Introduction在线视频

Abstract: 

The problem of theory choice can be hard even for a realist but still important even for an instrumentalist. Namely, we can find some cases in which useful truths are underdetermined by the kinds of data we can have access to ethically or practicably, but even an infinity of such data would not remove the underdetermination in question. This talk will address a causal version of the above problem: the problem of inferring causal relations from non-experimental data (which is important for, say, epidemiologists and policy makers). To this causal version of the problem, the now-standard solution proceeds with an assumption about how causation and chance are actually related, an assumption that works simply by ruling out some skeptical scenarios that incur underdetermination. (For those who are familiar with causal Bayesian networks, the assumption I have in mind is the so-called Faithfulness assumption.) The goal of this talk is to show you how the problem can be solved without making such a factual assumption about causation. The crux lies not in the factual, but in the normative, the evaluative, and the mathematical---or so I will argue. The result is a new way of doing formal epistemology and a new class of (learning-theoretic) theorems in statistics and machine learning. 

 Speaker

Hanti Lin is an assistant professor of philosophy at the University of California, Davis. His philosophical work concerns the epistemology of scientific inference. His technical work belongs to the more theoretical areas of machine learning and statistics, such as learning theory and causal discovery/inference.