OUR Featured Researcher: Gaurab Pokharel '21
Gaurab Pokharel (he/him) is an Oberlin College Research Fellow (OCRF) majoring in Computer Science. He conducts mentored research under Professor Adam Eck. His project is titled "Recognition of Emotion Expression from Still Images using CNNs".
Please describe your research:
As technology continues to develop, it has become more and more clear that human computer interaction is just as equally important as the computers capabilities themselves. This interaction would become more natural if computers were able to recognize non-verbal cues from the users, like emotions, and adapt dynamically to what the user is feeling. There have been several efforts to recognize emotions in users already using video and audio input. This project uses a simpler method to see if it is as effective, if not more, than existing methods and use the result to display on-screen messages based on the emotion detected to demonstrate that it is possible to in fact have computers adapt dynamically to user emotions. The project uses VGGNet as a base and the model has been trained over KDEF dataset to recognize six different emotions in users. Preliminary results indicate that, it is possible for VGGNet to recognize emotions effectively with about 60% accuracy. However, with dropout added, that accuracy increases to 70% and finally up to about 85% with Batch Normalization.
Why is your research important?
This project makes use of only still images to predict with high accuracy what the user is experiencing. This is different from existing methods because they use both video and audio data which is computationally expensive to process and are more invasive of user privacy. We managed to re-create comparable results using only still images, which is not computationally expensive to process and is less invasive.
What does the process of doing your research look like?
Once I have a dataset ready to go, I train models using the data. Once the model has been trained, I analyze the results to compare performance, depending on which I can make tweaks to the model to make it perform better. The idea is to have the best performing model.
What knowledge has your research contributed to your field?
We can, with equal accuracy, predict user emotions using only still images, which is less computationally expensive and relatively less complicated to implement in real-world scenarios.
In what ways have you showcased your research?
I have written a paper on my findings with the help of my Mentor and have been accepted into MCURCSM 2019 which is a conference for undergraduate researchers doing research in Mathematics and Computer Science.
How did you get involved in research? What drove you to want to seek out research experiences in college?
I just happened to have a lot of friends doing research on campus, whether it was through OUR or with professors. So, when the opportunity presented itself, I jumped at it and got involved. Because I want to go to graduate school, I felt that having research experience would really help.
What is your favorite part about engaging in this work?
Hands down the results. When you realize that your hypothesis is indeed correct and hours on end of work turns out the way you expect it, nothing can beat that feeling! Also, the learning involved is pretty neat as well.
How has working with your mentor impacted the development of your research project? How has it impacted you as a researcher?
I have managed to learn a lot from him, from scratch too. I have never had any sort of exposure in the field of machine learning before I started working with Professor Eck but now I have a paper written under his supervision. Working with him has definitely helped me settle down on a career path that I think I want to pursue.
How has the research you’ve conducted contributed to your professional or academic development?
I have learned a lot about machine learning and artificial intelligence, how the process works and what it means to do research in this field. Before this, all of the ideas I learnt was akin to magic with a sense of mysticism shrouding the theory. Now that I know how it works, I have come to appreciate the art of it all.
What advice would you give to a younger student wanting to get involved in research in your field?
Just try it out. You never know if you'd like it or not. Even if you don't plan on going to graduate school anytime soon, it can definitely be a rewarding experience. You meet and make connections with a lot of amazing people.