Nikola Jevtic 3859 Miramar Street, Apt. #F La Jolla, CA 92037 Phone 858-450-6619 E-mail: nikola@cts.ucsd.edu Research Interests: Speech Recognition, Information Theory, Machine Learning , Artificial Intelligence Education - Ph.D. in Electrical and Computer Engineering (Speech Recognition) University of California San Diego GPA: 3.975./4.00 Expected: December 2004 - M.S. in Electrical and Computer Engineering University of California San Diego GPA: 3.975/4.00 December 2000 - B.Sc. in Electrical Engineering School of Electrical Engineering, University of Belgrade, Yugoslavia GPA: 9.58/10.00 September 1998 Work Experience Research Assistant, Electrical and Computer Engineering, University of California San Diego, 1998 – present: Participated in the development of the speech recognition software in Java, key person for language modeling tools and decoder. Also participated in the development of the HMM subsystem. Authored research papers and presented them at technical conferences. Teaching Assistant, Electrical and Computer Engineering, University of California San Diego, Fall 2000: Organized and taught discussion sections for the graduate Information Theory and undergraduate Discrete Mathematics course. Prepared and graded homework assignments, midterms and finals. Held office hours, counseled students individually. Papers: - A. Klautau, N. Jevtic, A. Orlitsky “Server-assisted speech recognition over the Internet” IEEE International Conference on Acoustics, Speech and Signal Processing, Istanbul., June 2000 - N. Jevtic, A. Klautau, A. Orlitsky “Estimated rank pruning and Java based speech recognition” Automatic Speech Recognition and Understanding Workshop, Madonna Di Campiglio, December 2001 - N. Jevtic, A. Orlitsky, N. P. Santhanam “Universal compression of unknown alphabets” International Symposium on Information Theory, Lausanne., July 2002 - A. Klautau, N. Jevtic, A. Orlitsky “Combined binary classifiers with applications to speech recognition” International Conference on Spoken Language Processing, Denver, September 2002 - A. Klautau, N. Jevtic, A. Orlitsky “On nearest-neighbor ECOC with application to all-pairs multiclass SVM” Journal of Machine Learning Research, April 2003, vol. 4, 1-15 - A. Klautau, N. Jevtic, A. Orlitsky “Discriminative Gaussian Mixture Models: A Comparison with Kernel Classifiers” International Conference on Machine Learning, Washington DC, August 2003 - N. Jevtic, A. Orlitsky “On the Relation Between Additive Smoothing and Universal Coding” Automatic Speech Recognition and Understanding Workshop, St.Thomas, US Virgin Islands December 2003 - N. Jevtic, A. Orlitsky, N. P. Santhanam “A lower bound on compression of unknown alphabets” Accepted for publication in Theoretical Computer Science. - N. Jevtic, A. Orlitsky “A Universal Compression Perspective of Smoothing” Submitted to Computer Speech and Language. Awards: - Best poster award, Electrical and Computer Engineering Department, UCSD Research Review 2000. - Several First and Second place awards during my undergraduate education, in Math competitions in Elektrijada, annual competitions among all Electrical Engineering Universities in Serbia and Montenegro. - A couple of First places on Serbian Math Olympics during Elementary and High School Scholarships: - Summer research program at University of Illinois at Chicago, Merit award given annually to the 40 best students in Yugoslav universities, July 1996 - Seminar on Telecommunications, National Technical University of Athens. Merit award given annually to the 20 best students at the School of Electrical Engineering, University of Belgrade, Yugoslavia, February 1997 - Scholarship from Yugoslav Foundation for Young Talents in Arts and Sciences, 1989-1998 Projects: - Language modeling tools: Implemented a Java toolkit for n-gram language modeling for LVCSR. Investigated smoothing methods from a universal-compression perspective. - Large vocabulary decoder: Implemented a Java decoder for LVCSR. Designed new approaches for efficiently combining information sources (language model, pronunciation dictionary, acoustic models). Small memory footprint for the inherently large structures involved was achieved and efficient search strategies for recombination of active hypotheses “in place” in adaptive beam search showed 14 times faster recognition than MSU open source C++ decoder. - Kernel classifiers for speech recognition: Investigated the use of kernel based classifiers such as Support, Relevance, and Informative Vector Machines for speech recognition. Optimal combination of binary classifiers was considered and all-pairs ECOC matrix performed best. However discriminatively trained mixture of Gaussian distributions outperformed kernel methods on several general machine learning tasks, including some speech related. - Server-Assisted Speech recognition over the Internet: Demonstrated a new approach for Speech recognition over the Internet. The system was written in java, and enables browsing the Internet using voice. Another demo was playing the tic-tac-toe over the Internet using voice. - Speech coding: Implemented CELP speech coder for a class project. - Speech Recognition based on Wavelets with HMT structure: Studied a new approach for using wavelets in speech recognition, by imposing Hidden Markov Tree structure to capture additional correlation from wavelet coefficients, within the standard temporal HMM setting. - Image Coding, Coding of similar images: Implemented a coder combining a JPEG coding and VQ to reduce the bit rate of coded images with assumed similarity, while in the same time producing identical output as a simple JPEG coder (class project). - Turbo Codes, Simulation: Simulated Turbo code demonstrating performance for various SNR (class project). Skills: - Programming: Expert knowledge of C, C++, Java. Good knowledge of shell scripts, Latex and Matlab. - Development Tools: Visual Studio, Emacs, CVS, Macromedia Dreamweaver, Borland JBuilder - Foreign Languages: Spanish and French References: - Dr. Alon Orlitsky, Professor UCSD Electrical and Computer Engineering and Computer Science and Engineering Departments Rm. 4602, EBU1 9500 Gilman Drive La Jolla, CA 92093-0409 Phone: (858) 822-0228, Fax: (858) 822-3426 E-mail: alon@ucsd.edu - Dr. Pamela Cosman, Associate Professor UCSD Electrical and Computer Engineering Department Rm. 4604, EBU1 9500 Gilman Drive La Jolla, CA 92093-0409 Phone: (858) 822-0157, Fax: (858) 822-3426 E-mail: pcosman@ucsd.edu - Dr. Sanjoy Dasgupta, Assistant Professor UCSD Computer Science and Engineering Department Rm. 4848, AP&M Building 9500 Gilman Drive La Jolla, CA 92093-0114 Phone: (858) 822-5270, Fax: (858) 534-7029 E-mail: dasgupta@cs.ucsd.edu - Dr. Charles Elkan, Associate Professor UCSD Computer Science and Engineering Department Rm. 4856, AP&M Building 9500 Gilman Drive La Jolla, CA 92093-0114 Phone: (858) 534-8897, Fax: (858) 534-7029 E-mail: elkan@cs.ucsd.edu