Explore a 20-minute conference talk from the FAccT 2021 virtual event that delves into the human-like biases present in image representations learned through unsupervised pre-training. Presented by R. Steed and A. Caliskan, this research-focused presentation covers the Implicit Association Test, methodologies employed, and key findings. Gain insights into image generation techniques and future directions in this field. Understand how unsupervised machine learning models can inadvertently incorporate societal biases, mirroring human prejudices in visual representations.
Image Representations Learned With Unsupervised Pre-Training Contain Human-like Biases
Association for Computing Machinery (ACM) via YouTube
Overview
Syllabus
Introduction
Implicit Association Test
Methods
Results
Key Observations
Image Generation
What Next
Taught by
ACM FAccT Conference