A5100636286- Privacy-Preserving Technologies in Data
- Recommender Systems and Techniques
- Domain Adaptation and Few-Shot Learning
- AI-based Problem Solving and Planning
- Topic Modeling
- Data Mining Algorithms and Applications
- Text and Document Classification Technologies
- Cryptography and Data Security
- Data Management and Algorithms
- Human Mobility and Location-Based Analysis
- Web Data Mining and Analysis
- Stochastic Gradient Optimization Techniques
- Multimodal Machine Learning Applications
- Machine Learning and Data Classification
- Face and Expression Recognition
- Machine Learning and Algorithms
- Mobile Crowdsensing and Crowdsourcing
- Logic, Reasoning, and Knowledge
- Adversarial Robustness in Machine Learning
- Advanced Graph Neural Networks
- Context-Aware Activity Recognition Systems
- Advanced Image and Video Retrieval Techniques
- Indoor and Outdoor Localization Technologies
- Constraint Satisfaction and Optimization
- Machine Learning and ELM
Hong Kong University of Science and Technology
2016-2025
University of Hong Kong
2016-2025
Heilongjiang University of Chinese Medicine
2023-2025
East China University of Science and Technology
2022-2025
Beijing Institute of Petrochemical Technology
2023-2025
Xihua University
2025
Guizhou University
2025
Sun Yat-sen University
2013-2025
Anhui University
2025
Yangtze University
2014-2024
A major assumption in many machine learning and data mining algorithms is that the training future must be same feature space have distribution. However, real-world applications, this may not hold. For example, we sometimes a classification task one domain of interest, but only sufficient another where latter different or follow In such cases, knowledge transfer, if done successfully, would greatly improve performance by avoiding much expensive data-labeling efforts. recent years, transfer...
Domain adaptation allows knowledge from a source domain to be transferred different but related target domain. Intuitively, discovering good feature representation across domains is crucial. In this paper, we first propose find such through new learning method, transfer component analysis (TCA), for adaptation. TCA tries learn some components in reproducing kernel Hilbert space using maximum mean miscrepancy. the subspace spanned by these components, data properties are preserved and...
Today’s artificial intelligence still faces two major challenges. One is that, in most industries, data exists the form of isolated islands. The other strengthening privacy and security. We propose a possible solution to these challenges: secure federated learning. Beyond federated-learning framework first proposed by Google 2016, we introduce comprehensive framework, which includes horizontal learning, vertical transfer provide definitions, architectures, applications for survey existing...
A large family of algorithms - supervised or unsupervised; stemming from statistics geometry theory has been designed to provide different solutions the problem dimensionality reduction. Despite motivations these algorithms, we present in this paper a general formulation known as graph embedding unify them within common framework. In embedding, each algorithm can be considered direct its linear/kernel/tensor extension specific intrinsic that describes certain desired statistical geometric...
In recent years, mobile devices are equipped with increasingly advanced sensing and computing capabilities. Coupled advancements in Deep Learning (DL), this opens up countless possibilities for meaningful applications, e.g., medical purposes vehicular networks. Traditional cloud-based Machine (ML) approaches require the data to be centralized a cloud server or center. However, results critical issues related unacceptable latency communication inefficiency. To end, Mobile Edge Computing (MEC)...
Traditional machine learning makes a basic assumption: the training and test data should be under same distribution. However, in many cases, this identical-distribution assumption does not hold. The might violated when task from one new domain comes, while there are only labeled similar old domain. Labeling can costly it would also waste to throw away all data. In paper, we present novel transfer framework called TrAdaBoost, which extends boosting-based algorithms (Freund & Schapire, 1997)....
Multi-Task Learning (MTL) is a learning paradigm in machine and its aim to leverage useful information contained multiple related tasks help improve the generalization performance of all tasks. In this paper, we give survey for MTL from perspective algorithmic modeling, applications theoretical analyses. For definition then classify different algorithms into five categories, including feature approach, low-rank task clustering relation approach decomposition as well discussing...
Many applications of collaborative filtering (CF), such as news item recommendation and bookmark recommendation, are most naturally thought one-class (OCCF) problems. In these problems, the training data usually consist simply binary reflecting a user's action or inaction, page visitation in case webpage bookmarking scenario. Usually this kind extremely sparse (a small fraction positive examples), therefore ambiguity arises interpretation non-positive examples. Negative examples unlabeled...
In October 2005, we took an initiative to identify 10 challenging problems in data mining research, by consulting some of the most active researchers and machine learning for their opinions on what are considered important worthy topics future research mining. We hope insights will inspire new efforts, give young (including PhD students) a high-level guideline as where hot located Due limited amount time, were only able send out our survey requests organizers IEEE ICDM ACM KDD conferences,...
Abstract As a promising area in machine learning, multi-task learning (MTL) aims to improve the performance of multiple related tasks by leveraging useful information among them. In this paper, we give an overview MTL first giving definition MTL. Then several different settings are introduced, including supervised unsupervised semi-supervised active reinforcement online and multi-view learning. For each setting, representative models presented. order speed up process, parallel distributed...
Sentiment classification aims to automatically predict sentiment polarity (e.g., positive or negative) of users publishing data reviews, blogs). Although traditional algorithms can be used train classifiers from manually labeled text data, the labeling work time-consuming and expensive. Meanwhile, often use some different words when they express in domains. If we directly apply a classifier trained one domain other domains, performance will very low due differences between these In this...
Region-level demand forecasting is an essential task in ridehailing services. Accurate ride-hailing can guide vehicle dispatching, improve utilization, reduce the wait-time, and mitigate traffic congestion. This challenging due to complicated spatiotemporal dependencies among regions. Existing approaches mainly focus on modeling Euclidean correlations spatially adjacent regions while we observe that non-Euclidean pair-wise possibly distant are also critical for accurate forecasting. In this...
Multi-Task Learning (MTL) is a learning paradigm in machine and its aim to leverage useful information contained multiple related tasks help improve the generalization performance of all tasks. In this paper, we give survey for MTL from perspective algorithmic modeling, applications theoretical analyses. For definition then classify different algorithms into five categories, including feature approach, low-rank task clustering relation approach decomposition as well discussing...
Large language models (LLMs) are gaining increasing popularity in both academia and industry, owing to their unprecedented performance various applications. As LLMs continue play a vital role research daily use, evaluation becomes increasingly critical, not only at the task level, but also society level for better understanding of potential risks. Over past years, significant efforts have been made examine from perspectives. This paper presents comprehensive review these methods LLMs,...
Memory-based approaches for collaborative filtering identify the similarity between two users by comparing their ratings on a set of items. In past, memory-based approach has been shown to suffer from fundamental problems: data sparsity and difficulty in scalability. Alternatively, model-based proposed alleviate these problems, but this tends limit range users. paper, we present novel that combines advantages introducing smoothing-based method. our approach, clusters generated training...
With the increasing popularity of location-based services, such as tour guide and social network, we now have accumulated many location data on Web. In this paper, show that, by using based GPS users' comments at various locations, can discover interesting locations possible activities that be performed there for recommendations. Our research is highlighted in following location-related queries our daily life: 1) if want to do something sightseeing or food-hunting a large city Beijing, where...
Today's AI still faces two major challenges. One is that in most industries, data exists the form of isolated islands. The other strengthening privacy and security. We propose a possible solution to these challenges: secure federated learning. Beyond learning framework first proposed by Google 2016, we introduce comprehensive framework, which includes horizontal learning, vertical transfer provide definitions, architectures applications for survey existing works on this subject. In addition,...
Machine learning relies on the availability of a vast amount data for training. However, in reality, most are scattered across different organizations and cannot be easily integrated under many legal practical constraints. In this paper, we introduce new technique framework, known as federated transfer (FTL), to improve statistical models federation. The federation allows knowledge shared without compromising user privacy, enables complimentary transferred network. As result, target-domain...
Data preparation is a fundamental stage of data analysis. While lot low-quality information available in various sources and on the Web, many organizations or companies are interested how to transform into cleaned forms which can be used for high-profit purposes. This goal generates an urgent need analysis aimed at cleaning raw data. In this paper, we first show importance analysis, then introduce some research achievements area preparation. Finally, suggest future directions development.
As architecture, systems, and data management communities pay greater attention to innovative big systems architectures, the pressure of benchmarking evaluating these rises. Considering broad use benchmarks must include diversity workloads. Most state-of-the-art efforts target specific types applications or system software stacks, hence they are not qualified for serving purposes mentioned above. This paper presents our joint research on this issue with several industrial partners. Our...