Pawan Kumar

About

I did my PhD and postdocs in Europe, which gave me strong foundations in research. For the past 5 years I have been a full-time data science researcher at IIIT Hyderabad, India, with more than 10 years of hands-on research experience in machine learning, optimization, high performance computing, and applications to the sciences.

I have guided 8 thesis students and currently guide 10. I have obtained industrial grants from Microsoft, Qualcomm and TATA. I like working on all aspects of machine learning: theoretical, algorithmic and applied, and publish in venues such as NeurIPS, TMLR, EMNLP, AAMAS, SIAM J. of Optimization, Numerische Mathematik, SDM, ECML and WACV.

Grants and awards

• Best Paper Award, IJCNN 2023
• Marie Curie Actions Fellowship
• Qualcomm Faculty Award
• Microsoft Academic Partnership Grant
• KCIS Grant (TATA group): ML for biosensors
• GATE percentile 96.6 (rank 88), 2007

Professional service

• Program committee, SIGMETRICS 2027
• Program committee, AAAI 2027
• Program committee, AAAI 2026

Publications

44 papers and one book, newest first within each theme. P. Kumar marks my name.

Book1

1. Mathematics of Generative Modelling. Open source P. Kumar, et al. AI Wranglers Foundation. Open source online book with LaTeX source, an HTML reader and teaching material.
   2024–26

Safety and Alignment of AI8

1. Auditing and Suppressing Targeted Reconstruction in Diffusion Language Models. D. Chaudhary, V. S. Dhara, Preeti, P. Kumar. CODS.
   2027
2. PID-DiffRL: Safe Offline Reinforcement Learning as Noisy Delayed Feedback Control. A. Kushwaha, Preeti, K. Ravish, P. Kumar, A. Mahajan. Transactions on Machine Learning Research (TMLR).
   2026
3. Test-Time Visual Concept Anchoring via Entropic Optimal Transport. P. Kumar. CVPR, VisCon Workshop.
   2026
4. ConceptOT: Fine-Grained Vision-Language Alignment via Low-Rank Unbalanced Optimal Transport. P. Kumar. CVPR, VisCon Workshop.
   2026
5. MA-SafeDiffuser: Safe Multi-Agent Planning with Diffusion Probabilistic Models. K. Ravish, et al. AAMAS.
   2026
6. Safe Offline Reinforcement Learning using Diffusion Policies. A. Kushwaha, et al. AAMAS.
   2026
7. Alignment and Safety of Diffusion Models via Reinforcement Learning and Reward Modeling: A Survey. P. Lamba, et al. arXiv, under submission.
   2025
8. A Survey of Safe Reinforcement Learning and Constrained MDPs. A. Kushwaha, et al. arXiv, under submission.
   2025

Large Language Models5

1. Beyond Current Confidence: Trajectory Memory for Selective Revision in Masked Generation. Preeti, K. Ravish, A. Kushwaha, P. Kumar. CODS.
   2027
2. TAP: Policy Interfaces for Diffusion Language Models. Preeti, A. Kushwaha, K. Ravish, P. Kumar. EMNLP (Main Conference).
   2026
3. Quantizing Masked Diffusion Language Models: What Compresses, What Breaks, and What Transfers. N. Mishra, P. Kumar. EMNLP (Industry Track).
   2026
4. Speculative Refinement: A Hybrid Autoregressive Diffusion Decoding Strategy and Its Behavior Across Benchmarks. A. Gupta, N. Mishra, K. Trivedi, P. Kumar. ACL, GEM Workshop.
   2026
5. Hierarchical Sparse Plus Low Rank Compression of LLM. A. Gupta, et al. CODS.
   2025

Multimodal AI and Computer Vision6

1. Semantically Aware Optimal Transport for Dense Label Transfer. Preeti, et al. ACL, ALVR Workshop.
   2026
2. Fast and Efficient Modern BERT based Text-Conditioned Diffusion Model for Medical Image Segmentation. Siddharth, et al. CVIP.
   2025
3. Nonnegative Low-Rank Tensor Completion via Dual Formulation with Applications to Image and Video Completion. T. Sinha, et al. WACV.
   2022
4. Structured Low-Rank Tensor Learning. J. Naram, T. Sinha, P. Kumar. NeurIPS Workshop.
   2021
5. A Deflation Based Fast and Robust Preconditioner for Bundle Adjustment. S. Das, et al. WACV.
   2021
6. Multigrid Preconditioned Solver for Bundle Adjustment. S. Katyan, et al. WACV.
   2020

Machine Learning for Sciences: Biosciences, Circuit Design5

1. A Label-Free Sensing of Creatinine using Radio Frequency-Driven Lab-on-Chip (LOC) System. S. Sinha, et al. Engineering Express Journal, IOP Science.
   2024
2. A Label-Free Low-Cost Radio Frequency Driven Noninvasive Lab-on-Chip System for Creatinine Detection. S. Sinha, et al. ICST.
   2024
3. Qualitative Data Augmentation for Performance Prediction in VLSI Circuits. P. Srivastava, P. Kumar, Z. Abbas. Integration, the VLSI Journal, Elsevier.
   2024
4. Data Augmentation for Performance Prediction in VLSI Circuits. P. Srivastava, P. Kumar, Z. Abbas. ISCAS, Singapore.
   2024
5. Enhancing ML Model Accuracy for Digital VLSI Circuits using Diffusion Models. P. Srivastava, Z. Abbas, P. Kumar. MLSys Workshop, NeurIPS.
   2021

Reinforcement Learning, Data Mining and Machine Learning12

1. Ambiguous or Wrong? Semantic Optimal Transport and Nested Bilevel Triage for Dataset Auditing. A. Mishra, V. S. Dhara, P. Kumar. CODS.
   2027
2. Discrete Diffusion Language Models Are Training-Free Multi-Label Classifiers. P. Kumar. SIAM International Conference on Data Mining (SDM).
   2026
3. Stable 3D Tokens for Foundation-Model Robot Planning. P. Kumar. RSS Workshop on Robot Planning in the Era of Foundation Models.
   2026
4. A Planner-Centered Study of Robust Latent World Models. P. Kumar. RSS Workshop on Robot Planning in the Era of Foundation Models.
   2026
5. alphaElimination: Using Deep Reinforcement Learning for Sparse Gaussian Elimination. A. Dasgupta, P. Kumar. ECML.
   2023
6. marl-jax: Multi-Agent Reinforcement Learning Framework for Social Generalization. K. Mehta, A. Mahajan, P. Kumar. ECML.
   2023
7. Effects of Spectral Normalization in Multi-Agent Reinforcement Learning. K. Mehta, A. Mahajan, P. Kumar. IJCNN.
   2023
8. Light Weight Deep Extreme Multilabel Classification. Best Paper Award U. Mishra, et al. IJCNN.
   2023
9. A Riemannian Approach to Extreme Classification Problems. J. Naram, T. Sinha, P. Kumar. CODS-COMAD.
   2022
10. Hybrid Tokenization and Datasets for Solving Mathematics and Science Problems Using Transformers. P. Mandlecha, et al. SDM.
    2022
11. SCIMAT: Science and Mathematics Dataset. S. Chatakonda, et al. DCAI Workshop, NeurIPS.
    2021
12. DXML: Distributed Extreme Multilabel Classification. P. Kumar. BDA, Springer.
    2021

Optimization Methods and Preconditioners5

1. Riemannian Hamiltonian Methods for Min-Max Optimization on Manifolds. A. Han, et al. SIAM Journal on Optimization (SIOPT).
   2023
2. Angle Based Dynamic Learning Rate for Gradient Descent. N. Mishra, P. Kumar. IJCNN.
   2023
3. Adaptive Consensus Optimization Method for GANs. S. Danisetty, S. Mylaram, P. Kumar. IJCNN.
   2023
4. Fast Preconditioned Solver for Truncated Saddle Point Problem in Nonsmooth Cahn-Hilliard Model. Book chapter.
   2016
5. Modified Tangential Frequency Filtering Decomposition and its Fourier Analysis. Numerische Mathematik.
   2010

High Performance and Scientific Computing3

1. Efficient FPGA Implementation of Conjugate Gradient Methods for Laplacian System using HLS. FPGA.
   2019
2. Multilevel Communication Optimal Least Squares Solver. ICCS, vol. 51.
   2015
3. High Performance Solvers for Implicit Particle in Cell Simulation. ICCS, vol. 18.
   2013

Teaching

Selected talks and workshops

• Fast and Efficient Modern BERT based Text-Conditioned Diffusion Model for Medical Image Segmentation, CVIP, IIT Ropar, December 2025
• Hierarchical Sparse Plus Low Rank Compression of LLM, CODS, IISER Pune, December 2025
• Effects of Spectral Normalization in MARL, Gold Coast, Australia, 2023
• Winter school on Hierarchical Matrices, Leipzig, Germany, 2014
• Research visit (Prof. M. Gander), University of Geneva, 2014
• Parallel aggregation based algebraic multigrid, Domain Decomposition, France, 2012

Computer skills