As machine learning systems are increasingly used to make real world legal and financial decisions, it is of paramount importance that we develop algorithms to verify that these systems do not discriminate against minorities. We design a scalable algorithm for verifying fairness specifications. Our algorithm obtains strong correctness guarantees based on adaptive concentration inequalities; such inequalities enable our algorithm to adaptively take samples until it has enough data to make a decision. We implement our algorithm in a tool called VeriFair, and show that it scales to large machine learning models, including a deep recurrent neural network that is more than five orders of magnitude larger than the largest previously-verified neural network. While our technique only gives probabilistic guarantees due to the use of random samples, we show that we can choose the probability of error to be extremely small.
Wed 23 OctDisplayed time zone: Beirut change
14:00 - 15:30 | |||
14:00 22mTalk | Duet: An Expressive Higher-Order Language and Linear Type System for Statically Enforcing Differential Privacy OOPSLA Joseph P. Near University of Vermont, David Darais University of Vermont, Chike Abuah University of Vermont, Tim Stevens University of Vermont, Pranav Gaddamadugu University of California, Berkeley, Lun Wang University of California, Berkeley, Neel Somani University of California, Berkeley, Mu Zhang University of Utah, Nikhil Sharma University of California, Berkeley, Alex Shan University of California, Berkeley, Dawn Song University of California, Berkeley DOI | ||
14:22 22mTalk | Improving Bug Detection via Context-Based Code Representation Learning and Attention-Based Neural Networks OOPSLA Yi Li New Jersey Institute of Technology, USA, Shaohua Wang New Jersey Institute of Technology, USA, Tien N. Nguyen University of Texas at Dallas, Son Nguyen The University of Texas at Dallas DOI | ||
14:45 22mTalk | Probabilistic Verification of Fairness Properties via Concentration OOPSLA Osbert Bastani University of Pennsylvania, Xin Zhang Massachusetts Institute of Technology, Armando Solar-Lezama Massachusetts Institute of Technology DOI | ||
15:07 22mTalk | Generating Precise Error Specifications for C: A Zero Shot Learning Approach OOPSLA Baijun Wu University of Louisiana at Lafayette, John Peter Campora University of Louisiana at Lafayette, He Yi University of Louisiana at Lafayette, Alexander Schlecht University of Louisiana at Lafayette, Sheng Chen University of Louisiana at Lafayette DOI | ||