Fair Algorithms

Yang Liu

Computer Science and Engineering, UC Santa Cruz

Machine learning (ML) is increasingly used in domains that have a profound effect on people's opportunities and well-being, including healthcare, law enforcement and consumer finance. A ML-trained system may appear to be fair without human intervention, this talk argues that this is far from being the case. A typical ML system consists of a pipeline of the following major components: data collection, model training and model deployment, and I’d like to give a high-level overview of the potential biases or discriminations that may arise in each of above components. This talk will also raise awareness of the following questions: 1) How to guarantee the quality of data collected from potentially careless and even malicious human agents, instead of treating the data as if they were clean and representative (blind trust)?  2) How to build ML methods that are robust despite noise in the data (bias in training data)? 3) How to guarantee fair and transparent treatment of people when ML models are deployed (bias in model and algorithm)