Page 68 - Undergraduate Research Opportunities Program Proceedings 2017-18

School of Engineering
Department of Computer Science & Engineering

constrains on the vertices are also constructed and analyzed. Moreover, discussions on possible further extensions and
research ideas are also proposed

Machine Learning on Wearable Devices

Supervisor: HUI Pan / CSE
Student: CHO Hyun Dong / COGBM Course: UROP1100, Spring

With the advances in sensors, electronic devices have enjoyed the luxury of carrying more sensors at a fraction of their
previous costs. An abundance of sensors enable a comprehensive collection of data regarding the surroundings of the
device, and this has been most evidently perceivable in our smartphones. One of the most useful information that could
possibly be deduced from such data is the emotional state of the user of the wearable device, as emotional information
has implications for mental health, accurate recommendation systems, etc. Understanding the importance of such
information, this research seeks to mimic previous work in taking advantage of data collected through mobile phones in
realistic, non-intrusive settings, and discover areas for improvement during that process.

Supervisor: HUI Pan / CSE
Student: HU Zeou / COSC Course: UROP1100, Fall

Machine learning technology is widely used in computer vision nowadays to solve many real-life classification problems.
In this project, we will try to use Android phone (as a replacement for wearable devices) to capture images of human
facial expressions and try to use machine learning techniques to predict the correct emotion. We will use some free
well-known facial expression datasets as our foundation of work, mainly FER2013 and CK+ dataset. In order to get good
classification result, this project will try to use Convolutional Neural Network (CNN) to train and predict outcomes, with
utilization of some deep learning frameworks (mainly Tensorflow), and also use transfer learning to ease the training
process of deep CNN networks.

Supervisor: HUI Pan / CSE
Student: LIU Heshan / COMP Course: UROP1100, Fall

The research aims to find a best algorithm to complete the real-time emotion recognition task on the wearable devices
such as mobile phone and google glasses. In this research, we first design several machine learning emotion recognition
algorithms based on some existing facial recognition algorithm and test them on some famous emotion dataset such as
CK+ and Yale dataset to find the best algorithm. Then we try to implement it in android system to see whether it also
works well in real-time recognition.

Supervisor: HUI Pan / CSE
Student: SUN Yusen / PHYS Course: UROP1100, Summer

Nowadays, photos of food are ubiquitous. People are used to upload photos of their food to the social networking
website or Apps such as Facebook and Instagram before eating. Besides, most restaurants also upload their food photos
online as an advertisement to attractive customers. However, while some of the food images look attractive, some do
not. Some features of a photo determine its attractiveness. In this project, we analyzed the attractiveness of the food
photos by extracting and analyzing different handcraft features such as colors and texture. By utilizing features from
these two main categories, the accuracy to test whether a food is attractive or not solely from its photo has successfully
reached 80.74%.

58