Presented a poster and hardware demonstration illustrating how both electromagnetic and acoustic interference can compromise systems that place blind trust in sensor readings.
Abstract: The central problem in empirical economics is to identify the causal effect of a “treatment” on outcomes of interest. This task is complicated, however, by potential correlations between latent variables and the outcome. Economists traditionally overcome this problem by use of instrumental variable analysis, which relies on strong assumptions of both linearity and homogeneity. We do away with these assumptions by using deep neural networks to estimate both treatments and outcomes, and validate our model by replicating the results of a seminal economics paper that studies the relationship between Chinese imports and American manufacturing employment.
Course project in which we implemented a 2-factor-authentication scheme which requires absolutely no user interaction. To accomplish this, we used the electromagnetic radiation emitted by a machine's memory bus lines to authenticate with a user's cell phone.
My first task as a Ph.D. student was to spend a month reproducing the results of a recent paper. I chose GSMem, a paper from Usenix Security Symposium 2015 in which the authors demonstrated how to establish a covert channel on top of the electromagnetic radiation emitted by a machine's memory bus lines, and thus exfiltrate data from an airgapped machine.