Vikram Ramaswamy

Timeline

1. 2022

   HIVE and human evaluation

   In 2022, he coauthors HIVE, a human-evaluation framework that tests visual explanations in AI-assisted decisions. The team evaluates four computer-vision interpretation methods in human studies with nearly 1,000 participants.

2. 2023

   A PhD on vision bias

   After bachelor's and master's degrees at IIT Madras, he completes a computer science PhD at Princeton in 2023 under adviser Olga Russakovsky. His 2023 final public oral, “Tackling bias within Computer Vision Models,” examines balanced image datasets and methods for interpreting visual models.

3. 2023

   Testing concept explanations

   In 2023, he coauthors “Overlooked Factors in Concept-Based Explanations,” finding that the probe dataset can change an explanation and that explanations with more than 32 concepts are less useful to people.

4. 2023–2024

   GeoDE spans six regions

   In 2023, he coauthors GeoDE, an object-recognition dataset with 61,940 images from 40 classes and six world regions. A Princeton Engineering feature in 2024 identifies him as the project's lead author and a lecturer in computer science.

5. 2025

   Attention IoU for bias

   In 2025, he coauthors Attention IoU, a metric that uses attention maps to reveal bias inside visual models. Its analysis of CelebA uncovers attribute correlations beyond accuracy disparities.

6. 2026

   Reward model bias

   In 2026, he coauthors an audit of text-to-image reward models that finds they encode demographic biases. Reward-guided optimization can reinforce gender and racial stereotypes and reduce demographic diversity in generated images.

7. As of October 2026

   Lecturer at Princeton

   As of October 2026, he is a lecturer in Princeton's Computer Science department. In Fall 2026, he supervises grader and lab teaching-assistant positions for COS 324, Introduction to Machine Learning. His research interests include fairness and interpretability in visual systems.