Research Page for Kyungim Baek

Research Interests:

• Computer Vision

My primary research focus lies in computational models and mechanisms of biological visual perception. While the human visual system exhibits an amazing capability to identify objects and analyze complex scenes under widely varying circumstances, current state-of-the-art machine vision systems can perform only rudimentary tasks in highly constrained situations. Building computational models of biological vision systems has been a difficult task because of our poor understanding of the functional mechanisms and anatomical architectures of the visual process. Although it is still far from complete, the field of cognitive neuroscience continues to make progress in understanding the neural mechanisms underlying the visual process. As a result, it has provided hypothesis of how visual information can be represented in the brain, how the visual task can be solved at the computational level, and how the resulting algorithms can be implemented on the known architecture of cortical areas in the primate visual system. Understanding computational mechanisms underlying visual perception and building an artificial vision system based on biological models and theories not only provides a baseline for building more complex, end-to-end vision systems, but also facilitates interactions between computational and biological vision studies by providing feedback to both communities.


• Network models for visual processing

Since information present in the raw sensory data is inherently incomplete, ambiguous, and noisy, visual system has to integrate available sources of information to perform inferential reasoning about a scene. Architecturally, visual cortex appears to be designed for integrating multiple sources of information, yet the underlying computational mechanisms remain mostly unexplained. Recently, much focus has been on probabilistic frameworks for understanding the neural mechanisms and computational principles underlying inference within the brain. They naturally provide the ability to deal with complex uncertainty associated with ambiguous and noisy signal, and to integrate multiple sources of information across space and modality. (For more information, please refer to LIINC website at Columbia University.)


• Biomimetic computer vision: modeling expert object recognition

This project involved developing a computational analogue to expert object recognition in human ventral visual pathway. It was largely inspired by a Kosslyn's psychological model of visual perception, and Tarr's theory of viewpoint-dependent recognition mechanisms covering different levels of perceptual classification and expertise. This work provided an algorithmic analysis of psychological and anatomical models of the expert object recognition, presumably a distinct visual skill implemented by a dedicated anatomic pathway, using the machine vision technologies, such as feature extraction, feature clustering, local linear subspace projection, and matching. (For more information, please refer to Biomimetic Vision website at CSU.)


• Evaluation of face recognition algorithms

A comparison between PCA and ICA on recognizing facial actions and identities was performed. For ICA, InfoMax and FastICA algorithms with two different architectures were considered. For more broader and thorough work on statistical face recognition algorithms, please refer to the HumanID project website at CSU.


• Adaptive object recognition (ADORE)

As a graduate student, I was involved in the development of an adaptive object recognition system called ADORE. The idea of the project was that recognition strategies are sequences of visual procedures, but that the exact set of procedures depends on the target. In ADORE, the task-specific object recognition strategies are learned from examples. The underlying approach is that object recognition is modeled as a Markov Decision Problem, and control policies are trained to sequence vision procedures in order to recognize specific objects in known domains. Backpropagation neural networks have been used to apply reinforcement learning, and bagging (Bootstrap Aggregating) technique was implemented to improve the performance of predictors.


• Bioinformatics


• Metagenomic sequence annotation

In most metagenomic studies, genetic material from a diversity of organisms is sampled from the environment and sequenced using Sanger or 454 sequencing. This process typically results in thousands of DNA-fragments that need to be assembled and annotated before any inferences or conclusions can be drawn from the data in hand. Current metagenomic sequence analysis processes consists of the preliminary assembly of DNA-fragments and their subsequent annotation. However, both the length of the sequenced DNA-fragments and the high level of diversity present in a community result in an increased number of unassembled and unannotated DNA-fragments limiting our capability to better understand the community.
We have been able to improve the taxonomic annotation of unassembled DNA-fragments generated by metagenomic sequencing projects using an automated method relying a) on the Markov clustering (MCL) of all the protein sequences belonging to the same taxon and b) on constructing each taxon's genetic variations (skeleton) using Hidden Markov Model (HMM) profiles. For more information, please refer to Anacle: A skeleton method for taxonomic annotation of metagenomic DNA-fragments.

The documents distributed here have been provided as a means to ensure timely dissemination of scholarly and technical work on a noncommercial basis. Copyright and all rights therein are maintained by the authors or by other copyright holders, not withstanding that they have offered their works here electronically. It is understood that all persons copying this information will adhere to the terms and constraints invoked by each author's copyright. These works may not be reposted without the explicit permission of the copyright holder.

Journal Articles and Book Chapters

• Sajda, P., Baek, K., and Finkel, L., Bayesian networks for modeling cortical integration. Handbook of Neural Engineering, M. Akay (Editor), Wiley/IEEE Press, 2007.
• Draper, B.A., Elliott, D.L., Hayes, J., and Baek, K., EM in high dimensional spaces. IEEE Transactions on Systems, Man and Cybernetics, Vol.35, No.3, pp.571-577, June, 2005. [pdf]
• Baek, K. and Sajda, P., Inferring figure-ground using a recurrent integrate-and-fire neural circuit. IEEE Transactions on Neural Systems and Rehabilitation Engineering, Vol.13, No.2, pp.125-130, June, 2005. [pdf]
• Sajda, P. and Baek, K., Integration of form and motion within a generative model of visual cortex. Neural Networks: Special Issue on Vision and Brain, Vol.17, pp.809-821, June, 2004. [pdf]

Also appeared in S. Grossberg, L. Finkel, and D.J. Field, editors, Vision and Brain: how the brain sees - new approaches to computer vision, Amsterdam: Elsevier Science.
• Draper, B.A., Baek, K., and Boody, J., Implementing the expert object recognition pathway. Machine Vision and Applications, Vol.16, No.1, pp.27-32, 2004. @Springer-Verlag, [pdf]
• Draper, B.A., Baek, K., Bartlett, M.S., and Beveridge, J.R., Recognizing faces with PCA and ICA. Computer Vision and Image Understanding, Vol.91, pp.115-137, July, 2003. [pdf]
• Draper, B.A., Bins, J., and Baek, K., ADORE: Adaptive object recognition. Videre, Vol.1, No.4, pp.86-99, 2000. [pdf]
• Lee, R., Baek, K., Han, H., Jung, W., Jung, H., Im, K.,  Park, M., and Ihm, I., SGRT: Design and implementation of an advanced rendering system. Journal of Korea Information Science Society (C), Vol.4, No.5, pp.633-643, October, 1998.
• Baek, K. and Ihm, I., SGVR: A collaborative volume visualization system. Journal of Korea Information Science Society (A), Vol.24, No.5, pp.417-428, May, 1997.
• Lee, R., Baek, K., and Ihm, I., Enhancing the speed of splatting with indexing and implementation of the user interface. Journal of Korea Information Science Society (A), Vol.23, No.5, pp.443-453, May, 1996.

• Menor, M., Baek, K., Belcaid, M., Gingras, Y., and Poisson G., Virus DNA-fragment Classification using Taxonomic Hidden Markov Model Profiles. ACM-SAC 2010 Conference Track on Bioinformatics and Computational Systems Biology, Sierre, Switzerland, March 22-26, 2010. (Accepted)
• Baek, K., Kim, D.H., and Sajda, P., Inferring direction of figure using a recurrent integrate-and-fire neural circuit. The 26th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBS), San Francisco, California, September 1-5, 2004.
• Baek, K. and Sajda, P., A probabilistic network model for integrating visual cues and inferring intermediate-level representations. IEEE Workshop on Statistical and Computational Theories of Vision (SCTV'03), Nice, France, October 12, 2003. [pdf]
• Draper, B.A., Baek, K., and Boody, J., Implementing the expert object recognition pathway. Winner: Best Cognitive Vision Paper Award, International Conference on Vision Systems, Graz, Austria, April 1-3, 2003.
  Lecture Notes in Computer Science, Springer-Verlag, Vol. 2626, pp.1-11, 2003. [pdf]
• Baek, K. and Draper, B.A., Factor analysis for background suppression. International Conference on Pattern Recognition (ICPR '02), Quebec, Canada. August 11-15, 2002. [pdf]
• Draper, B.A., Baek, K., and Boody, J., A Biologically Plausible Approach to Cat and Dog Discrimination. Joint IAPR International Workshops on Statistical, Syntactical and Structural Pattern Recognition (SSSPR'02), Windsor, Canada, August 6-9, 2002.
  Lecture Notes in Computer Science, Springer-Verlag, Vol. 2396, pp.779-788, 2002. [pdf]
• Baek, K., Draper, B.A., Beveridge, J.R., and She, K., PCA vs. ICA: A comparison on the FERET data set. The 6th Joint Conference on Information Sciences, Durham, North Carolina, March 8-14, pp.824-827, 2002. [pdf]
• Draper, B.A. and Baek, K., Unsupervised learning of biologically plausible object recognition strategies. IEEE International Workshop on Biologically Motivated Computer Vision (BMCV '00), Seoul, Korea, May 15-17, 2000.
  In S.-W. Lee, H. H. Buelthoff, and T. Poggio, editors, Biologically Motivated Computer Vision, Lecture Notes in Computer Science, Springer-Verlag, Vol. 1811, pp.238-247, 2000. [pdf]
• Draper, B.A., Bins, J., and Baek, K., ADORE: Adaptive object recognition. International Conference on Vision Systems, Los Palmas, Gran Canaria, Spain, Jan. 13-15, 1999.
  Lecture Notes in Computer Science, Springer-Verlag, Vol. 1542, pp.522-537, 1999. [pdf]
• Draper, B.A. and Baek, K., Bagging in Computer Vision. IEEE Conference on Computer Vision and Pattern Recognition, CVPR'98, Santa Barbara CA, June 23-25, pp. 144-149, 1998. [pdf]
• Lee, R., Baek, K., Han, H., Jung, W., Eo, S., and Ihm, I., Design and implementation of an advanced rendering system. The Korea Information Science Society Fall Conference, Yongin, Korea, October, 1996.
• Rhee, T., Baek, K., Lee, R., and Ihm, I., Case Study: Construction of sogang volume rendering system V0.55. Computer Graphics Society '95, pp.112-114, Seoul, Korea, November, 1995.
• Baek, K., Rhee, T., Lee, R., and Ihm, I., SGVR: Construction of Scientific Visualization System for Collaboration. Proceedings of  the 22nd Korea Information Science Society Fall Conference, Vol.22, No.2, pp. 563-566, Incheon, Korea, October, 1995.

• Menor, M., Baek, K., and Poisson G., Multiple-Genome Annotation of Genome Fragments Using Hidden Markov Model Profiles, Technical Report, Dept. of ICS, University of Hawaii at Manoa, Jan. 2008.
• Baek, K. and Sajda, P., A probabilistic network model of the influence of local figure-ground representations on the perception of motion. The 5th Annual Vision Sciences Society (VSS) Meeting, Sarasota, Florida, May 6-11, 2005.
[pdf]