About Me
I am a postdoctoral researcher at the MIT Department of Brain and Cognitive Sciences, primarily working with Dr. Xavier Boix in the Sinha Lab for Developmental Research. My current research is focused on understanding the computational principles that allow for the robust recognition behavior of humans and translating this insight into machine vision algorithms. See my Publications and CV.
Education and Academic Appointments
- Mar. 2008 - Aug. 2014, B.Sc. in Comptuer Science
- Korea University, South Korea, Advisor: Prof. Hee-Jo Lee
- Sep. 2014 - Aug. 2016, M.E. in Brain and Cognitive Engineering
- Korea University, South Korea, Advisor: Prof. Jong-Hwan Lee
- Sep. 2016 - Aug. 2021, Ph.D. in Psychology
- Vanderbilt University, United States, Advisor: Prof. Frank Tong
- Sep. 2021 - Aug. 2022, Postdoctoral Researcher in Psychology
- Vanderbilt University, United States, Advisor: Prof. Frank Tong
- Sep. 2022 - Present, Postdoctoral Researcher in Brain and Cognitive Sciences
- Massachusetts Institute of Technology, United States, Advisors: Prof. Pawan Sinha & Dr. Xavier Boix
Research Interests
At the center of my academic journey has been a desire to puzzle out the complex algorithms that underlie human visual perception. Back in 1999, the movie “The Matrix,” fueled me with a spark of inspiration, and this became my professional and intellectual aspiration to study both Computer Science and Neuroscience. My research objectives revolve around two major themes: 1) studying the computational principles and neural mechanisms that allow us to maintain reliable representations of the visual world, and 2) designing a better machine vision model, inspired by neuroscience literature, that can benefit real-world applications. My current research interests are particularly focused on robust object recognition, face holistic processing, and recurrent computations.
Projects
Understanding a computational basis for holistic face processing
2021 - 2022
Tong, F., & Jang, H. (2022, May). Convolutional neural networks optimized for face recognition reveal a computational basis for holistic face processing [Poster presentation]. Vision Sciences Society, St. Pete Beach, Florida, United States.
Why do we recognize faces in a holistic manner? Does this phenomenon indicate the specificity of faces over non-face objects? In this study, we systematically compared face and object processing with different manipulations that are known to involve holistic processing by using convolutional neural networks.
Potential benefits of blurry vision in robust object recognition
2020 - 2021
Jang, H., & Tong, F. (2022, May). Lack of experience with blurry visual input may cause CNNs to deviate from biological visual systems [Oral presentation]. Vision Sciences Society, St. Pete Beach, Florida, United States.
In a general sense, blurry vision is considered to need correction However, this notion may underestimate the potential benefits of blurry vision in real-world object recognition. We showed that machine models equipped with blurry vision provided a better alignment with biological vision, exhibiting higher neural predictivity under various viewing conditions, stronger noise robustness, and higher shape bias.
Revealing critical differences between face and object processing with a developmental sequence of blurry to clear vision
2019 - 2020
Jang, H., & Tong, F. (2021). Convolutional neural networks trained with a developmental sequence of blurry to clear images reveal core differences between face and object processing. Journal of vision, 21(12), 6-6.
Infant vision is initially very poor but gradually improves over time. We showed that face and object recognition differently benefited from this developmental sequence, suggesting that face and object recognition are fundementally different in terms of utilizing low- and high-spatial frequency information.
Exploring the neural principles of visual crowding in the ventral stream
2017 - 2019
Jang, H., Tong, F. (2019). Visual crowding disrupts the cortical representation of letters in early visual areas. Journal of Vision 2019;19(10):65c. doi: https://doi.org/10.1167/19.10.65c.
Visual crowding is a well-known phenomenon, but its neural principles are not yet fully understood. We examined how visual crowding affects the representations of digits particularly using a decoding approach.
Category-specific and category-general effects of perceptual learning with visual noise
2016 - 2021
Jang, H., Tong, F. (2020). Do noise-trained DNNs process noisy visual images in a more human-like manner?. Journal of Vision 2020;20(11):1776. doi: https://doi.org/10.1167/jov.20.11.1776.
Can training on category A with visual noise generalize to category B? We compared learning effects in humans and machines in terms of category-specific and category-general effects and found that they involved different hierarchical levels of visual processing.
Investigating the representational similarity between noise-trained deep neural networks and the human visual system under challenging conditions
2016 - 2021
Jang, H., McCormack, D., & Tong, F. (2021). Noise-trained deep neural networks effectively predict human vision and its neural responses to challenging images. PLoS Biology, 19(12), e3001418.
How can humans achieve highly reliable performance at recognizing objects under challenging noisy conditions? We investigated the representational aspect of noise-robustness with behavioral, neural, and computational data.
Revealing the hierarchical structure of resting-state networks by deep belief network
2014 - 2016
Kim, H. C., Jang, H., & Lee, J. H. (2020). Test–retest reliability of spatial patterns from resting-state functional MRI using the restricted Boltzmann machine and hierarchically organized spatial patterns from the deep belief network. Journal of Neuroscience Methods, 330, 108451.
We investigated the utility of unsupervised deep learning models (i.e., restricted Boltzmann machine and deep belief network) in revealing resting-state brain networks from the Human Connectome Project.
Developing weight sparsity constraints to avoid the curse of dimensionality issue of deep neural networks for brain data analysis
2014 - 2016
Jang, H., Plis, S. M., Calhoun, V. D., & Lee, J. H. (2017). Task-specific feature extraction and classification of fMRI volumes using a deep neural network initialized with a deep belief network: Evaluation using sensorimotor tasks. NeuroImage, 145, 314-328.
We applied deep learning models to classifying task-based fMRI volumes across four sensorimotor tasks (i.e., left-hand clenching, right-hand clenching, auditory attention, and visual stimulus) and examined the hierarchical weight representations of the networks.
Contact
If you have any questions or inquires, please email me at the address below.
Location:
Massachusetts Institute of Technology, 77 Massachusetts Avenue, Cambridge, MA 02139
Email:
jangh@mit.edu