I’ve had the honor of being chairperson of Georgia Tech’s Big O Theoretical Computer Science Club for the past two years. Today I moderated a panel on PvNP with Richard Lipton, Scott Aaronson, Lance Fortnow, Ryan Williams, and our club’s founder, Dylan McKay! It speaks to how much the TheoryCS community values inspiring and educating young computer scientists that these giants in complexity theory were willing to spend their night discussing the classic and fundamental problem of PvNP: whether any problem with an efficiently checkable solution is also efficiently solvable. While there are many interesting approaches that may have the potential to solve this problem, the consensus was that we are not close to a solution, but the answer is likely that P does not equal NP (although this was not...</span></li>
We explore how recommendation techniques can be adapted and applied to big data science. Using Globus we derive features specific to big data science and develop a set of data location prediction heuristics. We combine these heuristics into a single recommendation engine using a deep recurrent neural network. We show, via analysis of historical Globus data, that our approaches can predict the storage locations used in user-submitted data transfers with 78.2% and 95.5% accuracy for top-1 and top-3 recommendations, respectively. We presented this work as a SRC poster and an extension as a workshop paper at Supercomputing ‘16. The SRC poster won Best Undergraduate Poster! My amazing mentor Kyle Chard also won the Early Career Researchers in High Performance Computing Award!
Mobile devices currently rely on ineffective passive cooling to manage their temperature. Device temperatures frequently get very high, forcing phones to throttle their CPU, reducing performance. I’m currently researching alternative methods for managing phone temperatures and for balancing performance and heat production to provide a consistently responsive user experience. The following video accompanying our Tiny TOCS 2016 submission summarizes this problem and one solution our group has considered, offloading. We also described some of our work in a poster at HotMobile ‘16. In the video below we demonstrate the temperature reduction our system provides with icecubes!