Tata Institute of Fundamental Research
Fully Differentiable Networks for Retrieval of Graph-Structured Objects
STCS Seminar
Speaker:
Abir De (IIT Bombay)
Organiser:
Raghuvansh Saxena
Date:
Thursday, 8 Oct 2026, 16:00 to 17:00
Venue:
A-201 (STCS Seminar Room)
(Scan to add to calendar)
Abstract:
Given a query graph and a database of corpus graphs, a graph retrieval system aims to deliver the most relevant corpus graphs. Graph retrieval has a wide variety of applications, e.g., molecular fingerprint detection, circuit design, software analysis, and question answering. In such applications, a corpus graph is relevant to a query graph, if the query graph is (perfectly or approximately) similar to the corpus graph. In this talk, we will discuss a family of neural networks that can approximate similarity between graphs under different similarity notions. We consider graph retrieval based on subgraph matching. We discuss ISONET, a novel interpretable neural edge alignment formulation, which can better learn the edge-consistent mapping necessary for subgraph matching. It incorporates a new scoring mechanism that enforces an asymmetric relevance score tailored to subgraph matching. We also propose both late and early interaction neural formulations for this task. Late interaction methods compute dense representations for the query and corpus graph separately, and compare these representations using simple similarity functions at the last stage, leading to highly scalable systems. Early interaction methods combine information from both graphs right from the input stages, and are usually considerably more accurate, but slower. They are both based on a continuous relaxation of a node alignment matrix between query and corpus nodes.
Bio: Abir De is Bhide Family Chair Associate Professor at the Department of Computer Science and Engineering in IIT Bombay. Before that, he did his postdoctoral research at Max Planck Institute for Software Systems at Kaiserslautern, Germany. He received his PhD from Department of Computer Science and Engineering, IIT Kharagpur. His PhD work was supported by Google India PhD Fellowship 2013. Prior to this he did his BTech in Electrical Engineering and MTech in Control Systems Engineering both from IIT Kharagpur. His research interests are learning on structured objects with focus on graphs and sets. He is a recipient of Indian National Academy of Engineering Young Engineer Award 2021 and Krithi Ramamritham Award for Creative Research 2020 at IIT Bombay. He is the recipient of the 2026 Young Alumnus Achiever Award from IIT Kharagpur
| Speaker: | Abir De (IIT Bombay) |
| Organiser: | Raghuvansh Saxena |
| Date: | Thursday, 8 Oct 2026, 16:00 to 17:00 |
| Venue: | A-201 (STCS Seminar Room) |
(Scan to add to calendar)
Given a query graph and a database of corpus graphs, a graph retrieval system aims to deliver the most relevant corpus graphs. Graph retrieval has a wide variety of applications, e.g., molecular fingerprint detection, circuit design, software analysis, and question answering. In such applications, a corpus graph is relevant to a query graph, if the query graph is (perfectly or approximately) similar to the corpus graph. In this talk, we will discuss a family of neural networks that can approximate similarity between graphs under different similarity notions. We consider graph retrieval based on subgraph matching. We discuss ISONET, a novel interpretable neural edge alignment formulation, which can better learn the edge-consistent mapping necessary for subgraph matching. It incorporates a new scoring mechanism that enforces an asymmetric relevance score tailored to subgraph matching. We also propose both late and early interaction neural formulations for this task. Late interaction methods compute dense representations for the query and corpus graph separately, and compare these representations using simple similarity functions at the last stage, leading to highly scalable systems. Early interaction methods combine information from both graphs right from the input stages, and are usually considerably more accurate, but slower. They are both based on a continuous relaxation of a node alignment matrix between query and corpus nodes.
Bio: Abir De is Bhide Family Chair Associate Professor at the Department of Computer Science and Engineering in IIT Bombay. Before that, he did his postdoctoral research at Max Planck Institute for Software Systems at Kaiserslautern, Germany. He received his PhD from Department of Computer Science and Engineering, IIT Kharagpur. His PhD work was supported by Google India PhD Fellowship 2013. Prior to this he did his BTech in Electrical Engineering and MTech in Control Systems Engineering both from IIT Kharagpur. His research interests are learning on structured objects with focus on graphs and sets. He is a recipient of Indian National Academy of Engineering Young Engineer Award 2021 and Krithi Ramamritham Award for Creative Research 2020 at IIT Bombay. He is the recipient of the 2026 Young Alumnus Achiever Award from IIT Kharagpur