Tianbao Yang

Associate Professor and Herbert H. Richardson Faculty Fellow
Computer Science and Engineering, Texas A&M University
Email: [first-name]-[last-name] at tamu.edu

LibAUC: A Deep Learning Library for X-risk Optimization

The LibAUC Library is an open source library that we launched in April 2021. We have developed several practical and efficient algorithms for large-scale training of deep neural networks for optimizing a family of X-risks (e.g., AUROC, partial AUC, AUPRC, AP, NDCG, Contrastive Loss) for learning deep neural networks. Our LibAUC library has achieved great success on various problems (e.g., 1st Place at Stanford CheXpert Competition, 1st Place at MIT AICures Challenge, beating Tensorflow Ranking Library). We will continuously develop the library to benefit the community. For more detials, please check our LibAUC library website. If you are interested in the research of LibAUC on your data, we are glad to collaborate.

Birds: A Distributed Library for Classification and Regression

The birds library is a distributed library aiming to solve big data classification and regression problems in a distributed enviorment (a cluster of machines) or parallel fashion (a multi-core machine). It implements a practical distributed stochastic dual coordinate ascent proposed in the following paper: Trading Computation for Communication: Distributed Stochastic Dual Coordinate Ascent published at NIPS, 2013. The two ideas proposed in this work namely local updates and model averaging have become the norm of Federated Learning.

The library can solve the following linear classification and regression problems:

  • SVM (hinge loss and squared hinge loss)
  • Logistic Regression
  • Least Square Regerssion
  • with L2 regularization, with/without L1 regularization
  • one-vs-all multi-class classification