Title: A Plea for History and Philosophy of Statistics and Machine Learning—with a Little Story of Achievabilism

Hanti Lin (University of California, Davis)

Abstract

Philosophy of science encompasses a broad spectrum of approaches, ranging from the use of mathematical tools in formal epistemology to the focus on scientific practice in history and philosophy of science (HPS). Despite their differences, integrating these approaches is essential, especially when examining the scientific fields that study scientific inference, such as statistics and machine learning. I illustrate this integration with a historical case study of an epistemological idea I call achievabilism—the thesis that the standards for assessing inference methods and learning algorithms should not be invariant but instead sensitive to what is achievable within specific problem contexts. Achievabilism appears crucial to the foundations of statistics and machine learning, yet it has rarely been explicitly articulated and has instead been practiced largely implicitly. This has led to its repeated reinvention by figures such as Putnam (1965) and Gold (1967) in formal learning theory, and Devroye et al. (1996) in statistical/machine learning theory, although its origin can be traced back to Neyman and Pearson (1936) in classical statistics.

Date: Mar. 14th, 2025

Time: 11am – 12pm (CT)

Zoom Link:

https://rutgers.zoom.us/j/99554502801?pwd=rXCCcpqG9LUTdvBDaZIOqnDbp0wFS9.1

Zoom Meeting ID: 995 5450 2801

Passcode: 699310