Pardis Semnani

In today's world, machines manage much of our daily lives, from self-checkout systems to self-driving cars. Ensuring the algorithms behind these machines function properly is crucial. My research centers around devising efficient algorithms that work under realistic assumptions and are theoretically sound. In many fields, a central task is learning the causal effects among variables of interest. Examples include assessing the effectiveness of a treatment on a certain disease or discovering the effects of different factors on a nation’s economic growth. Part of my research concerns such questions, with a focus on scenarios where at least two variables exhibit mutual causal effects on one another, possibly mediated by other variables. Such situations are common in the real world. For instance, the emission of greenhouse gases both contributes to global warming and is impacted by the melting of permafrost, itself a consequence of global warming.

Nevertheless, the number of algorithms able to handle these cases is limited, which motivates my intention to address them. Another direction I pursue involves creating algorithms that can learn the general pattern of the outcomes of a series of random experiments after merely observing a finite number of these outcomes. This enables nearly accurate predictions of future outcomes. The challenge, however, lies in the fact that one needs to rely on some assumptions about these patterns in order to learn them. While more realistic assumptions result in more accurate predictions, they also demand more complicated algorithms. My objective is to achieve the right balance between the restrictiveness of these assumptions and the efficiency of the resulting algorithms.