[Paper][Code][Poster]

Workshop@ICLR : Data-centric Machine Learning Research, 2023

Paper: In this paper, we propose a controllable framework for data-centric trustworthy AI- VTruST, that allows users to control the trade-offs between the different trustworthiness metrics of the constructed training datasets. We propose a novel online version of the Orthogonal Matching Pursuit (OMP) algorithm for solving this problem

[Paper][Code]

Journal: IEEE Transactions on Artificial Intelligence (IEEE TAI), 2023

Paper: In this paper, we argue that data valuation techniques should be flexible, accurate, robust, and efficient (FARE). We propose a two-phase approach towards achieving these objectives, where the first phase, checkpoint selection, extracts important model checkpoints while training on a related dataset, and the second data valuation and subset selection phase extracts the high-value subsets.

[Paper] [Video] [Code] [Slides] [Poster]

Conference: Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), 2021

Paper: In this work, we study the problem of subset selection in the tasks of autonomous driving and semantic segmentation. We design a constrained optimisation problem that can use either pairwise or pointwise criteria or both for obtaining a subset that can perform at par with that of the whole dataset.

[Paper] [Video] [Code] [Slides] [Poster]

Conference: European Conference on Machine Learning and Knowledge Discovery in Databases (ECML-PKDD), 2021

Paper: In this work, we study the problem of finding high value training data points that can be used in subset selection, finding mislabelled examples and more such tasks. We design a learnable framework for online subset selection that is amenable to differentiable convex paradigm. 

[Paper]

Journal: Pattern Recognition Letters, 2020

Paper: In this work, we study the problem of subset selection in the task of semantic segmentation. We incorporate different pointwise criteria besides existing pairwise criteria in the optimisation problem to obtain an appropriate subset for any task.