Monday, October 2, 2017

UB CSE 50 celebrations, The Graduate Research Conference

Over the last couple of days, we celebrated the 50th anniversary of our department, CSE at University at Buffalo. We celebrated the anniversary with a technical conference and panels, so it was an opportunity to learn new things for everyone. With the attendance of many prominent UB-CSE alumni, it has been a really amazing 2.5 days. Here is a link to the CSE 50 conference program.

On Thursday evening, the event was kicked off with a reception and an undergraduate poster session. The thing that surprised me in this poster session was how quickly the PM2.5 sensors miniaturized. PM2.5 refers to atmospheric particulate matter (PM) that have a diameter less than 2.5 micrometers, which is about 3% the diameter of a human hair. PM2.5 is a traffic-related pollutant implicated in a wide variety of adverse health outcomes. I was involved in a NIH/NIEHS project for using smartphone-based time-activity data for air pollutant exposure estimation from 2010-12. At that time PM2.5 sensors were mini-fridge sized and expensive to buy and deploy. On the poster demo, my jaw hit the floor when I saw the new generation of PM2.5 sensors that are palm-sized and are connected to Arduino boards.

The Friday conference consisted of 3 keynote presentations and 4 sessions. The sessions were a mix of invited alumni talks and our own graduate students unpublished original paper presentations.

I was the program chair for the Friday conference, and was responsible for selecting the graduate papers. I used EasyChair to collect the paper submissions and review them. We formed a technical program committee of 22 alumni/faculty. Out of 21 paper submissions, we selected 8 for the Friday program. While all the submissions were high quality, we had to be selective to keep to the time constraints. We also processed 50 poster submissions, and chose 29 papers among them for the graduate poster presentation on Saturday.

Here are my notes from some of the talks on Friday.

Keynote 1 - Dr. Victor Bahl (Microsoft Research) "Democratization of Streaming Video Analytics & the Emergence of Edge Computing"

Victor got a BS & MS degree from ECE at UB at the end of 80s. He is a  Distinguished Scientist and Director of Mobile & Networking Research at Microsoft. 

His talk was about edge computing, looking beyond cloud computing. In a 2013 meeting in Microsoft, he had claimed that by 2020, cloud computing would be disaggregated and augmented by edge/fog computing. He defended edge computing putting forth latency/bandwidth, expense/service, and battery-saving reasons.

Since then he was involved in proving the utility of edge computing with killer applications. He talked about "Glimpse: continuous realtime object recognition of mobile devices" from Sensys 2014 as one application. Another application is the connected car. In 2015, they came up with ParkMaster, edge-powered in vehicle analytics for detecting open parking spaces in urban environments. As you drive your smartphone detects (and then uploads to cloud) empty parking spaces for others to later park in that street. Your car provides service to others, and in return others provide the same service to you.

Yet, as another application of edge computing, he pursued surveillance of public buildings. The idea is to do the filtering/analysis of video feeds right in the building machines, instead of uploading the videos to cloud for remote offline analysis.

And finally, most recently, he has been applying the edge computing concept to  live video analytics of traffic cameras at intersections. This project serves by collecting traffic video analytics and data for the Vision Zero project. Vision Zero is a multi-national road traffic safety project started in 1997 that aims to achieve a highway system with no fatalities or serious injuries involving road traffic. The project is deployed and in use in Bellevue and Seattle streets, and is in progress to be deployed in Cambridge UK.

Invited talk: Prof. Bruce Shriver (Liddy Shriver Sarcoma Initiative), "Unconventional Computer Architectures"

Bruce started his PhD at University at Buffalo CS department in 1968 and got his PhD in 1971. His talk was about rethinking/rebooting computation and computers and touched on many topics including neuromorphic engineering. (This topic was also revisited by Dr. Steve Kang, another of our alum, in his Saturday's keynote titled "The 4th Industrial Revolution and Future of Nanoscience and Engineering".)

Bruce has been interested in how the human brain organizes, stores, accesses and understands sensory input and its accumulated knowledge, and yet run with such a small power requirement. The recent success and wide adoption of CRISPR has invigorated the area. Bruce's presentation referred to several recent articles in the area, including:

  • DNA Fountain enables a robust and efficient storage architecture, Elrich et al, Science, March 2017
  • Model-based design of RNA hybridization networks implemented in living cells, Rodrigo et al, Nucleic Acids Research, September 2017
  • Complex cellular logic computation using ribocomputing devices, Green at al, Nature, July 2017
  • Silicon quantum processor with robust long-distance qubit couplings, Tosi et al, Nature Communications, September 2017
  • A Brain Built From Atomic Switches Can Learn, Quanta Magazine, Andreas von Bubnoff, September 2017
  • On-chip generation of high-dimensional entangled quantum states and their coherent control, Kues et al, Nature, June 2017
  • A million spiking-neuron integrated circuit with a scalable communication network and interface, Merolla et al, Science, August 2014
Bruce pointed out that these unconventional new architectures would need new type of algorithms, a type we do not have yet. He urged the audience to think about what type of algorithms, software, programming language, OS, hardware, and programmers would be needed to address these challenges. Bruce conjectured that we should see breakthroughs via molecular computing I/O.

Student paper presentations

Among the student papers, some of the most interesting ones for me were the following.

  • "Metadata-based Feature Aggregation Network for Face Recognition" by Nishant Sankaran, Sergey Tulyakov, Srirangaraj Setlur and Venugopal Govindaraju. The idea here is to use metadata (yaw, pitch, roll, gender) unperturbed by the feature extraction process to gauge quality of facial image.
  • "Differentially Private Empirical Risk Minimization with Non-convex Loss Function" by Di Wang and Jinhui Xu.
  • "Emulating Quantum Circuits Via Boolean Formulas and #SAT Solving" by Chaowen Guan and Kenneth Regan.

No comments: