The Abacus.AI team has published two papers to appear at the Conference on Neural Information Processing Systems (NeurIPS) 2020. NeurIPS is a top machine learning conference held every December. In this post, we give a short description and a 3 minute video for each paper.
Encodings for Neural Architecture Search
Our first paper is about encodings for neural architecture search, and it was accepted as a spotlight paper. Automated machine learning (AutoML) and neural architecture search (NAS) are rapidly growing areas of research, but it is not well-understood how to choose the best architecture encodings for each AutoML algorithm. This paper provides the first formal study of encodings for neural architecture search. For more information, see the links below.
Post-Hoc Methods for Debiasing Neural Networks
Our second paper is about methods for debiasing machine learning models. As deep learning models become tasked with more and more decisions that impact human lives, such as hiring, criminal recidivism, and loan repayment, bias is becoming a growing concern. This paper initiates the study of a new class of debiasing techniques, called intra-processing methods.
Blog post: https://abacus.ai/blog/2020/07/14/debiasing-facial-prediction-models-with-adversarial-fine-tuning/
If you’re interested in learning more, you can come to our poster sessions at the NeurIPS Conference, December 7-10. It is held virtually this year. Register at https://neurips.cc/.