Jie Lu - Selected Publications#
NOTATIONS
Prof Lu is the corresponding author on all 10 papers.
Where she is not first author, it is one of her PhD students or postdoctoral associates. Where she is immediately after the student/postdoc, in the cases she has contributed extensively as the first author’s supervisor. The last author usually led the project and/or is the first author’s PhD co-supervisor.
IF= Impact factor, S=Scopus, GS=Google Scholar, CS = computer science.
[1] Fang, Z, Lu, J, Liu, F, Zhang, G. (2022) ‘Semi-supervised heterogeneous domain adaptation: theory and algorithms’, IEEE Transactions on Pattern Analysis and Machine Intelligence, DOI: 10.1109/ TPAMI.2022.3146234.
Journal IF: 16.39, ranked 1/139 in Computer Science & AI (CS&AI)
This breakthrough gives a theoretical foundation for algorithms to train a classifier for a target domain in which only unlabelled and a small number of labelled data are available. This semi-supervised heterogeneous domain adaptation theory has been a long-standing problem in machine learning. The study is the first to explain the nature of the problem, so will guide new and better solutions.
[2] Fang, Z., Lu, J., Liu, F., Xuan, J., Zhang G. (2021) ‘Open set domain adaptation: theoretical bound and algorithm’, IEEE Transactions on Neural Networks and Learning Systems, 32 (10), pp. 4309–22.
Journal: IF: 10.5; ranked 10/139 in CS&AI
Citations: 46 (GS)
This breakthrough has solved a long-standing theoretical issue in transfer learning. We devised a ‘learning bound’ – a novel theory for unsupervised domain adaptation. This is now an important theoretical base for researching unsupervised transfer learning.
[3] Liu, A., Lu, J. Zhang G. (2021), ‘Concept drift detection via equal intensity k-means space partitioning’, IEEE Transactions on Cybernetics, 51 (6), pp. 3198–3211.
Journal: IF: 11.5; ranked 6/139 in CS&AI
Citations: 23 (GS)
We propose an innovative solution to a major challenge: real-time detection of unpredictable distribution changes in data streams, called ‘concept drift’. The solution significantly improves drift detection accuracy and sensitivity in complex unsupervised settings, and has been successfully applied in real-world applications e.g., Sydney Trains carriage-load prediction in real time.
[4] Liu, F., Zhang, G., Lu, J. (2020) ‘Heterogeneous domain adaptation: An unsupervised approach’, IEEE Transactions on Neural Networks and Learning Systems, 31 (12), pp. 5588–5602
Journal: 10.5; ranked 10/139 in CS&AI
Citations: 49 (GS), top 3% Scopus
We report a theoretical breakthrough in cross-domain transfer learning: unsupervised knowledge transfer theory that guarantees the effectiveness of transferring knowledge to a heterogeneous domain and serves as a robust measure of the distance between two heterogeneous domains.
[5] Liu, F, Zhang, G., Lu, J (2020) ‘Multisource heterogeneous unsupervised domain adaptation via fuzzy relation neural networks’, IEEE Transactions on Fuzzy Systems 29 (11), 3308-3322
Journal: IF 12.029 ranked 4/139 in CS&AI.
Citations: 21 (GS)
A technological breakthrough in fuzzy transfer learning. We provided an innovative solution for multi-source domain unsupervised transfer learning based on fuzzy relation neural networks. This achieves significantly improved prediction accuracy and a totally new way for cross multi-source heterogeneous transfer learning.
[6] Zhang, Q., Lu, J., Wu, D., Zhang, G. (2019), ‘A cross-domain recommender system with kernel-induced knowledge transfer for overlapping entities’, IEEE Transactions on Neural Networks and Learning Systems, 30 (7), pp. 1998–2012.
Journal: 10.5; ranked 10/139 in CS&AI,
Citations: 70 (GS), top 3% Scopus
The first solution of knowledge transfer between different items to personalised recommender systems, which overcomes the data insufficient/sparsity problem in recommendation generation. The idea has also been applied by industry partners e.g., in the Woolworths supermarket chain, in their recommender system in Sydney via developments by Lu’s graduates.
[7] Lu, J. Zuo, H., Zhang G. (2019) ‘Fuzzy multiple-source transfer learning’, IEEE Transactions on Fuzzy Systems, 28 (12), pp. 3418–31.
Journal: IF 12.0; ranked 4/139 in CS&AI.
Citations: 31 (GS), top 5% Scopus
The first-ever theory and demonstration of transferring knowledge from multiple source domains (homogeneous and heterogeneous) to assist prediction in a target domain that lacks data to train a prediction model. The innovation involved developing and integrating fuzzy clustering into the knowledge transfer process. This is becoming an important theoretical component of multi-source transfer learning research.
[8] Lu, J., Xuan, J., Zhang, G., Luo, X. (2018), ‘Structural property-aware multilayer network embedding for latent factor analysis, Pattern Recognition, 76, pp. 228–41.
Journal: IF: 7.7; ranked 17/139 in CS&AI
Citations: 48 (GS), top 10% Scopus
An important theoretical contribution to incorporating various complex network properties as constraints in latent factor analysis. The results can support data analysis in complex situations, solving multilayer network embedding issues and knowledge transfers between networks in a multi-layer network model.
[9] Mao, M, Lu, J, Zhang, G. Zhang, J. (2017), ‘Multirelational social recommendations via multigraph ranking’, IEEE Transactions on Cybernetics, 47, pp. 4049–61.
Journal: IF: 11.4; ranked 6/139 in CS&AI
Citations: 101 (GS), top 3% Scopus
A highly innovative method of mining multi-relational social networks to solve a key issue – the data sparsity problem – for personalised recommender systems. This enables the system to effectively identify relevant items for particular users in large-scale online applications such as e-commence, e-government, and e-learning. Cited by Echostar Technologies in two patents (US 10268689 B2; US 10390084 B2).
[10] Pratama, M., Lu, J., Zhang G. (2016), ‘Evolving type-2 fuzzy classifier’, IEEE Transactions on Fuzzy Systems, 24 (3), pp. 574–89.
Journal: IF 12.029 ranked 4/139 in CS&AI.
Citations: 119 (GS), top 1% Scopus
A novel algorithm and accompanying theoretical guarantee for effectively handling non-stationary data streams with highly uncertain characteristics in features. The learning process can be started from scratch; fuzzy rules can be automatically grown, pruned, recalled, and merged in real time. This overcomes problems that earlier systems had with outliers and rules becoming outdated, while retaining lower complexity.
The novelty and significance of the advance resulted in a prestigious award: The 2019 IEEE Transactions on Fuzzy Systems Outstanding Paper” (one awarded per year).
