Subhajit Dandapat

Subhajit Dandapat

Postdoctoral Research Fellow at the National University of Singapore (NUS), and former PhD researcher at the Tata Institute of Fundamental Research (TIFR), Mumbai.

About me

👋 Hi there!

I am an AI-driven computational scientist with expertise in large-scale statistical data analysis, probabilistic modeling, and high-dimensional inference systems. I build GPU-accelerated machine learning systems for extracting weak signals from noisy, high-dimensional time-series data, reducing probabilistic inference time from days to hours.

My work focuses on applied AI/ML, physics-informed machine learning, and large-scale Bayesian inference, with an emphasis on real-world decision systems, time-series modeling, and uncertainty-aware AI.

🔬 What I work on

  • Build and deploy transformer-based and generative AI models (normalizing flows) for high-dimensional and time-series data
  • Engineer GPU-accelerated ML systems that replace simulation-heavy pipelines (e.g., MCMC) with fast, scalable neural inference
  • Develop uncertainty-aware models for robust decision-making under noise and incomplete data
  • Design end-to-end ML pipelines, from data processing to training, validation, and large-scale inference

These systems are designed for scalability, robustness, and real-world data environments.

🧠 Current Work

I am currently a Postdoctoral Research Fellow at the National University of Singapore, where I work on:

  • Physics-informed transformer architectures for time-series analysis
  • Simulation-based Bayesian inference using generative models (normalizing flows)
  • Large-scale probabilistic ML pipelines on GPU/HPC systems

I also collaborate with IBM Research on foundation-model-style AI systems, focusing on scalable inference, noise characterization, and signal classification.

While my current applications are in scientific data, the underlying systems directly apply to:

  • Physics-informed ML for real-world systems (e.g., forecasting, simulations)
  • Anomaly detection and pattern discovery in noisy, high-dimensional data
  • Scalable probabilistic modeling and uncertainty-aware decision systems
  • Predictive modeling and data-driven optimization
  • Financial time-series modeling and risk forecasting

🌍 Background

I completed my Ph.D. at the
Tata Institute of Fundamental Research (TIFR), Mumbai (2024),
where I developed large-scale Bayesian inference systems for extracting weak signals from noisy observational data, in the context of gravitational-wave data analysis and high-dimensional time-series inference.

I collaborate with major international research collaborations, including in technical leadership roles:

I also completed my BS–MS Dual Degree in Physics from Indian Institute of Science Education and Research, Bhopal (2019), where my Master’s thesis focused on Chiral Anomalies in Quantum Field Theory.

🎯 Areas of Interest

I am interested in applying AI/ML and probabilistic modeling to complex, real-world systems across science and industry, particularly in areas involving high-dimensional data, uncertainty, and large-scale decision-making.

  • AI/ML engineering and applied machine learning for real-world systems
  • Physics-informed ML for complex physical and engineering systems
  • Scalable probabilistic modeling and decision-making under uncertainty
  • Quantitative modeling and financial machine learning

⚡ Core Expertise

  • Machine Learning: Transformers, Generative Models, LLMs, Normalizing Flows
  • Data Science: Time-Series Modeling, Statistical Analysis, Feature Engineering, Model Validation
  • Probabilistic AI: Bayesian Inference, Simulation-Based Inference, Uncertainty Quantification
  • Systems: Python, PyTorch, Scikit-learn, GPU/HPC computing, scalable ML pipelines

📌 Additional Context

My work sits at the intersection of physics-informed machine learning and large-scale Bayesian inference, focused on extracting signal from noisy, high-dimensional time-series data.

During my Ph.D., I developed scalable MCMC-based inference systems, statistical validation frameworks, and contributed to international collaborations on complex inference problems.

At NUS, my work has evolved toward AI-driven systems, including physics-informed transformers and normalizing-flow-based posterior estimators, enabling GPU-accelerated pipelines that reduce multi-day computations to hours.

I have also contributed to AI-for-Science initiatives, including securing significant AI and HPC research funding, and I collaborate with IBM Research on foundation models for physics-informed ML, focused on scalable inference and noise characterization.