Grad Cohort 16 Review

I attended Grad Cohort 2016 with three goals in mind: finding an internship position aligned with my research interests for summer, connecting with PhD students in HCI (Human Computer Interaction) and understanding their experience and the path they are following, making the most of the workshops geared toward the third year PhD students as the audience.

To fulfill the first goal, I secured an internship as a HCI researcher at Intel Labs, fully aligned with my research interests. The hiring manager reached out to me few days before the conference, explaining that she found my name as one of the Grad Cohort attendees and she was thinking that my research background would be a fit with her open internship position. We exchanged a few emails regarding the position and the research papers we had read. Even though I did not meet with the hiring manager (she did not attend the conference), meeting with  the head of HR at Intel Labs who provided my name to the hiring manager in the first place, left an impression which played a role in the process of receiving the internship offer. Had I not won the scholarship to attend Grad Cohort, the hiring manager would have never reached out to me, and I would not have been considered for such a competitive internship role at Intel Labs.

To accomplish the second goal, I managed to meet with a few scholars in HCI, and attended the 3rd year PhD students’ sessions. I met with Saeideh Bakhshi (an HCI researcher from Yahoo) and attended her talk on dark moments in everyone’s PhD life and how to come in terms with them by keeping the positive attitude. That motivated me to attended her research talk at CHI 2016 conference where she presented her paper on use of GIF files as a substitute to emoticons. I also met with A.J. Brush (Microsoft) but unfortunately could not attend her presentation, as it conflicted with “how to move to candidacy” talk geared toward 3rd year PhD students. In addition, I had a chance to meet with Deb Agarwal again, whom I first met with at Grace Hopper few years ago. It was a pleasure to talk to her again and also attend her talk on Career path. She told me that once I have two publications in CHI, I should reach out to her for a post-doc position at UC, Berkeley. I still have that in mind and it keeps me motivated submitting papers to CHI.

Fulfilling the third goal, I met with other HCI female researchers to understand their perspective on HCI research, the challenges they are facing at other schools, tools and techniques they are using, and their prospects of the future directions of the field. My reflections on discussions with some of HCI researchers are as follows:

I learned that most of the HCI PhD students, specially those attending top HCI programs such as CMU, are now invested in educating themselves in machine learning and natural language processing. Therefore, getting more expertise in deep learning (use of deep architectures of recurrent, recursive, or convolutional neural networks to perform predictions, language translations, image labeling, etc.) as one of the advanced areas of machine learning, is not too of a crazy idea for an HCI researcher. Indeed, it offers a unique advantage, if a researcher is to either stay in academia or join a research lab.

To elaborate the advantage of deep learning knowledge, first I need to state one common pitfall of HCI researchers; feature engineering thinking. In this mentality, any phenomenon consists of extractable features, and an experienced researcher can potentially find the most influential features. Perhaps, this thinking is the reason why HCI researchers take qualitative data analysis courses and gain expertise in “coding”(finding patterns in data). However, such thinking does not scale well when it comes to understanding complex phenomena (e.g., predicting human emotion in response to a novel experience via concurrent inputs of heart rate, respirations, and facial expressions). Can a researcher safely claim that, given a certain heart rate and facial expression, the predicted emotion would always be an excitement? Maybe or maybe not. Considering this example and many others, it seems that feature engineering thinking is a useful technique but have severe limitations.

Recently, an alternative thinking of having neural networks learn features through seeing a lot of examples (deep learning) suggested to be a more promising future direction in machine learning. I think, adopting such a new mindset for HCI researchers would also critical. By wearing neuro-science lenses, an HCI researcher can delegate the task of feature extraction (“coding”) of a complex phenomenon to neurons. Then, by learning neurons patterns and excitement behaviors, an HCI researcher can gain a more realistic perspective as to what features and how frequently those features excite the neurons that leads to desired outcome.

My speculation of use of more machine learning methods, specifically deep learning, in HCI, was further validated as the future direction of the field when I attended CHI 2016 (one of the top conferences in HCI) this year(please read my post on CHI16 here on the topic).

The discussions were over Grad Cohort first and second day two-hour lunch meetings with other HCI researchers. We wish he had an assigned mentor (an HCI expert either from academia or industry) who had joined our table during the lunch sessions. This is, indeed, one of my suggestions with regard to improving Grad Cohort experience, as I highly cherish and encourage topic-specific lunch meetings led by mentors.

In addition, I suggest that Grad Cohort provides an opportunity for attendees to have collaborative group-work experience. This idea was put into practice by Dr.Beck <> from University of Washington hosting “Big Data” workshop in 2016, funded by NSF. I was one of the 100 scholars who attended the workshop and I enjoyed the experience very much.

The idea was to group attendees and set goals and expectations for them through the workshop. The process involved teammates’ brainstorming sessions under mentors’ supervision, answering pre-defined questions, and finally presenting a short talk at the closing event.

To illustrate the process, before arrival to the conference, attendees were grouped and given a name (for example, fish, bird, cat, etc.). Everyone in a group were PhD students in their x-year of studies, each from a different university. For example, the gold fish group were ten first-year PhD students in HCI from UC, San Diego, CMU, etc. Each group had a mentor, from industry or academia, who helped them through the process. A shared Google doc was assigned to each group with a set of pre-defined questions, concerning the big data context. For example, in the context of Grad Cohort conference, the relevant pre-defined questions could be about how students define their field of study, what challenges they face in the field, how they envision future of the field, what tools and educational material they recommend to new scholars joining the field, etc.

The students met in a group during the lunch and dinner time with their group mates and mentor to brainstorm and create a short (5 to 7 min) presentation addressing the pre-defined questions. During the closing night event, some of these groups were selected at random to perform their presentations to the audience (including scholars from both academia and industry). Then, the best presentation was selected by the judging committee and were granted an award.

I speculate that this would be a great closing to Grad Cohort, as we would learn from each other and have access to every team’s Google documents as a reference. In addition, we would be more invested to attend the conference sessions and use what we learn in them, in the final presentation. I presume lots of attendees would enjoy such closing to the Grad Cohort.

Overall, I enjoyed wisely picked sessions, meeting with great speakers, and of course good poster sessions.


Posted in LDA.