Erasure of Unaligned Attributes from Neural Representations
This addresses bias removal in AI systems for fairness applications, but it is incremental as it builds on existing attribute removal methods.
The paper tackles the problem of erasing implicit, unaligned attributes from neural representations, presenting the AMSAL algorithm that alternates between assignment and projection steps. Results show bias can often be removed across multiple datasets, including Twitter and BiasBench, though limitations exist when task and information are strongly entangled.
We present the Assignment-Maximization Spectral Attribute removaL (AMSAL) algorithm, which erases information from neural representations when the information to be erased is implicit rather than directly being aligned to each input example. Our algorithm works by alternating between two steps. In one, it finds an assignment of the input representations to the information to be erased, and in the other, it creates projections of both the input representations and the information to be erased into a joint latent space. We test our algorithm on an extensive array of datasets, including a Twitter dataset with multiple guarded attributes, the BiasBios dataset and the BiasBench benchmark. The last benchmark includes four datasets with various types of protected attributes. Our results demonstrate that bias can often be removed in our setup. We also discuss the limitations of our approach when there is a strong entanglement between the main task and the information to be erased.