Edgar Minasyan
MIT EECS | Lal Undergraduate Research and Innovation Scholar
Parametric Inversion of Non-Invertible Functions
2017–2018
EECS
- Artificial Intelligence & Machine Learning
Armando Solar Lezama
Many of the most important problems in artificial intelligence can be formulated as inverse simulations of computer programs; however, only rarely are programs literally simulated in reverse. Inverse simulation is hard because (a) non-injectivity of the forward program renders the inverse grossly ill-posed, and (b) it requires search over complex, highly constrained spaces. We introduce parametric inversion in response to both of these concerns and present algorithms to simulate non-trivial programs in reverse. A parametric inverse is a generalized inverse function; it uses a parameter to disambiguate between the multiple inputs that a non-invertible forward function maps to the same output.
During my three years at MIT, I have developed a strong passion for research, so I want to participate in SuperUROP. I’ve already been doing UROP for my whole junior year and will continue extensive research with the same group and graduate student. Eventually, I hope the experience I gain from this program will help me in my pursuit of graduate degrees.