Dr. Lin guest talk on Errors in Machine Learning
The guest talk is part of a collaboration with University of Arts (Helsinki) in the context of Helsinki Biennial 2023.
Oplysninger om arrangementet
Online By Zoom, click register for link
For access to the online Zoom link, please contact professor Jussi Parikka at email@example.com
Welcome to this online guest talk by Dr Cindy Lin on Rethinking Errors in Machine Learning. The talk is organized jointly by the Design and Aesthetics for Environmental Data project (Aarhus University) with the University of Arts (Helsinki) as part of the joint research studio on “Environment, Data, Contamination”. The studio is part of a collaboration in the context of Helsinki Biennial 2023.
Social scientists and critical computing scholars point to errors in machine learning (ML) to highlight how ML applications have undesirable effects, from the values that prefigure facial recognition to the lack of transparency in reproducing ML results. More recently, environmental scientists have also highlighted the errors of ML in the environmental sciences, such as the lack of geo-diverse and representative training data to the underprediction of extreme weather events in rural and remote areas. But what if error is central to the creation, organization, and infrastructure of ML knowledge? Drawing from a series of case studies on environmental mapping, this talk examines the value of error within ML work. The talk presents an ethnographic analysis of how ML datasets are developed and evaluated to argue two points: First, errors disclose existing structures of collaboration, often undervalued or overseen in supposedly working systems, and second, errors rework old actors and sites in new ways, that re-enter and/or devalue the position of different actors.
Dr. Cindy Lin is an information scientist and ethnographer. Her work centers on the data practices, exchanges, and expertise of climate change and their relationship to race and environmental governance in Indonesia and United States. Her work is situated at the intersections of feminist science and technology studies, critical data studies, and postcolonial studies. Prior to her professorship at Penn State, she was a visiting postdoctoral fellow at Cornell Tech's Digital Life Initiative as well as a postdoctoral fellow at Cornell Atkinson Centre for Sustainability and Cornell's Department of Information Science. She earned a doctoral degree from the School of Information at the University of Michigan. Dr. Lin's scholarship has appeared in computing venues such as ACM CHI, DIS, and PD as well as humanistic venues such as Social Text, e-flux, University of Nebraska Press, and elsewhere. She has also co-authored two multigraphs, Technoprecarious (MIT Press/Goldsmiths Press, Nov 2020) and Digital Energetics (University of Minnesota Press, June 2023).