural language questions about information contained in a KG. State-of-the-art methods of ConvQA often struggle with inexplicit question-answer pairs. These inputs are easy for human beings to understand given a conversation history, but hard for a machine to interpret, which can degrade ConvQA performance. To address this problem, we propose a reinforcement learning (RL) based model, CornNet, which utilizes question reformulations generated by large language models (LLMs) to improve ConvQA performance. CornNet adopts a teacher-student architecture where a teacher model learns question representations using human writing reformulations, and a student model to mimic the teacher model's output via reformulations generated by LLMs. The learned question representation is then used by an RL model to locate the correct answer in a KG. Extensive experimental results show that CornNet outperforms state-of-the-art convQA models.

cs.CL↗

arXiv2023

Neural Multi-network Diffusion towards Social Recommendation

Graph Neural Networks (GNNs) have been widely applied on a variety of real-world applications, such as social recommendation. However, existing GNN-based models on social recommendation suffer from serious problems of generalization and oversmoothness, because of the underexplored negative sampling method and the direct implanting of the off-the-shelf GNN models. In this paper, we propose a succinct multi-network GNN-based neural model (NeMo) for social recommendation. Compared with the existing methods, the proposed model explores a generative negative sampling strategy, and leverages both the positive and negative user-item interactions for users' interest propagation. The experiments show that NeMo outperforms the state-of-the-art baselines on various real-world benchmark datasets (e.g., by up to 38.8% in terms of NDCG@15).

cs.LG↗

arXiv2020

KompaRe: A Knowledge Graph Comparative Reasoning System

Reasoning is a fundamental capability for harnessing valuable insight, knowledge and patterns from knowledge graphs. Existing work has primarily been focusing on point-wise reasoning, including search, link predication, entity prediction, subgraph matching and so on. This paper introduces comparative reasoning over knowledge graphs, which aims to infer the commonality and inconsistency with respect to multiple pieces of clues. We envision that the comparative reasoning will complement and expand the existing point-wise reasoning over knowledge graphs. In detail, we develop KompaRe, the first of its kind prototype system that provides comparative reasoning capability over large knowledge graphs. We present both the system architecture and its core algorithms, including knowledge segment extraction, pairwise reasoning and collective reasoning. Empirical evaluations demonstrate the efficacy of the proposed KompaRe.

cs.AI↗

arXiv2019

SuRF: a New Method for Sparse Variable Selection, with Application in Microbiome Data Analysis

In this paper, we present a new variable selection method for regression and classification purposes. Our method, called Subsampling Ranking Forward selection (SuRF), is based on LASSO penalised regression, subsampling and forward-selection methods. SuRF offers major advantages over existing variable selection methods in terms of both sparsity of selected models and model inference. We provide an R package that can implement our method for generalized linear models. We apply our method to classification problems from microbiome data, using a novel agglomeration approach to deal with the special tree-like correlation structure of the variables. Existing methods arbitrarily choose a taxonomic level a priori before performing the analysis, whereas by combining SuRF with these aggregated variables, we are able to identify the key biomarkers at the appropriate taxonomic level, as suggested by the data. We present simulations in multiple sparse settings to demonstrate that our approach performs better than several other popularly used existing approaches in recovering the true variables. We apply SuRF to two microbiome data sets: one about prediction of pouchitis and another for identifying samples from two healthy individuals. We find that SuRF can provide a better or comparable prediction with other methods while controlling the false positive rate of variable selection.

stat.ME↗

arXiv2014

Holographic renormalization in no-boundary quantum cosmology

Recent results of Hartle-Hawking wave functions on asymptotic dS boundaries show non-normalizability, while the bulk origin is not clear. This paper attempts to addresse this problem by studying (Kerr-)dS_3 cosmology in Einstein gravity deformed by a minimally coupled scalar field. Various saddle-point contributions exp[i(saddle-point action)] to the Hartle-Hawking wave functions are computed with mini-superspace formalism. The saddle-point actions are first obtained in the spacetime bulks by direct computation, then on the asymptotic dS boundary by holographic renormalization. It is found that the imaginary part of the saddle-point actions, as functions of scalar field deformation, are generally bounded in the bulk, but can diverge to -infty on the boundary. This can probably be a source of the scalar field-related non-normalizability of the boundary Hartle-Hawking wave functions. For Kerr-dS_3 cosmology, some saddle-point actions have imaginary parts diverging to -infty both in the bulk and on the boundary, when the boundary T^2 is stretched to infinitely long. This can be an origin of non-normalizability of the boundary Hartle-Hawking wave functions related to temperature divergence. Finally the holographic renormalization is extended to scalar-deformed dS_{d+1} (d>= 3) cosmologies for the imaginary part of the saddle-point actions. The result is tested on a 5d example, and saddle points leading to divergent contribution to the boundary Hartle-Hawking wave function are shown to exist.

hep-th↗

arXiv2014

Lecture notes on thermodynamics of ideal string gases and its application in cosmology

In these lecture notes I give a pedagogical introduction to the thermodynamics of ideal string gases. The computation of thermodynamic quantities in the canonical ensemble formalism will be shown in detail with explicit examples. Attention will be given mainly to the thermodynamical consequences of string degrees of freedom, where I will especially address i) the Hagedorn temperature, a critical temperature above which the canonical ensemble description breaks down, which can be the onset point of some instability of the string gas; ii) the phase structure arising from compactification, embodied in the moduli-dependence of the Helmholtz free energy, which corrects the tree-level vacuum and can provide mechanism for moduli stabilization. Then I will briefly explain the implementation of string gas thermodynamics in cosmology, showing a simple example which gives rise to a radiation-dominated early universe. Further phenomenological issues and open questions will be discussed qualitatively with references indicated, including the Hagedorn instability in the resolution of the initial singularity, moduli stabilization, generation of hierarchy, radiative symmetry breaking and primordial cosmological fluctuations.

hep-th↗

arXiv2012

Moduli stabilization in type II Calabi-Yau compactifications at finite temperature

We consider the type II superstring compactified on Calabi-Yau threefolds at finite temperature. The latter is implemented at the string level by a free action on the Euclidean time circle. We show that all Kahler and complex structure moduli involved in the gauge theories geometrically engineered in the vicinity of singular loci are lifted by the stringy thermal effective potential. The analysis is based on the effective gauged supergravity at low energy, without integrating out the BPS states becoming massless at the singular loci. The universal form of the action in the weak coupling regime at low enough temperature is determined in two cases. Namely the conifold locus, as well as a locus where the internal space develops a genus-g curve of A{N-1} singularities, realizing an SU(N) gauge theory coupled to g hypermultiplets in the adjoint. In general, the favored points of stabilization sit at the intersection of several singular loci. Thus the entire vector multiplet moduli space can be lifted, together with hypermultiplet moduli. The scalars are dynamically stabilized during the cosmological evolution induced by the back-reaction of the thermal effective potential. When the universe expands and the temperature T drops, the scalars converge to minima, with damped oscillations. Moreover, they store an energy density that scales as T^4, which never dominates over radiation. The reason for this is that the mass they acquire at one-loop is of order the temperature scale, which is time-dependent. As an example we analyze the type IIA compactification on a Calabi-Yau space with Hodge numbers h{11}=2 and h{12}=128. In this case, both Kahler moduli are stabilized, where the internal space develops a node and an enhanced SU(2) gauge theory coupled to 2 adjoint hypermultiplets. This shows that in the dual thermal heterotic picture on K3xT^2, the torus modulus and the axio-dilaton are stabilized.

hep-th↗

arXiv2012

Moduli stabilisation in early superstring cosmology

We study moduli stabilization by thermal effects in the cosmological context. The implementation of finite temperature, which spontaneously breaks supersymmetry, induces an effective potential at one loop level. At the points where extra massless states appear in the string spectrum, the potential develops local minima whose depth depends on the temperature. Moduli attracted to these points acquire dynamical masses which decrease with cosmological evolution. This makes the coherent scalar oscillations dilute before nucleosynthesis, and the cosmological moduli problem is avoided. In particular, we study the effective potential induced by a maximally supersymmetric heterotic string gas for spacetime dimension D>=4, and a gas of type II strings compactified on Calabi-Yau three-folds (D=4). In the former case, the local minima of the potential arise at enhanced gauge symmetry points, which can stabilize all moduli but the dilaton. In the latter case, the local minima are reached at the loci where 2-cycles or 3-cycles in the Calabi-Yau space shrink to zero size, accompanied with either conifold transitions or non Abelian gauge symmetries. This stabilizes the type II moduli which characterize the deformation of these shrinking cycles. Moduli stabilization in the dual string models is also investigated by heterotic/type I S-dualities and type II/heterotic dualities.

hep-th↗

arXiv2011

Massless D-strings and moduli stabilization in type I cosmology

We consider the cosmological evolution induced by the free energy F of a gas of maximally supersymmetric heterotic strings at finite temperature and weak coupling in dimension D>=4. We show that F, which plays the role of an effective potential, has minima associated to enhanced gauge symmetries, where all internal moduli can be attracted and dynamically stabilized. Using the fact that the heterotic/type I S-duality remains valid at finite temperature and can be applied at each instant of a quasi-static evolution, we find in the dual type I cosmology that all internal NS-NS and RR moduli in the closed string sector and the Wilson lines in the open string sector can be stabilized. For the special case of D=6, the internal volume modulus remains a flat direction, while the dilaton is stabilized. An essential role is played by light D-string modes wrapping the internal manifold and whose contribution to the free energy cannot be omitted, even when the type I string is at weak coupling. As a result, the order of magnitude of the internal radii expectation values on the type I side is (lambda_I alpha')^{1/2}, where lambda_I is the ten-dimensional string coupling. The non-perturbative corrections to the type I free energy can alternatively be described as effects of "thermal E1-instantons", whose worldsheets wrap the compact Euclidean time cycle.

hep-th↗

arXiv2010

A topological limit of gravity admitting an SU(2) connection formulation

We study the Hamiltonian formulation of the generally covariant theory defined by the Lagrangian 4-form L=e_I e_J F^{IJ}(ω) where e^I is a tetrad field and F^{IJ} is the curvature of a Lorentz connection ω^{IJ}. This theory can be thought of as the limit of the Holst action for gravity for the Newton constant G goes to infinity and Immirzi parameter goes to zero, while keeping their product fixed. This theory has for a long time been conjectured to be topological. We prove this statement both in the covariant phase space formulation as well as in the standard Dirac formulation. In the time gauge, the unconstrained phase space of theory admits an SU(2) connection formulation which makes it isomorphic to the unconstrained phase space of gravity in terms of Ashtekar-Barbero variables. Among possible physical applications, we argue that the quantization of this topological theory might shed new light on the nature of the degrees of freedom that are responsible for black entropy in loop quantum gravity.

gr-qc↗

arXiv2008

Stability of BTZ black strings

We study the dynamical stability of the BTZ black string against fermonic and gravitational perturbations. The BTZ black string is not always stable against these perturbations. There exist threshold values for $m^2$ related to the compactification of the extra dimension for fermonic perturbation, scalar part of the gravitational perturbation and the tensor perturbation, respectively. Above the threshold values, perturbations are stable; while below these thresholds, perturbations can be unstable. We find that this non-trivial stability behavior qualitatively agrees with that predicted by a thermodynamical argument, showing that the BTZ black string phase is not the privileged stable phase.

hep-th↗

arXiv2007

The energy absorption problem of a brane-world black hole

We have studied the wave dynamics and the energy absorption problem for the scalar field as well as the brane-localized gravitational field in the background of a braneworld black hole. Comparing our results with the four-dimensional Schwarzschild black hole, we have observed the signature of the extra dimension in the energy absorption spectrum.

hep-th↗