Dr. David Miller

Research Results

Machine Learning

novel semisupervised learning algorithms, including semisupervised class discovery, semisupervised learning with instance-level constraints, and semisupervised EM;

deterministic annealing techniques for clustering, tree-structured clustering, supervised classification, and piecewise regression;

maximum entropy statistical inference for classification, general inference, and collaborative filtering;

ensemble classification techniques, including transductive distributed classification, maximum entropy-improved iterative scaling ensemble classification, and critic-driven ensembles;

clustering/mixture modeling in high dimensions: parsimonious mixtures with integrated model order and feature selection (and application to document clustering);

improved mean-field annealing techniques for image segmentation;

data mining for network digesting and intrusion detection

Source Coding and Source-Channel Coding

novel joint source-channel decoding techniques, including MMSE decoders, MAP decoders for varaible-length encoded data, source-channel decoding for predictive coding systems, and joint source-channel decoding of digital images.

optimal encoding for variable rate source coding with dependent data streams;

entropy-constrained tree-structured vector quantization;

deterministic annealing for source-channel, tree-structured, and trellis quantization

Citations

M. Park and D. Miller， ``Joint source-channel decoding for variable-length encoded data by exact and approximate {MAP} sequence estimation'', IEEE Trans. on Communications, vol. 48, no. 1, 2000, pp. 1-6; cited 73 times measured by Googlescholar.

D. J. Miller and H. S. Uyar, ``A mixture of experts classifier with learning based on both labelled and unlabelled data'', Neural Information Processing Systems, 1997, pp. 571-577; early paper on semisupervised learning via Expectation-Maximization, cited 102 times on Googlescholar and cited by Nigam, Mccallum, Thrun, Mitchell's ``Text classification from labeled and unlabeled documents using EM'', which has been cited 736 times on Googlescholar.

A subset of 15 papers has been cited more than 450 times, measured by Googlescholar.

Research Sponsorship

Principal Investigator on two National Science Foundation grants in machine learning area

Co-investigator on 3-year NSF Collaborative research grant on protecting network infrastructure, 2003 - 2006

Co-investigator on NIH bioinformatics proposal in single nucleotide polymorphism (SNP) biomarker discovery

Sponsored research support from AFRL Phase II SBIR in video tracking, 2004-2006

Sponsored research support from ONR in maximum entropy data and decision fusion, 2004-2006

Consultant on NASA, NIH, and ONR projects

Awards

NSF CAREER Award in 1996.

Best Presentation Award, Monday poster session, at the International Joint Conference on Neural Networks, 1999, held in Washington, D.C.

Best student paper finalist at ICASSP'2005 for paper on semisupervised learning.


	Research Results Machine Learning novel semisupervised learning algorithms, including semisupervised class discovery, semisupervised learning with instance-level constraints, and semisupervised EM; deterministic annealing techniques for clustering, tree-structured clustering, supervised classification, and piecewise regression; maximum entropy statistical inference for classification, general inference, and collaborative filtering; ensemble classification techniques, including transductive distributed classification, maximum entropy-improved iterative scaling ensemble classification, and critic-driven ensembles; clustering/mixture modeling in high dimensions: parsimonious mixtures with integrated model order and feature selection (and application to document clustering); improved mean-field annealing techniques for image segmentation; data mining for network digesting and intrusion detection Source Coding and Source-Channel Coding novel joint source-channel decoding techniques, including MMSE decoders, MAP decoders for varaible-length encoded data, source-channel decoding for predictive coding systems, and joint source-channel decoding of digital images. optimal encoding for variable rate source coding with dependent data streams; entropy-constrained tree-structured vector quantization; deterministic annealing for source-channel, tree-structured, and trellis quantization Citations M. Park and D. Miller， ``Joint source-channel decoding for variable-length encoded data by exact and approximate {MAP} sequence estimation'', IEEE Trans. on Communications, vol. 48, no. 1, 2000, pp. 1-6; cited 73 times measured by Googlescholar. D. J. Miller and H. S. Uyar, ``A mixture of experts classifier with learning based on both labelled and unlabelled data'', Neural Information Processing Systems, 1997, pp. 571-577; early paper on semisupervised learning via Expectation-Maximization, cited 102 times on Googlescholar and cited by Nigam, Mccallum, Thrun, Mitchell's ``Text classification from labeled and unlabeled documents using EM'', which has been cited 736 times on Googlescholar. A subset of 15 papers has been cited more than 450 times, measured by Googlescholar. Research Sponsorship Principal Investigator on two National Science Foundation grants in machine learning area Co-investigator on 3-year NSF Collaborative research grant on protecting network infrastructure, 2003 - 2006 Co-investigator on NIH bioinformatics proposal in single nucleotide polymorphism (SNP) biomarker discovery Sponsored research support from AFRL Phase II SBIR in video tracking, 2004-2006 Sponsored research support from ONR in maximum entropy data and decision fusion, 2004-2006 Consultant on NASA, NIH, and ONR projects Awards NSF CAREER Award in 1996. Best Presentation Award, Monday poster session, at the International Joint Conference on Neural Networks, 1999, held in Washington, D.C. Best student paper finalist at ICASSP'2005 for paper on semisupervised learning.