Me Papers CV Research Contact Useful Link

Hsiao-Yu Fish Tung

Machine Learning Department

Carnegie Mellon University

mail: htung at cs.cmu.edu

About me

some_text

I am a first-year PhD student in the Machine Learning Department at CMU. I work with Professor Alex Smola on deep learning and scalable machine learning algorithms. My research interests include Deep Learning, Spectral Methods, Statistical Modeling, Data Mining and Application.

I received my M.S. in CMU MLD and my B.S. in Electrical Engineering from National Taiwan University in 2013. During my master, I worked on spectral method for Bayesian models and succefully designed efficient and provable algorithms. In my undergraduate, I was one of the member in Professor Chih-Jen Lin 's team "Algorithm @ National Taiwan University" and won the KDD CUP 2013 Championship on both Track1 and Track2.

Conference Papers

Fast and Guaranteed Tensor Decomposition via Sketching [Paper]

Yining Wang, Hsiao-Yu Tung, Alexander J. Smola, Animashree Anandkumar
Neural Information Processing Systems, NIPS 2015, spotlight.

Spectral Methods for Indian Buffet Process Inference. [Paper]

Hsiao-Yu F. Tung, Alexander J. Smola
Neural Information Processing Systems, NIPS 2014.

Novel Traffic Light Timing Adjustment Strategy Based On Genetic Algorithm. [Paper]

Hsiao-Yu Tung, Wei-Chiu Ma, and Tian-Li Yu
IEEE Congress on Evolutionary Computation, IEEE CEC 2013, Oral.

Effective String Processing and Matching for Author Disambiguation. [Paper]

W.-S. Chin, Y.-C. Juan, Y.-Zhuang, Felix Wu, H.-Y. Tung, T. Yu, J.-P. Wang, C.-X. Chang, C.-P. Yang, W.-C. Chang, K.-H. Huang, T.-M. Kuo, S.-W. Lin, Y.-S. Lin, Y.-C. Lu, Y.-C. Su, C.-K. Wei, T.- C. Yin, C.-L. Li, T.-W. Lin, C.-H. Tsai, S.-d. Lin, H.-T. Lin, C.-J. Lin
KDD Cup 2013 Workshop, KDD 2013

Combination of Feature Engineering and Ranking Models for Paper-Author Identification in KDD Cup 2013 [Paper]

C.-L. Li, Y.-C. Su, T.-W. Lin, C.-H. Tsai, W.-C. Chang, K.-H. Huang, T.-M. Kuo, S.-W. Lin, Y.-S. Lin, Y.-C. Lu, C.-P. Yang, C.-X. Chang, W.-S. Chin, Y.-C. Juan, H.-Y. Tung, J.-P. Wang, C.-K. Wei, Felix Wu, T.-C. Yin, T. Yu, Y. Zhuang, S.-d. Lin, H.-T. Lin, C.-J. Lin
KDD Cup 2013 Workshop, KDD 2013

Journal Paper

Effective String Processing and Matching for Author Disambiguation. [Paper]

W.-S. Chin, Y.-C. Juan, Y.-Zhuang, Felix Wu, H.-Y. Tung, T. Yu, J.-P. Wang, C.-X. Chang, C.-P. Yang, W.-C. Chang, K.-H. Huang, T.-M. Kuo, S.-W. Lin, Y.-S. Lin, Y.-\ C. Lu, Y.-C. Su, C.-K. Wei, T.- C. Yin, C.-L. Li, T.-W. Lin, C.-H. Tsai, S.-d. Lin, H.-T. Lin, C.-J. Lin
Journal of Machine Learning Research, 2014

CV, Resume(Short version)[ Last Update: 2014/9 ]

Education


Carnegie Mellon University

2015-present

PhD Student
Machine Learning Department, School of Computer Science

Carnegie Mellon University

2013-2015

Master Student
Machine Learning Department, School of Computer Science

National Taiwan University

2009-2013

B.S. in Electrical Engineering

Work Experience


Intel Lab

2015 Summer

Summer Intern, Parallel Computing group

MediaTek

2012 Summer

Summer Intern, Home Entertainment group

Honor and Awards


Member of Eta Kappa Nu

An academic honor society

2013 KDD Cup Award, Track1&2 Champion

The leading Data Mining competition in the world, organized by ACM.
Competition topic on Author-Paper recognition for data from Microsoft Academic Search

2 times NTU President Award

Awarded to top 5% of students in each department of National Taiwan University.
Rank 1/245 in 2012 Spring

NTU 2013 Student Outstanding Performance Scholarship

2012 Altera Innovate Asia FPGA Design Competition, Outstanding Achievement

Most prestigious FPGA design competition in Asia.
Project on FPGA-based gesture recognition 3-D control panel

2nd Place in NTU Green Grass Award (without a first prize winner)

NTU Art Festival Installation Art Contest

Research Ideas


Spectral Methods on non-parametric models

Spectral methods are known for its amazingly short computing time for parameter and topic inference, which leads to nice scalability of the algorithm and can be quite useful with huge amount of data. However, the methods are limited to parametric models in recent works. So, I considered the problem of applying such models on non-parametric model such Indian Buffet Process and Hierarchical Dirichlet Process. You can check my work for more details.

Identification of Overlapping biclusterings

Given a matrix, find a set of submatrices such that the contents in this submatrix follow a specific patterns. This kind of problems is useful in real-world problem like examining on gene expression data and find out the relationship between genes and certain drugs. In previous work, the algorithms and analysis are focused mainly on non-overlapping or single biclusters. If the problems contains overlapping biclusters, we can use either Bayesian and frequentist approaches for finding the submatrices.

Spectral Methods vs optimization

One big problem in spectral learning is that it is actually not solving an optimization problem like variational inference or sampling methods. However, if we want to add restriction to the solution like sparsity, it will be useful that we turn the process of spectral learning into optimization problem and regularization term.

Selceted Projects


Identification of Songbird Species in Field Recordings [Report] [Poster]

Using machine learing (ML) methods to identify bird species from continuous audio recordings has been a hot topic in recent conference competitions. In this project, we used the audio data of songbirds collected by Carnegie Museum of Natural History. First, Audio data are first preprocessed into spectrograms which are further cleaned by applying background noise reduction and image processing techniques. Second, features are then extracted and selected from different sources, e.g., file statistics, segment statistics and probabilities, and mel-frequency cepstral coefficients (MFCCs). Third, the classification is then done by using multiple algorithms, e.g., naive Bayes, k-nearest neighbors (k-NN), support vector machines (SVM), etc. Finally, we used ensemble methods for further exploring some properties on overall performance by combining the predictions of models.

The software developed for this project will be used by the Carnegie Museum of Natural History, and possibly shared with other land managers, researchers, and educators to enhance the use of flight calls as a method to study the patterns of migratory songbirds.

Kinect Game: Soup-Or-Fish-All

This is a multi-player posture-controlled Kinect fishing game. The application has beautiful interface and different level of challenges. Besides, we used elastic formula to simulate the movement of fishing line and hook under the water.

Android Game: CASTLE BUSTLE

A multiplayer tower defense games on nexus 7. The application has professionally-designed interface and characters. To play the game, you need to use a "find friend" method to find your partner. You will build the map together and make your best to attack and defend.

Contact Information

Address

GHC-8208, Gates Hillman Center,
Machine Learning Department,
Carnegie Mellon University,
5000 Forbes Ave,
Pittsburgh, PA 15213

Email

htung at cs.cmu.edu
or
sfish0101 at gmail.com

Phone

(412)961-1335

Links :