4 Key Works for Research Assessment Exercise (RAE) 2020
1. Na Li and M.W. Mak, "SNR-Invariant PLDA Modeling in Nonparametric Subspace for Robust Speaker Verification", IEEE/ACM Trans. on Audio Speech and Language Processing, vol. 23, no. 10, pp. 1648-1659, Oct. 2015.
This paper is the first attempt to explore the relationship between the noise-level variability in speech and the multi-modal distribution of i-vectors for robust speaker verification. The key contribution is the development of a general PLDA mixture model in which the mixture posteriors can be obtained not only from the data distribution but also from the meta information (e.g., SNR, duration, etc.) of speech signals. The paper is the result of the collaboration with Prof. J.T. Chien from National Chiao Tung University. It is also the main paper of a General Research Grant funded by the RGC of Hong Kong.
2. M.W. Mak, X.M. Pang and J.T. Chien, "Mixture of PLDA for Noise Robust I-Vector Speaker Verification", IEEE/ACM Trans. on Audio Speech and Language Processing, vol. 24, No. 1, pp. 130-142, Jan. 2016.
This is the first speaker recognition paper that introduces an SNR subspace to the PLDA model for robust speaker recognition. The work is significant in that it leads to a general model that opens up opportunity for researchers to model all sort of variability (e.g., duration and reverberation) in utterances, not only limited to noise level variability. The work is the result of a General Research Fund (GRF). It has laid down the foundation for our successful GRF bid in the following year and has trained up a PostDoc who is now a Research Scientist in Tencent AI Lab.
3. S.B. Wan, M.W. Mak, and S.Y. Kung, "HybridGO-Loc: Mining Hybrid Features on Gene Ontology for Predicting Subcellular Localization of Multi-Location Proteins", PLoS ONE, vol. 9, no. 3, March 2014. http://dx.doi.org/10.1371/journal.pone.0089545
This paper is the first attempt to address a missing element in today’s multi-labeled subcellular localization predictors: The interpretation of the prediction decisions. The innovative part is the application of sparse regressions to exploit gene ontology for both predicting and interpreting the subcellular localization of single- and multi-location proteins. The work has also led to a web-server that not only allows other researchers to use our predictors in their work but also attracts attention from the bioinformatics community. It has also trained up a research student who is now a PostDoc in Princeton University.
4. S.B. Wan, M.W. Mak and S.Y. Kung, "Sparse Regressions for Predicting and Interpreting Subcellular Localization of Multi-label Proteins", BMC Bioinformatics, vol. 17:97, DOI 10.1186/s12859-016-0940-x, Feb. 2016
This paper represents the first bioinformatics study that leverages the inter GO-term relationship for protein subcellular localization. The key contribution is the discovery of the interesting relationship between the frequency of occurrences of GO terms and their semantic similarity so that exploiting their hybridization can lead to significant performance gain. The work is part of the 15-year collaboration with Prof. S.Y. Kung from Princeton University. It has also led to a web-server that not only allows other researchers to use our predictors in their work but also attracts attention from the bioinformatics community.