picture of pankaj
Pankaj Rajak
Postdoctoral Appointee
Argonne National Laboratory

Bio

I am a Post Doctoral Research Associate in the Advanced Leadership Computing Facility (ALCF) at Argonne National Laboratory. My current research work is on Deep Learning methods such as reinforcement learning, active learning and generative models for autonomous design of material’s atomic structure with desired physical and mechanical properties. To this end, some of my current projects includes predictive synthesis and design of quantum materials using reinforcement learning, graph-neural network based potential energy function for simulation of material’s dynamics and active learning assisted accelerated search of best thermoelectric 2D materials. Beside this, my research interest also includes meta-learning and current challenges in these models such as problems related to task-memorization in few-shot learning.

I received my Ph.D. in Computational Material Science and MS in Computer Science with Data Science specialization from University of Southern California (USC) under the guidance of Prof. Priya Vashishta, Prof. Aiichiro Nakano and Prof. Rajiv Kalia. During my Ph.D, the focus of dissertation topic was on force field development for materials simulation, large scale reactive molecular dynamics simulation of materials consisting of billions of atoms and development of machine learning models for the analysis of these large scale simulation data, where some of these simulations are performed at blue gene/Q supercomputer at Argonne National Laboratory.

Research Interest: Deep Generative Models, Meta-Learning, Reinforcement Learning, Graph Neural Network, Active Learning, High Performance Computing, Force Field Development and Large Scale Molecular Dynamics Simulation.

News

Mar 2021 Will be giving invited talk on Multi-Task Reinforcement Learning for Autonomous Material Design at APS March Meeting 2021.
Nov 2020 Our paper Quantum Material Synthesis by Reinforcement Learning got accepted for poster presentation at NeurIPS workshop, Machine Learning and the Physical Sciences.
Nov 2020 Our paper Denoising Autoencoders for High-Qubit Quantum Dynamics Simulations on Quantum Computers got accepted for poster presentation at NeurIPS workshop NeurIPS workshop, Machine Learning and the Physical Sciences.
May 2020 Presented at SCADL workshop at IPDPS conference on Neural Network Molecular Dynamics at Scale.
Sep 2019 Gave invited talk on "Generative and Reinforcement Learning for Interfaces and Metamaterial Design" at CINT Annual Meeting organised by ointly operated by Los Alamos and Sandia National Laboratorie.

Education

University of Southern California
Ph.D. in Computational Material Science , Aug. 2011 - July 2018

The dissertation topic was on large scale molecular dynamics (MD) simulation of nano-structured material at atomistic level during events such as fracture, shock and chemical reacation. Beside that my research work also included development of interaction potential for reactive MD simulation using multi-objective genetic algorithm, accelerated discovery of materials with active learning and deep generative models such as variational autoencoder.
University of Southern California
Masters in Computer Science , Aug. 2016 - May 2018

I did masters in Computer Science from USC with specilication in Data Science and Machine Learning.
Standford University
Artifical Intelligence Certificate , Aug. 2019 - Presetnt

Currently, I am doing another Artifical Intelligence Certificate from Stanford for which I have taken classes such as multi-task and meta learning, probabilistic graphical models.
Indian Institute of Technology, Kharagpur
Bachelors in Material Science , August 2006 - May 2011

I completed my Bachelor of Technology in Material Science from IIT Kharagpur in India.

Research

Quantum Material Synthesis using Reinforcement Learning
Designing experimental synthesis condition of materials is a challenging task as it involves decision making process of time-dependent synthesis parameters. Further, these experiments take very long time to perform, and thus its not feasible to use the active learning approaches to directly learn the optimal synthesis condition. Here, we have used Offline Model based Reinforcement Leaning (RL) to propose the optimal synthesis condition of 2D quantum material. After trainng, RL agent not only outperformed the random baseline model by more then 50% by desiging optimal policy of synthesis condtiom of reaction temperature and reactant gas concentration but also provided mechanistic insight of synthesis process itself in terms of synthesis parameters
Nov' 2019 - Jul' 2020
Graph Attention Neural Network for Accelerated Polymer Discovery
Designed a simulation framework to estimate electrical properties of polymers using Molecular Dynamics Simulation. This framework is a high throughput framework capable of handling 1000s of polymer simulations at once. Further, we used Graph Neural Networks to learn various complex features for each polymers to predict the based structures which can give desirable electrical properties.
Jan' 2020 - Present
Equivariant Deep Neural Network Force Field for Molecular Dynamics Simulation
Conventional Energy based Neural Network based force field for molecular dynamics(NNMD) model are expensive to perform large scale and long time simulation as computation of atomic fotces requies derivative of Neural Network (NN) at each time step. We have developed a Equivariant NN based field model that is designed to directly predict atomic forces at density functional theory (DFT) level accuracy.
Nov' 2018 - Dec' 2019
Active Learning for Material Design
Properies such as band gap and thermoelectric behaviour of vertically stacked 2D materials can be tunned by careully choosing the compostion of materials at each layer. However, the state space of material goes exponentially with the number of layers and qunatim mechanical (QM) calculation of each structure is very expensive and time consuming. Thus, brute force or regression based models are not suitable choice to find the optimial structure from this exponentially large search space. Here, we have developed an active learning model based on Bayesian optimization to find the best stacked 2D material structure with desired properties with minimum QM calculation.
Jan' 2017 - Feb' 2018
Deep Generative Model for Phase Transformation in Materials
Learning the mechanism of phase transformation in matwrials is a complex and time consuming process. Here, we have developed a generative model of the phase transformation pathways in MoWSe2 during dyanamic fracture using variational autoencoder(VAE). The trained VAE model correctly identifies transformation pathways connecting the semiconducting (2H) and metallic (1T) phases via novel intermediate structures, which is also observed experimently. Further, the mechanistic insight of transformation provided by VAE is much faster then tradtional molecular dynamics simulation that takes 2-3 days.
Jan' 2018 - July'2018

Project

Meta-Regularization by Enforcing Mutual-Exclusiveness
We propose a regularization technique for meta-learning models that minimizes task-overfitting by maximizing the distance between tasksummary statistics as meta-regularization function during training in memory augumented meta-learning models. Maximizing this proposed regularization function reduces task-overfitting and shows an accuracy boost of ∼ 36% on the Omniglot dataset for 5-way, 1-shot classification problem.
Project, Sep'2020 - Nov'2020
Generative Models for 3D Point Clouds
We developed a generative model for 3D point cloud using variatonal autoencoder where the encoder is represented using 1D convolution layer, latent space using normalizing flow and the decoder with a autoregressive progressive multilayer perceptron.
Project, Oct'2019 - Dec'2019

Selected Publications

  • Full Publication List (38 Articles): Google Scholar
    • 2020
  • Predictive Synthesis of Quantum Materials by Model-Based Reinforcement Learning

    P. Rajak, A. Krishnamoorthy, A. Mishra, R. K. Kalia, A. Nakano and P. Vashishta, under review
    Preprint

  • Neural network molecular dynamics at scale

    P. Rajak, K. Liu, A. Krishnamoorthy, R. K. Kalia, A. Nakano, K. Nomura, S. C. Tiwari and P. Vashishta International Parallel and Distributed Processing Symposium Workshops, 2020
    Conference Paper

    • 2019
  • Structural Phase Transitions in MoWSe2 Monolayer – Molecular Dynamics Simulations and Variational Autoencoder Analysis

    P. Rajak, A. Krishnamoorthy, R. K. Kalia, A. Nakano and P. Vashishta, Phys. Rev. B 100, 014108: 1-7 (2019)
    Paper

  • Defect Healing in Layered Materials: A Machine Learning-Assisted Characterization of MoS2 Crystal-Phases

    S. Hong, K. Nomura, A. Krishnamoorthy, P. Rajak, C. Sheng, R. K. Kalia, A. Nakano and P. Vashishta, J. Phys. Chem. Lett. 10, 2739- 2744 (2019)
    Paper

  • Neural Network Analysis of Dynamic Fracture in a Layered Material

    P. Rajak, R. K. Kalia, A. Nakano and P. Vashishta, MRS Adv. 4, 1109-1117 (2019)
    Conference Paper

  • Graph neural network analysis of layered material phasesGraph neural network analysis of layered material phases

    K. Liu, K. Nomura, R. K. Kalia, A. Nakano, P. Vashishta and P. Rajak, Proceedings of the High Performance Computing Symposium, (2019)
    Conference Paper

    • 2018
  • Active Learning for Accelerated Design of Layered Materials

    L. Bassman*, P. Rajak*, R. K. Kalia, A. Nakano, F. Sha, J. Sun, D. J. Singh, M. Aykol, P. Huck, K. Persson and P. Vashishta, npj Comput. Mater. 4, 74: 1-9 (2018), *equal contribution
    Paper

  • Multiobjective Genetic Training and Uncertainty Quantification of Reactive Force Fields

    A. Mishra, S. Hong, P. Rajak, C. Sheng, K. Nomura, R. K. Kalia, A. Nakano and P. Vashishta, npj Comput. Mater. 4, 42: 1-7 (2018)
    Paper

  • Structural Phase Transformation in Strained Monolayer MoWSe2 alloy

    A Apte*, V Kochat*, P Rajak*, A Krishnamoorthy, P Manimunda, J Hachtel, J. C. Idrobo, A. Nakano, R. K. Kalia, P Vashishta, C. S. Tiwary, and P. M. Ajayan, ACS Nano 12, 3468-3476 (2018), *equal contribution
    Paper

  • Faceting, Grain Growth, and Crack Healing in Alumina

    P. Rajak, R. K. Kalia, A. Nakano and P. Vashishta, ACS Nano 12, 9005-9010 (2018)
    Paper

    • 2017
  • “Gel Phase in Hydrated Calcium Dipicolinate

    P. Rajak, A. Mishra, C. Sheng, S. Tiwari, A. Krishnamoorthy, R. K. Kalia, A. Nakano and P. Vashishta. Appl. Phys. Lett. 111, 213701: 1-5 (2017)
    Paper

    • Awards

    • Viterbi Fellowship for Excellence in Graduate Studies, University of Southern California (2011-2015)

    • Research Article featured on the cover page of nature publishing journal computational materials and Applied Physics Letters

    • Travel scholarship, From Passive to Active: Generative and Reinforcement Learning with Physics,UCLA (Sep, 2019)

    • Best Paper Award, International Conference on High Performance Computing in Asia-Pacific Region (2020)