CARAML Lab effiCient, fAir, Robust, and Active ML Lab

LEVERAGING ADVANCES IN OPTIMIZATION FOR ENABLING EFFICIENT, ROBUST, AND SCALABLE MACHINE LEARNING

CARAML Lab is a research group at the University of Texas at Dallas, Department of Computer Science led by Prof. Rishabh Iyer. We are interested in developing efficient, fair, robust, and scalable machine learning algorithms by leveraging theoretical and algorithmic insights from various fields such as combinatorial optimization and subset selection, information theory, continuous, and bi-level optimization.

Specificaly, we are interested in making deep model training and inference 5x to 10x faster using subset selection, label-efficient deep learning via active learning and semi-supervised learning, robust deep learning specifically with label noise, imbalance, distribution shift, and out-of-distribution data, fair learning to data biases, and personalized machine learning. Some of our recent theoretical focus areas are submodularity and combinatorial optimization, submodular information measures (like the submodular mutual information and conditional gain measures), coresets, and bi-level optimization.

Recent News

  • One paper from CARAML lab got accepted at ECCV 2022! Congrats Suraj and Saikat!
  • One paper from CARAML lab got accepted at ICML 2022! Congrats Suraj!
  • Our demonstration was accepted at the IEEE Intelligient Vehicles Symposium, IV 2022. Congrats Suraj!
  • Suraj received the prestigious Jan Van der Ziel Fellowship at UT Dallas!
  • Suraj’s dissertation was awarded the runner-up at the 3-Minute-Thesis (3MT) competition hosted by UT Dallas!
  • One paper form CARAML lab got accepted at CVPR 2022! Congrats Krishnateja!

  • One paper from CARAML lab got accepted at Findings of ACL 2022! Congrats Krishnateja!

  • Two papers from CARAML lab are accepted at AAAI 2022! Congrats Krishnateja and Suraj!

  • Our work on Submodular Information Measures was accepted to Transactions of Information Theory Journal.

  • Prof. Rishabh Iyer is be giving a tutorial at AAAI 2022 on Subset Selection in Machine Learning: Theory, Applications, and Hands-on. Stay tuned for more updates!

  • Three papers from CARAML lab are accepted at NeurIPS 2021! Congrats Krishnateja, Ping, Nathan, and Suraj!

  • We received an NSF Collaborative Medium Grant on Submodular Information Functions with Applications to Machine Learning. Thanks, NSF!

  • We received an gift funding from Adobe for Targeted Subset Selection! Thanks, Adobe!

  • We received an gift funding from Google on Continuous Learning! Thanks, Google!

  • Together with Prof. Abir De, Prof. Ganesh Ramakrishnan, and Prof. Jeff Bilmes, Prof. Rishabh Iyer is co-organizing a workshop on Subset Selection in Machine Learning: From Theory to Applications at ICML 2021 on July 24th 2021! Workshop page

  • CARAML Lab is excited to release SubModLib (Github), a submodular optimization toolkit. Credits to Vishal Kaushal for leading this effort.

  • CARAML Lab is excited to release CORDS (Github), a PyTorch-based open-source efficient deep model training and autoML library! Credits to Krishnateja Killamsetty for leading this.

  • CARAML Lab is excited to release DISTIL (Github), a PyTorch-based open-source active learning toolkit for deep learning! Credits to Nathan Beck and Durga Sivasubramanian for leading this.

  • Two papers from CARAML Lab, GRAD-MATCH and SELCON are accepted to ICML 2021! Congrats Krishnateja and Durga!

  • Two papers on rule augmented learning are accepted at Findings of ACL 2021 (one short and one long). Congrats Krishnateja!

  • Prof. Rishabh Iyer is invited as a Speaker at the London Symposium on Information Theory (LSIT) 2021.(Youtube Link to the Talk)

  • Happy to announce that we have released VISIOCITY, a dataset comprising of long videos for video summarization, and more broadly video understanding!

  • Our work on “A Clustering based Selection Framework for Cost Aware and Test-time Feature Elicitation” received Best Paper Honorable Mention at CODS-COMAD 2021! Congrats Srijita and Sriraam!

  • Prof. Rishabh Iyer will be presenting a tutorial on Combinatorial Approaches for Data, Topic and Feature Selection and Summarization at IJCAI 2020 with Ganesh Ramakrishnan (presented a similar one at ECAI 2020 earlier this year).

  • Our paper on Data Subset Selection (GLISTER) is accepted to AAAI 2021! Congrats Krishnateja and Durga!

  • Prof. Rishabh Iyer gave an Invited Talk in the Special Session Deep Learning and Information Theory at SPCOM 2020 (Virtual)

Funding

Our work is made possible by funding from several organizations.