UM Data Science Research Seminar

Schedule

Presentation 1

Time: 12:00 - 12:30

Speaker: Tjitze Rienstra

Title: Ranked Programming

Abstract: 
In artificial intelligence, uncertainty is typically expressed using probabilities. Probabilities are, however, not appropriate for every task. Consider a fault diagnosis problem, where we know that components are “normally” good and “exceptionally” bad, but where we do not know (nor are we interested in) precise probabilities of these events. Or we are processing sensor data or user input, and we assume that each piece of information we receive is “normally” correct and “exceptionally” wrong, again without knowing anything in terms of probabilities.

In this talk I will first focus on ranking theory, an alternative to probability theory developed within epistemology that is suitable for reasoning about this kind of uncertainty. In ranking theory, uncertainty is measured with ranks, which are non-negative integers that represent degrees of exceptionality. Then I present a so called ranked programming language. Based on the probabilistic programming paradigm, ranked programming uses ranking theory as its underlying uncertainty calculus, which means that uncertainty is expressed with constructs that specify normal as well as exceptional program flow. I will discuss a number of inference tasks easily implemented with ranked programming as well as a number of directions for future work.

Presentation 2

Time: 12:30 - 13:00

Speaker: Marijn ten Thij

Title: Measuring distorted thinking in written text

Abstract: 
Depression is a leading cause of disability worldwide, but is often underdiagnosed and undertreated. Cognitive behavioural therapy holds that individuals with depression exhibit distorted modes of thinking, that is, cognitive distortions, that can negatively affect their emotions and motivation. In this talk, I will introduce a novel approach, called Cognitive Distortion Schemata (CDS), that measures these maladaptive patterns of thinking in written text, irrespective of the content in which they occur. Next, I will show how we applied CDS to measure that the language of individuals with a self-reported diagnosis of depression on social media is characterized by higher levels of distorted thinking compared with a random sample.