Hi! I am a 3rd year Computer Science graduate student at UCSB. I received my Bachelor of Science in both Physics and Math from HKUST in 2015, and master's degree in Computer Science from HKUST in 2017.
My research motivation is to create more intelligent artificial models. More specifically, I want to bridge real and artificial neural networks and find better schemes for automatically finding suitable model architectures (Neural Architecture Search) with the inspiration gotten from brain network modeling. I also did some projects in computer vision before.
I believe in, welcome, and want to contribute to Artificial general intelligence (AGI).
Research statement as of Oct. 2018
CS165A: Artificial Intelligence (Winter 2018)
CS180: Computer Graphics (Fall 2017)
COMP4331: Data Mining (Fall 2016)
The most exciting phrase to hear in science, the one that heralds new discoveries, is not 'Eureka!' (I've found it!), but 'That's funny...' -Isaac Asimov.
C   u   r   r   e   n   t
We are working on modeling the heterogeneity of brain MRI data through optimization approaches. By estimating the inverse covariance matrix of fMRI data under network constraints, we find pathways of brain activation when subjects are doing different tasks. We are also investigating the relationship between structural and functional networks, which can be useful for predicting brain activities based on specific structural connectivity. Taking sample variations into account, it is more reasonable to jointly estimate multiple models for heterogeneous groups, instead of estimating a single model for the whole population. We are also solving this through optimization approaches as well.
P   a   s   t
White noise analysis of deep neural networks
Classification images and spike triggered analysis have been widely used in psychophysics and neurophysiology to understand underlying mechanisms of sensory systems in humans and monkeys. This project leverages these techniques to investigate the inherent biases of deep neural networks and to obtain a first-order approximation of their functionality. We emphasize on CNNs, but also studied MLPs, RNNs, and logistic regression, with experiments over MNIST, Fashion-MNIST and CIFAR-10.
Every person has his/her own preference when paying attention to the surroundings. A general attention model is not enough when trying to find the personalized salient regions. This work proposes a convolutional neural network model to predict the personalized attention region when a person sees an image, utilizing established detection and saliency prediction techniques.
Augmented Reality (AR) is the next big thing for mobile and wearable devices, but it requires a lot of resources. Sound/Image processing is expensive and require large knowledge bases - they cannot be done locally on a device. The objective of the CloudAR project is to develop the mobile cloud computing technology and the in-cloud big data processing algorithms to enable an AR ecosystem.
Ubii: Ubiquitous interface and interaction
Ubii is an integrated system implemented on the Google glass that put AR and ubiquitous computing together. Upon it, users can interact with smart devices that are connected together over the network, using simple hand gestures as drag and drops. Currently users can transfer files between computers, drag files to printer for printing, display files on screens or walls, and pair bluetooth devices. Ubii provides a more convenient and intuitive natural user interface, attaining a seamless interaction between the physical and digital worlds.
Zhang, W., Lin, S., Bijarbooneh, F. H., Cheng, H. F., & Hui, P. CloudAR: A Cloud-based Framework for Mobile Augmented Reality. Thematic Workshops of ACM Multimedia 2017. File
[Master Thesis] Where's YOUR focus: Personalized Attention
Lin, S., Cheng, H. F., Li, W., Huang, Z., Hui, P., & Peylo, C. Ubii: Physical World Interaction Through Augmented Reality. IEEE Transactions on Mobile Computing. File