Welcome to Numerical Algorithms Group (NAG) of IIITH: NAG@IIITH

My name is pawan kumar. I am an assistant professor at IIIT-H since approximately 2017.

In my group, we are interested in designing provably fast numerical algorithms for vision (image and video completion, solvers for 3D reconstruction, SLAM) machine learning (optimization methods for deep learning and extreme classification) and scientific computing (PDE Solvers). My experience has been mostly in scientific computing, but recently I have been interested in numerical algorithms for vision, machine learning, and distributed optimization and federated learning (eg with blockchains). They offer new set of challenges for numerical algorithms and analysis.

The fast and robust algorithms for the problems mentioned above are obtained either by important mathematical insight, or via clever algorithm design. More specifically, in machine learning as well as in scientific computing, newer insights via numerical analysis for optimization algorithms, which are at the core of most of these domains, can lead to robust and fast algorithms. On the other hand, a clever use of data structures, and a mapping to parallel architectures (FPGA, GPU, Clusters, etc) for the existing algorithms can lead to algorithms that run fast on modern day hardware. In general, I believe that robust numerical algorithm and high performance algorithm design go hand in hand, and they cant be developed in isolation.

In my group that comprises of bright undergraduate and graduate students, we are trying to explore various approaches for fast and robust numerical algorithms. Generally, two types of students join my group: those who are very good with linear algebra and analysis for numerical algorithms, or those who are exceptional programmers, or have strong interest in learning HPC skills, and applying it to develop algorithms for ML and scientific computing. I am particularly interested in students who want to work in the interface of scientific computing and machine learning. I am also interested in students from non-CS background. If you are one of these, then feel free to write to me to know more, or join my group as an intern. See below for current research topics and publications.

Announcement: Full time MS-by-research, PhD, or Postdoc positions available!

Employment/Education History

• Postdoc, 2014-2016 Department of Mathematics and Computer Science, FU Berlin, Einstein Center of Mathematics, Berlin, Germany.
• Postdoc, 2013-2014 Fraunhofer ITWM, Funded by Marie Curie ERCIM Fellowship.
• Postdoc, 2011-2013 KU Leuven and Intel Exascience Lab, Leuven, Belgium.
• PhD, 2007-2010 INRIA Saclay, Ile-de-France, Saclay, France.
• MS, 2005-2007 IIT, Guwahati, India.

Project Pages (Codes, Papers, Datasets)

• Bundle Adjustment/SLAM Solvers for Bundle Adjustment and SLAM.

• Distributed Optimization and Multiagent RL Scalable Distributed Optimization and multi-agent RL.

• Robust Optimization Solvers for Deep Learning Scalable and Robust Solvers for Deep Learning.

• Efficient Algorithms for Edge or Embedded Devices Algorithms for Edge Computing.

• Optimization on Manifolds Robust Optimization Solvers on Manifolds.

• Automated Solution of Mathematics and Science Solve problems in mathematics and science using transformers..

Selected Publications All Publications (Click Me!)»

Scholar   Erdös #: 3   ORCID  

Ranking Sources: Core, ConferenceRank, Oxford

• SIOPT 2023 Riemannian Hamiltonian methods for min-max optimization on manifolds, Andi Han, Bamdev Mishra, Pratik Jawanpuria, Pawan Kumar, Junbin Gao, SIAM J. of Optimization, Revised, 2023, (Rank A).
• CODS-COMAD 2023 Generalized Structured Low Rank Tensor Learning, J. Naram, T. Sinha, P. Kumar, CODS-COMAD, 2023, (Rank: B).
• SDM 2022 Hybrid Tokenization and Datasets for Solving Mathematics and Science Problems Using Transformers, P. Mandlecha, S. Chatakonda, N. Kollepara, P. Kumar, SDM, 2022, (Rank: A).
• CODS-COMAD 2022 A Riemannian Approach to Extreme Classification Problems, J. Naram, T. Sinha, P. Kumar, CODS-COMAD, 2022, (Rank: B).
• WACV 2022 Nonnegative Low-Rank Tensor Completion via Dual Formulation with Applications to Image and Video Completion, T. Sinha, J. Naram, P. Kumar, WACV, 2022, (Rank: A).
• NeurIPS-W 2021 SCIMAT: Science and Mathematics Dataset, S. Chatakonda, N. Kollepara, P. Kumar, DCAI, NeurIPS-W, 2021, (Rank: A).
• NeurIPS-W 2021 Structured Low-Rank Tensor Learning, J. Naram, T. Sinha, P. K., NeurIPS-W, 2021, (Rank: A).
• SC 2021 A Fast Parameter-Free Preconditioner for Structured Grid Problems (arXiv), Summary (two pages), and Poster, A. Agarwal, S. Kakkar, P. Kumar, Supercomputing 2021, (Rank: A).
• BDA 2021 SCIMAT: An Extensive Dataset and Results with Transformer, S. Chatakonda, N. Kollepara, P. Kumar, BDA 2021, (Rank: National).
• WACV 2021 A Deflation Based Fast and Robust Preconditioner for Bundle Adjustment, Das, Shrutimoy and Katyan, Siddhant and Kumar, Pawan, WACV, pp. 1781-1788, 2021, (Rank: A) (Video).
• WACV 2020 Two-Grid Preconditioned Solver for Bundle Adjustment, S. Katyan and S. Das and P. Kumar, WACV, pp. 3588-3595, 2020, (Rank: A) (Video).
• NCVPRIPG 2020 Domain Decomposition Based Preconditioned Solver for Bundle Adjustment, S. Das, S. Katyan, PK, Computer Vision, Pattern Recognition, Image Processing, and Graphics, Springer Singapore, pp. 64-75, 2020, (Rank: National).
• FPGA 2019 Efficient FPGA Implementation of Conjugate Gradient Methods for Laplacian System Using HLS, S. Rampalli, N. Sehgal, I. Bindlish, T. Tyagi, P. Kumar, Extended Abstract, International Symposium on Field-Programmable Gate Arrays (FPGA '19), pp. 308-309, 2019, (Rank: A).
• BDA 2021 DXML: Distributed Extreme Multilabel Classification (arXiv), Springer (doi), P. Kumar, BDA 2021, (Rank: National).
• Springer 2016 Fast Preconditioned Solver for Truncated Saddle Point Problem in Nonsmooth Cahn-Hilliard Model, Pawan Kumar, Recent Advances in Computational Optimization, Springer, Book Chapter, Studies in Computational Intelligence, Vol 655, 2016. (Impact Factor = 0.86).
• IJCM 2016 On Relaxed Nested Factorization and Combination Preconditioning, P. Kumar, L. Grigori, F. Nataf and Q. Niu, International Journal of Computer Mathematics, Vol. 93, Issue 1, pp. 179-199, 2016, (Impact Factor = 1.9).
• AIP 2016 Preconditioners Based on Approximation of Non-Standard Norms for Phase Separation Applications, Pawan Kumar, AIP Proceedings, Vol. 1738, 480114, 2016.
• ICCS 2015 Multilevel Communication Optimal Least Squares, P. Kumar, ICCS 2015, Vol. 51, pp. 1838-1847, 2015 (Rank: A).
• ISPDC 2014 Multithreaded Direction Preserving Preconditioners, P. Kumar, 2014 IEEE 13th International Symposium on Parallel and Distributed Computing (ISPDC), pp. 148-155, 2014, (Rank: C).
• IJCM 2014 Aggregation Based on Graph Matching and Inexact Coarse Grid Solve for Algebraic Two Grid, Pawan Kumar, International Journal of Computer Mathematics, Volume 91, Issue 5, pp. 1061-1081, 2014. (Impact Factor = 1.9).
• HPCC 2014 Communication Optimal Least Squares Solver, P. Kumar, HPCC 2014, IEEE Computer Society, pp. 316-319, 2014, (Rank: B).
• ICCS 2013 High Performance Solvers for Implicit Particle in Cell Simulation, P. Kumar, et al., ICCS 2013, Vol 18, pp 2396-2405, 2013, (Rank: A).
• PARA 2013 Multi-threaded Nested Filtering Factorization Preconditioner, P. Kumar, K. Meerbergen, and D. Roose, PARA 2012, Applied Parallel and Scientific Computing, Springer, Berlin Heidelberg, pp. 220-234, 2013. (Rank: National, Nordic Countries).
• Numerische Mathematik 2010 Modified Tangential Frequency Filtering Decomposition and its Fourier Analysis, Q. Niu, L. Grigori, P. Kumar, F. Nataf, Numerische Mathematik, Vol. 116, pp. 123-148, 2010. (Impact Factor = 2.7).