Transfer learning aims to transfer knowledge or information from a source domain to a relevant target domain. It involves two key challenges: distribution shifts and trustworthiness concerns. Having these challenges in mind, my research focuses on understanding transfer learning from the perspective of knowledge transferability (e.g., IID and non-IID learning tasks) and trustworthiness (e.g., adversarial robustness, data privacy, and performance fairness).

In the era of foundation models, Out-of-Distribution (OOD) problems, i.e., the data discrepancy between the training environments and testing environments, hinder AI generalization. Further, relational data like graphs disobeying the Independent and Identically Distributed (IID) condition makes the problem more challenging, especially much harder when it is associated with time. Motivated by this, to realize the robust invariant learning over temporal graphs, we want to investigate what components in temporal graphs are most invariant and representative with respect to labels. With the Information Bottleneck (IB) method, we propose an error-bounded Invariant Link Selector that can distinguish invariant components and variant components during the training process to make the deep learning model generalizable for different testing scenarios. Besides deriving a series of rigorous generalizable optimization functions, we also equip the training with task-specific loss functions, e.g., temporal link prediction, to make pre-trained models solve real-world application tasks like citation recommendation and merchandise recommendation, as demonstrated in our experiments with state-of-the-art (SOTA) methods. Our code is available at \url{https://github.com/kthrn22/OOD-Linker}

The proliferation of open-sourced Large Language Models (LLMs) and diverse downstream tasks necessitates efficient model selection, given the impracticality of fine-tuning all candidates due to computational constraints. Despite the recent advances in LLM selection, a fundamental research question largely remains nascent: how can we model the dynamic behaviors of LLMs during fine-tuning, thereby enhancing our understanding of their generalization performance across diverse downstream tasks? In this work, we propose a novel theoretical framework that provides a proper lens to assess the generalization capabilities of LLMs, thereby enabling accurate and efficient LLM selection for downstream applications. In particular, we first derive a PAC-Bayesian Generalization Bound that unveils fine-tuning dynamics of LLMs and then introduce LensLLM, a Neural Tangent Kernel (NTK)-based Rectified Scaling Model that enables accurate performance predictions across diverse tasks while maintaining computational efficiency. Extensive empirical results on 3 large-scale benchmarks demonstrate that our model achieves up to 91.1% accuracy and reduces up to 88.5% computational cost in LLM selection, outperforming 5 state-of-the-art methods. We open-source our proposed LensLLM model and corresponding results at LensLLM.io.

Transfer learning aims to transfer knowledge or information from a source domain to a relevant target domain. In this paper, we understand transfer learning from the perspectives of knowledge transferability and trustworthiness. This involves two research questions: How is knowledge transferability quantitatively measured and enhanced across domains? Can we trust the transferred knowledge in the transfer learning process? To answer these questions, this paper provides a comprehensive review of trustworthy transfer learning from various aspects, including problem definitions, theoretical analysis, empirical algorithms, and real-world applications. Specifically, we summarize recent theories and algorithms for understanding knowledge transferability under (within-domain) IID and non-IID assumptions. In addition to knowledge transferability, we review the impact of trustworthiness on transfer learning, e.g., whether the transferred knowledge is adversarially robust or algorithmically fair, how to transfer the knowledge under privacy-preserving constraints, etc. Beyond discussing the current advancements, we highlight the open questions and future directions for understanding transfer learning in a reliable and trustworthy manner.

In federated learning, most existing robust aggregation rules (AGRs) combat Byzantine attacks in the IID setting, where client data is assumed to be independent and identically distributed. In this paper, we address label skewness, a more realistic and challenging non-IID setting, where each client only has access to a few classes of data. In this setting, state-of-the-art AGRs suffer from selection bias, leading to significant performance drop for particular classes; they are also more vulnerable to Byzantine attacks due to the increased variation among gradients of honest clients. To address these limitations, we propose an efficient two-stage method named BOBA. Theoretically, we prove the convergence of BOBA with an error of the optimal order. Our empirical evaluations demonstrate BOBA’s superior unbiasedness and robustness across diverse models and datasets when compared to various baselines.

Heterogeneous data widely exists in various high-impact applications. Domain adaptation and out-of-distribution generalization paradigms have been formulated to handle the data heterogeneity across domains. However, most existing domain adaptation and out-of-distribution generalization algorithms do not explicitly explain how the label information can be adaptively propagated from the source domains to the target domain. Furthermore, little effort has been devoted to theoretically understanding the convergence of existing algorithms based on neural networks.To address these problems, in this paper, we propose a generic distributional network of networks (TENON) framework, where each node of the main network represents an individual domain associated with a domain-specific network. In this case, the edges within the main network indicate the domain similarity, and the edges within each network indicate the sample similarity. The crucial idea of TENON is to characterize the within-domain label smoothness and cross-domain parameter smoothness in a unified framework. The convergence and optimality of TENON are theoretically analyzed. Furthermore, we show that based on the TENON framework, domain adaptation and out-of-distribution generalization can be naturally formulated as transductive and inductive distribution learning problems, respectively. This motivates us to develop two instantiated algorithms (TENON-DA and TENON-OOD) of the proposed TENON framework for domain adaptation and out-of-distribution generalization. The effectiveness and efficiency of TENON-DA and TENON-OOD are verified both theoretically and empirically.

Transfer learning refers to the transfer of knowledge or information from a relevant source domain to a target domain. However, most existing transfer learning theories and algorithms focus on IID tasks, where the source/target samples are assumed to be independent and identically distributed. Very little effort is devoted to theoretically studying the knowledge transferability on non-IID tasks, e.g., cross-network mining. To bridge the gap, in this paper, we propose rigorous generalization bounds and algorithms for cross-network transfer learning from a source graph to a target graph. The crucial idea is to characterize the cross-network knowledge transferability from the perspective of the Weisfeiler-Lehman graph isomorphism test. To this end, we propose a novel Graph Subtree Discrepancy to measure the graph distribution shift between source and target graphs. Then the generalization error bounds on cross-network transfer learning, including both cross-network node classification and link prediction tasks, can be derived in terms of the source knowledge and the Graph Subtree Discrepancy across domains. This thereby motivates us to propose a generic graph adaptive network (GRADE) to minimize the distribution shift between source and target graphs for cross-network transfer learning. Experimental results verify the effectiveness and efficiency of our GRADE framework on both cross-network node classification and cross-domain recommendation tasks.

In federated learning (FL), multiple clients collaborate to train machine learning models together while keeping their data decentralized. Through utilizing more training data, FL suffers from the potential negative transfer problem: the global FL model may even perform worse than the models trained with local data only. In this paper, we propose FedCollab, a novel FL framework that alleviates negative transfer by clustering clients into non-overlapping coalitions based on their distribution distances and data quantities. As a result, each client only collaborates with the clients having similar data distributions, and tends to collaborate with more clients when it has less data. We evaluate our framework with a variety of datasets, models, and types of non-IIDness. Our results demonstrate that FedCollab effectively mitigates negative transfer across a wide range of FL algorithms and consistently outperforms other clustered FL algorithms.

With decentralized data collected from diverse clients, a personalized federated learning paradigm has been proposed for training machine learning models without exchanging raw data from local clients. We dive into personalized federated learning from the perspective of privacy-preserving transfer learning, and identify the limitations of previous personalized federated learning algorithms. First, previous works suffer from negative knowledge transferability for some clients, when focusing more on the overall performance of all clients. Second, high communication costs are required to explicitly learn statistical task relatedness among clients. Third, it is computationally expensive to generalize the learned knowledge from experienced clients to new clients.

Deep transfer learning investigates the transfer of knowledge or information from a source domain to a relevant target domain via deep neural networks. In this tutorial, we dive into understanding deep transfer learning from the perspective of knowledge transferability and trustworthiness (e.g., privacy, robustness, fairness, transparency, etc.). To this end, we provide a comprehensive review of state-of-the-art theoretical analysis and algorithms for deep transfer learning. To be specific, we start by introducing the concepts of transferability and trustworthiness in the context of deep transfer learning. Then we summarize recent theories and algorithms for understanding knowledge transferability from two aspects: (1) IID transferability: the samples within each domain are independent and identically distributed (e.g., individual images), and (2) non-IID transferability: The samples within each domain are interdependent (e.g., connected nodes within a graph). In addition to knowledge transferability, we also review the impact of trustworthiness on deep transfer learning, e.g., whether the transferred knowledge is adversarially robust or algorithmically fair, how to transfer the knowledge under privacy-preserving constraints, etc. Finally, we highlight the open questions and future directions for understanding deep transfer learning in real-world applications. We believe this tutorial can benefit researchers and practitioners by rethinking the trade-off between knowledge transferability and trustworthiness in developing trustworthy transfer learning systems.

Transferable graph learning involves knowledge transferability from a source graph to a relevant target graph. The major challenge of transferable graph learning is the distribution shift between source and target graphs induced by individual node attributes and complex graph structures. To solve this problem, in this paper, we propose a generic graph-structured Gaussian process framework (GraphGP) for adaptively transferring knowledge across graphs with either homophily or heterophily assumptions. Specifically, GraphGP is derived from a novel graph structure-aware neural network in the limit on the layer width. The generalization analysis of GraphGP explicitly investigates the connection between knowledge transferability and graph domain similarity. Extensive experiments on several transferable graph learning benchmarks demonstrate the efficacy of GraphGP over state-of-the-art Gaussian process baselines.

Multi-source domain adaptation studies the knowledge transferability from multiple labeled source domains to an unlabeled target domain under a distribution shift. However, little effort has been devoted to studying the adversarial vulnerability of multi-source domain adaptation approaches. Specifically, most existing techniques focus on learning the domain-invariant representation to mitigate the distribution shift across domains. In this paper, we theoretically show that the domain-invariant representation cannot guarantee the success of multi-source domain adaptation, when no labeled samples are available in the target domain. This result motivates us to propose a unified framework (AdaptAttack) for data poisoning adversarial attacks on multi-source domain adaptation. The key idea is to maliciously manipulate the label-informed data distributions of source domains by injecting perceptibly unnoticeable noise into the source data. In addition, it requires that the generated adversarial attacks are invisible to multi-source domain adaptation algorithms, i.e., the source classification errors and marginal discrepancies across domains are not negatively affected. Extensive experiments on public domain adaptation benchmarks confirm the effectiveness and computational efficiency of our proposedAdaptAttackframework in both white-box and black-box attack scenarios.

In this paper, we study the dynamic transfer learning problem involving adaptive knowledge transfer from a static source domain to a time evolving target domain. One major challenge is the time evolving relatedness of the source domain and the current target domain as the target domain evolves over time. To address this challenge, we derive a generic error bound on the current target domain with flexible domain discrepancy measures. Moreover, we propose a label-informed $\mathcal{C}$-divergence to measure the shift of joint data distributions (over input features and output labels) across domains. The resulting tighter error bound with $\mathcal{C}$-divergence motivates us to develop a novel dynamic transfer learning algorithm TransLATE. Empirical results on various data sets confirm the effectiveness of our proposed algorithm in modeling the time evolving target domain.

Federated learning learns a neural network model by aggregating the knowledge from a group of distributed clients under the privacy-preserving constraint. In this work, we show that this paradigm might inherit the adversarial vulnerability of the centralized neural network, i.e., it has deteriorated performance on adversarial examples when the model is deployed. This is even more alarming when federated learning paradigm is designed to approximate the updating behavior of a centralized neural network. To solve this problem, we propose an adversarially robust federated learning framework, named Fed_BVA, with improved server and client update mechanisms. This is motivated by our observation that the generalization error in federated learning can be naturally decomposed into the bias and variance triggered by multiple clients' predictions. Thus, we propose to generate the adversarial examples via maximizing the bias and variance during server update, and learn the adversarially robust model updates with those examples during client update. As a result, an adversarially robust neural network can be aggregated from these improved local clients' model updates. The experiments are conducted on multiple benchmark data sets using several prevalent neural network models, and the empirical results show that our framework is robust against white-box and black-box adversarial corruptions under both IID and non-IID settings.

Transfer learning refers to the transfer of knowledge or information from a relevant source task to a target task. However, most existing works assume both tasks are sampled from a stationary task distribution, thereby leading to the sub-optimal performance for dynamic tasks drawn from a non-stationary task distribution in real scenarios. To bridge this gap, in this paper, we study a more realistic and challenging transfer learning setting with dynamic tasks, i.e., source and target tasks are continuously evolving over time. We theoretically show that the expected error on the dynamic target task can be tightly bounded in terms of source knowledge and consecutive distribution discrepancy across tasks. This result motivates us to propose a generic meta-learning framework L2E for modeling the knowledge transferability on dynamic tasks. It is centered around a task-guided meta-learning problem with a group of meta-pairs of tasks, based on which we are able to learn the prior model initialization for fast adaptation on the newest target task. L2E enjoys the following properties: (1) effective knowledge transferability across dynamic tasks; (2) fast adaptation to the new target task; (3) mitigation of catastrophic forgetting on historical target tasks; and (4) flexibility in incorporating any existing static transfer learning algorithms. Extensive experiments on various image data sets demonstrate the effectiveness of the proposed L2E framework.

Open-set domain adaptation aims to improve the generalization performance of a learning algorithm on a target task of interest by leveraging the label information from a relevant source task with only a subset of classes. However, most existing works are designed for the static setting, and can be hardly extended to the dynamic setting commonly seen in many real-world applications. In this paper, we focus on the more realistic open-set domain adaptation setting with a static source task and a time evolving target task where novel unknown target classes appear over time. Specifically, we show that the classification error of the new target task can be tightly bounded in terms of positive-unlabeled classification errors for historical tasks and open-set domain discrepancy across tasks. By empirically minimizing the upper bound of the target error, we propose a novel positive-unlabeled learning based algorithm named OuterAdapter for dynamic open-set domain adaptation with time evolving unknown classes. Extensive experiments on various data sets demonstrate the effectiveness and efficiency of our proposed OuterAdapter algorithm over state-of-the-art domain adaptation baselines.

In this paper, we study the problem of domain adaptation regression, which learns a regressor for a target domain by leveraging the knowledge from a relevant source domain. We start by proposing a distribution-informed neural network, which aims to build distribution-aware relationship of inputs and outputs from different domains. This allows us to develop a simple domain adaptation regression framework, which subsumes popular domain adaptation approaches based on domain invariant representation learning, reweighting, and adaptive Gaussian process. The resulting findings not only explain the connections of existing domain adaptation approaches, but also motivate the efficient training of domain adaptation approaches with overparameterized neural networks. We also analyze the convergence and generalization error bound of our framework based on the distribution-informed neural network. Specifically, our generalization bound focuses explicitly on the maximum mean discrepancy in the RKHS induced by the neural tangent kernel of distribution-informed neural network. This is in sharp contrast to the existing work which relies on domain discrepancy in the latent feature space heuristically formed by one or several hidden neural layers. The efficacy of our framework is also empirically verified on a variety of domain adaptation regression benchmarks.

Unsupervised domain adaptation has been successfully applied across multiple high-impact applications, since it improves the generalization performance of a learning algorithm when the source and target domains are related. However, the adversarial vulnerability of domain adaptation models has largely been neglected. Most existing unsupervised domain adaptation algorithms might be easily fooled by an adversary, resulting in deteriorated prediction performance on the target domain, when transferring the knowledge from a maliciously manipulated source domain.

Recommender systems are powerful tools for information filtering with the ever-growing amount of online data. Despite its success and wide adoption in various web applications and personalized products, many existing recommender systems still suffer from multiple drawbacks such as large amount of unobserved feedback, poor model convergence, etc. These drawbacks of existing work are mainly due to the following two reasons: first, the widely used negative sampling strategy, which treats the unlabeled entries as negative samples, is invalid in real-world settings; second, all training samples are retrieved from the discrete observations, and the underlying true distribution of the users and items is not learned.

Networks are natural representations in modeling adversarial activities, such as smuggling, human trafficking, and illegal arms dealing. However, such activities are often covert and embedded across multiple domains and sources. They are generally not detectable and recognizable from the perspective of an isolated network, and only become apparent when multiple networks are analyzed in a unified m anner. T o t his e nd, we propose Complex Analytics of Network of Networks (CANON), a mathematical and computational framework for modeling adversarial activities from large-scale, multi-sourced data inputs. Central to our framework is a network-of-networks model, where nodes and edges can be defined across different domains and at multiple resolutions. Based on this model, we address the key challenges in modeling adversarial activities via four technical components, including optimization-based network alignment, network embedding and conditioning, approximate subgraph matching, and investigative subgraph discovery.In this paper, we describe the design and implementation of the individual components as well as integrating these components into a unified system using a modular microservice architecture. Extensive experiments have been conducted in both synthetics and real-world datasets to demonstrate the effectiveness of our proposed system under the DARPA Modeling Adversarial Activity (MAA) program.

Networks are ubiquitous in many real-world applications due to their capability of representing the rich information in the data. One fundamental problem of network analysis is to learn a low- dimensional vector representation for nodes within the attributed networks. However, there is little work theoretically considering the information heterogeneity from the attributed networks, and most of the existing attributed network embedding techniques are able to capture at most k-th order node proximity, thus leading to the information loss of the long-range spatial dependencies between individual nodes across the entire network. To address the above problems, in this paper, we propose a novel MAnifold-RegularIzed Network Embedding (MARINE) algorithm inspired by minimizing the information discrepancy in a Reproducing Kernel Hilbert Space via Maximum Mean Discrepancy. In particular, we show that MARINE recursively aggregates the graph structure information as well as individual node attributes from the entire network, and thereby preserves the long-range spatial dependencies between nodes across the network. The experimental results on real networks demonstrate the effectiveness and efficiency of the proposed MARINE algorithm over state-of-the-art embedding methods.

Graph data widely exist in many high-impact applications. Inspired by the success of deep learning in grid-structured data, graph neural network models have been proposed to learn powerful node-level or graph-level representation. However, most of the existing graph neural networks suffer from the following limitations: (1) there is limited analysis regarding the graph convolution properties, such as seed-oriented, degree-aware and order-free; (2) the node's degreespecific graph structure is not explicitly expressed in graph convolution for distinguishing structure-aware node neighborhoods; (3) the theoretical explanation regarding the graph-level pooling schemes is unclear.

Imbalanced data widely exist in many high-impact applications. An example is in air traffic control, where among all three types of accident causes, historical accident reports with `personnel issues' are much more than the other two types (`aircraft issues' and `environmental issues') combined. Thus, the resulting data set of accident reports is highly imbalanced. On the other hand, this data set can be naturally modeled as a network, with each node representing an accident report, and each edge indicating the similarity of a pair of accident reports. Up until now, most existing work on imbalanced data analysis focused on the classification setting, and very little is devoted to learning the node representations for imbalanced networks. To bridge this gap, in this paper, we first propose Vertex-Diminished Random Walk (VDRW) for imbalanced network analysis. It is significantly different from the existing Vertex Reinforced Random Walk by discouraging the random particle to return to the nodes that have already been visited. This design is particularly suitable for imbalanced networks as the random particle is more likely to visit the nodes from the same class, which is a desired property for learning node representations. Furthermore, based on VDRW, we propose a semi-supervised network representation learning framework named ImVerde for imbalanced networks, where context sampling uses VDRW and the limited label information to create node-context pairs, and balanced-batch sampling adopts a simple under-sampling method to balance these pairs from different classes. Experimental results demonstrate that ImVerde based on VDRW outperforms state-of-the-art algorithms for learning network representations from imbalanced data.