Centres Of Excellence

To focus on new and emerging areas of research and education, Centres of Excellence have been established within the Institute. These ‘virtual' centres draw on resources from its stakeholders, and interact with them to enhance core competencies

Read More >>

Faculty

Faculty members at IIMB generate knowledge through cutting-edge research in all functional areas of management that would benefit public and private sector companies, and government and society in general.

Read More >>

IIMB Management Review

Journal of Indian Institute of Management Bangalore

IIM Bangalore offers Degree-Granting Programmes, a Diploma Programme, Certificate Programmes and Executive Education Programmes and specialised courses in areas such as entrepreneurship and public policy.

Read More >>

About IIMB

The Indian Institute of Management Bangalore (IIMB) believes in building leaders through holistic, transformative and innovative education

Read More >>

Research & Publications Office to host seminar ‘Heterogeneous Statistical Transfer Learning’, on 16 June

Seminar by Prof. Subhadeep Paul, The Ohio State University

10 June, 2026, Bengaluru: The Office of Research and Publications (R&P) will host a seminar on ‘Heterogeneous Statistical Transfer Learning’, to be led by Prof. Subhadeep Paul from the Statistics department, The Ohio State University, at 2.00 PM on 16 June 2026, in Classroom P-22.

Abstract:
The seminar by Prof. Paul, in the first part, will examine the problem of transfer learning under heterogeneity from a source domain to a new target domain in high-dimensional regression settings with differing feature sets. While most transfer learning approaches assume that source and target domains share the same feature space, real-world applications often involve missing or unobservable variables in data-constrained target environments. To address this challenge, he proposes a framework that first learns a feature map between missing and observed features using source-domain data and then imputes the missing features in the target domain. The approach subsequently applies a two-step transfer learning procedure for penalized regression.

The study develops theoretical guarantees for estimation and prediction errors under both linear and nonparametric feature-mapping settings, providing insights into how model complexity, sample size, feature-map quality, and cross-domain differences affect performance.

In the second part of the talk, Prof. Paul will discuss a transfer learning framework for nonparametric regression using random forests. The method assumes that source and target regression functions differ only across a small number of features and employs a residual-learning strategy using Centered Random Forests (CRF). The resulting theoretical analysis demonstrates the advantages of transfer learning in random forests and explains why shallower trees in the target-domain residual forest can serve as an effective form of implicit regularization.

Speaker Profile:

Subhadeep Paul is an Associate Professor in the Department of Statistics at The Ohio State University. He is also a faculty fellow and previously served as a co-director of the foundations of data science and AI at the Translational Data Analytics Institute at Ohio State. He received his PhD in Statistics from the University of Illinois at Urbana-Champaign in 2017. His research focuses on statistical analysis of complex network-linked data and transfer and federated statistical learning. His research has been funded by two NSF grants for the algorithms of threat detection and the mathematics of digital twins programs.

Webpage Link: https://stat.osu.edu/people/paul.963

Add to Calendar 2026-06-15 05:30:00 2026-06-11 10:33:19 Research & Publications Office to host seminar ‘Heterogeneous Statistical Transfer Learning’, on 16 June Seminar by Prof. Subhadeep Paul, The Ohio State University 10 June, 2026, Bengaluru: The Office of Research and Publications (R&P) will host a seminar on ‘Heterogeneous Statistical Transfer Learning’, to be led by Prof. Subhadeep Paul from the Statistics department, The Ohio State University, at 2.00 PM on 16 June 2026, in Classroom P-22. Abstract: The seminar by Prof. Paul, in the first part, will examine the problem of transfer learning under heterogeneity from a source domain to a new target domain in high-dimensional regression settings with differing feature sets. While most transfer learning approaches assume that source and target domains share the same feature space, real-world applications often involve missing or unobservable variables in data-constrained target environments. To address this challenge, he proposes a framework that first learns a feature map between missing and observed features using source-domain data and then imputes the missing features in the target domain. The approach subsequently applies a two-step transfer learning procedure for penalized regression. The study develops theoretical guarantees for estimation and prediction errors under both linear and nonparametric feature-mapping settings, providing insights into how model complexity, sample size, feature-map quality, and cross-domain differences affect performance. In the second part of the talk, Prof. Paul will discuss a transfer learning framework for nonparametric regression using random forests. The method assumes that source and target regression functions differ only across a small number of features and employs a residual-learning strategy using Centered Random Forests (CRF). The resulting theoretical analysis demonstrates the advantages of transfer learning in random forests and explains why shallower trees in the target-domain residual forest can serve as an effective form of implicit regularization. Speaker Profile: Subhadeep Paul is an Associate Professor in the Department of Statistics at The Ohio State University. He is also a faculty fellow and previously served as a co-director of the foundations of data science and AI at the Translational Data Analytics Institute at Ohio State. He received his PhD in Statistics from the University of Illinois at Urbana-Champaign in 2017. His research focuses on statistical analysis of complex network-linked data and transfer and federated statistical learning. His research has been funded by two NSF grants for the algorithms of threat detection and the mathematics of digital twins programs. Webpage Link: https://stat.osu.edu/people/paul.963 IIM Bangalore IIM Bangalore communications@iimb.ac.in Asia/Kolkata public
15 Jun 2026

Research & Publications Office to host seminar ‘Heterogeneous Statistical Transfer Learning’, on 16 June

Add to Calendar 2026-06-15 05:30:00 2026-06-11 10:33:19 Research & Publications Office to host seminar ‘Heterogeneous Statistical Transfer Learning’, on 16 June Seminar by Prof. Subhadeep Paul, The Ohio State University 10 June, 2026, Bengaluru: The Office of Research and Publications (R&P) will host a seminar on ‘Heterogeneous Statistical Transfer Learning’, to be led by Prof. Subhadeep Paul from the Statistics department, The Ohio State University, at 2.00 PM on 16 June 2026, in Classroom P-22. Abstract: The seminar by Prof. Paul, in the first part, will examine the problem of transfer learning under heterogeneity from a source domain to a new target domain in high-dimensional regression settings with differing feature sets. While most transfer learning approaches assume that source and target domains share the same feature space, real-world applications often involve missing or unobservable variables in data-constrained target environments. To address this challenge, he proposes a framework that first learns a feature map between missing and observed features using source-domain data and then imputes the missing features in the target domain. The approach subsequently applies a two-step transfer learning procedure for penalized regression. The study develops theoretical guarantees for estimation and prediction errors under both linear and nonparametric feature-mapping settings, providing insights into how model complexity, sample size, feature-map quality, and cross-domain differences affect performance. In the second part of the talk, Prof. Paul will discuss a transfer learning framework for nonparametric regression using random forests. The method assumes that source and target regression functions differ only across a small number of features and employs a residual-learning strategy using Centered Random Forests (CRF). The resulting theoretical analysis demonstrates the advantages of transfer learning in random forests and explains why shallower trees in the target-domain residual forest can serve as an effective form of implicit regularization. Speaker Profile: Subhadeep Paul is an Associate Professor in the Department of Statistics at The Ohio State University. He is also a faculty fellow and previously served as a co-director of the foundations of data science and AI at the Translational Data Analytics Institute at Ohio State. He received his PhD in Statistics from the University of Illinois at Urbana-Champaign in 2017. His research focuses on statistical analysis of complex network-linked data and transfer and federated statistical learning. His research has been funded by two NSF grants for the algorithms of threat detection and the mathematics of digital twins programs. Webpage Link: https://stat.osu.edu/people/paul.963 IIM Bangalore IIM Bangalore communications@iimb.ac.in Asia/Kolkata public

Seminar by Prof. Subhadeep Paul, The Ohio State University

10 June, 2026, Bengaluru: The Office of Research and Publications (R&P) will host a seminar on ‘Heterogeneous Statistical Transfer Learning’, to be led by Prof. Subhadeep Paul from the Statistics department, The Ohio State University, at 2.00 PM on 16 June 2026, in Classroom P-22.

Abstract:
The seminar by Prof. Paul, in the first part, will examine the problem of transfer learning under heterogeneity from a source domain to a new target domain in high-dimensional regression settings with differing feature sets. While most transfer learning approaches assume that source and target domains share the same feature space, real-world applications often involve missing or unobservable variables in data-constrained target environments. To address this challenge, he proposes a framework that first learns a feature map between missing and observed features using source-domain data and then imputes the missing features in the target domain. The approach subsequently applies a two-step transfer learning procedure for penalized regression.

The study develops theoretical guarantees for estimation and prediction errors under both linear and nonparametric feature-mapping settings, providing insights into how model complexity, sample size, feature-map quality, and cross-domain differences affect performance.

In the second part of the talk, Prof. Paul will discuss a transfer learning framework for nonparametric regression using random forests. The method assumes that source and target regression functions differ only across a small number of features and employs a residual-learning strategy using Centered Random Forests (CRF). The resulting theoretical analysis demonstrates the advantages of transfer learning in random forests and explains why shallower trees in the target-domain residual forest can serve as an effective form of implicit regularization.

Speaker Profile:

Subhadeep Paul is an Associate Professor in the Department of Statistics at The Ohio State University. He is also a faculty fellow and previously served as a co-director of the foundations of data science and AI at the Translational Data Analytics Institute at Ohio State. He received his PhD in Statistics from the University of Illinois at Urbana-Champaign in 2017. His research focuses on statistical analysis of complex network-linked data and transfer and federated statistical learning. His research has been funded by two NSF grants for the algorithms of threat detection and the mathematics of digital twins programs.

Webpage Link: https://stat.osu.edu/people/paul.963