SOURAJIT DAS

Hello! I am a Postdoctoral Researcher at Pennsylvania State University, working with Prof. Romit Maulik, where my research focuses on reinforcement learning for dynamical systems and scientific machine learning. My work aims to develop scalable, data-driven learning frameworks for modeling, control, and decision-making in complex physical and engineered systems.

I earned my PhD in Electrical and Systems Engineering from the University of Pennsylvania, where my dissertation explored Graph Neural Networks (GNNs) for multi-agent communication systems. During my PhD, I was a Graduate Research Assistant at AleLab, advised by Prof. Alejandro Ribeiro, and worked on state-augmented routing, communication-aware control, and graph-based decision-making in wireless networks and multi agent systems.

My research has been supported by large interdisciplinary initiatives, including projects sponsored by the Department of Energy (DOE), Intel Science and Technology Center (ISTC) on Wireless Autonomous Systems (WAS), Simons Institute and the Army Research Laboratory (ARL) through the Distributed and Collaborative Intelligent Systems (DCIST) program. Across these efforts, I have developed methods that bridge graph based learning, signal processing, control theory, and wireless communication, with an emphasis on real-world deployment and scalability. My publications list can be found here.

Prior to my work in machine learning and dynamical systems, I was a Graduate Student Fellow at the Singh Center for Nanotechnology, where I designed and fabricated Organic Thin-Film Transistors (OTFTs) and solar cells, gaining hands-on experience in nanofabrication, device characterization, and cleanroom processes.

My broader research interests include deep reinforcement learning for dynamical systems, scientific machine learning, graph signal processing, wireless communication networks, multi-agent systems, and distributed optimization. I enjoy working on problems that combine applications, algorithms, systems, and a bit of theory to back it. I am always open to research collaborations and discussions with people from different aspects of science and engineering.

PUBLICATIONS

Opportunistic routing in communications networks via learnable state-augmented policies

Sourajit Das, Satyajeet Das, and Kirtan Gopal Panda

Frontiers in Communications and Networks, 2026

This paper develops state-augmented graph neural network policies for opportunistic routing in wireless communication networks. By combining learned state representations with local network information, the method selects efficient forwarding actions across changing network conditions.

Paper

State-Augmented Opportunistic Routing in Wireless Communication Systems with Graph Neural Networks

Sourajit Das, Navid NaderiAlizadeh, and Alejandro Ribeiro

ICASSP 2025

This work applies state augmentation and graph neural networks to opportunistic routing in large-scale wireless networks. It learns decentralized relay-selection policies from local information and evaluates them against standard routing approaches.

Paper

Learning State-Augmented Policies for Information Routing in Communication Networks

Sourajit Das, Navid NaderiAlizadeh, and Alejandro Ribeiro

IEEE Transactions on Signal Processing, 2024

This paper formulates information routing as a constrained learning problem with only local observations. A state-augmentation strategy and unsupervised GNN training produce scalable routing policies for large communication networks.

Paper

State-Augmented Information Routing In Communication Systems With Graph Neural Networks

Sourajit Das, Navid NaderiAlizadeh, and Alejandro Ribeiro

ICASSP 2024

This work uses dual variables for flow constraints to augment each node's state and trains a GNN to make feasible routing decisions using local network information. Numerical experiments show improved routing performance over baseline methods.

Paper

Texture classification using combination of LBP and GLRLM features along with KNN and multiclass SVM classification

Sourajit Das and Uma Ranjan Jena

2016 2nd International Conference on Communication, Control & Intelligent Systems (CCIS), pp. 115–119

This paper combines Local Binary Pattern and Gray Level Run Length Matrix features for image-texture classification. It compares KNN and multiclass SVM classifiers and shows that the combined features improve classification performance.

Paper

Fabrication of Organic Thin Film Transistors using Inkjet Printing of PEDOT:PSS

Sourajit Das, Akshaya Venkatakrishnan, Hiromichi Yamamoto, and Pat Watson

Scholarly Commons: Protocols and Reports, Singh Center for Nanotechnology, University of Pennsylvania, 2019

This work fabricates organic thin-film transistors by inkjet-printing PEDOT:PSS and studies how printing conditions affect device characteristics. The fabricated devices demonstrate expected field-effect transistor behavior.

Paper

PREPRINTS

SC3D: Dynamic and Differentiable Causal Discovery for Temporal and Instantaneous Graphs

Sourajit Das, Dibyajyoti Chakraborthy, and Romit Maulik

Neural Networks (under review)

SC3D is a two-stage differentiable framework for discovering lagged and instantaneous causal structure in multivariate time series. It first identifies candidate edges and then refines a sparse, acyclic graph through likelihood optimization.

Paper

OppGNN: Deploying State-Augmented Policies for Opportunistic Routing in Real-Time Wireless Communication Networks

Sourajit Das and Alejandro Ribeiro

US patent filed

This work focuses on deploying state-augmented GNN routing policies in real-time wireless networks, combining decentralized opportunistic forwarding with a physical wireless testbed.

GITHUB

GitHub profile

Comparison of Round Robin scheduling and Control Aware Low Latency Scheduling in ROS

RR.mp4

Calls.mp4

Page updated

Google Sites

Report abuse