ABAB Pedia

Hansa Srinivasan

Overview

Contents7 sections
Key facts

Studying images and sound at MIT

A computational photography course drew her to computer vision. She took a machine learning course and did background research over the summer. As an MITMassachusetts Institute of TechnologyThe Massachusetts Institute of Technology (MIT) is a private research university in Cambridge, Massachusetts, United States. It received its charter from Massachusetts in 1861, admitted its first students in 1865 and moved from Boston to its present campus in 1916. As of October 2025 it had 11,816 students, and it counts 106 Nobel laureates among members of its community.Open the full entry EECS–DENSO Undergraduate Research and Innovation Scholar in 2015–2016, she worked with William T. Freeman on predicting how objects deform on impact. She used videos of a drumstick scraping or striking objects to locate and characterize deformations and added sound to test whether it improved predictions.⁠[4]

Studying the long-term effects of decisions

Writing as a Google Research software engineer, she introduced ML-fairness-gym. She explained how a decision changes the data available for the next decision, as lending can alter an applicant’s future opportunities. The tool uses simulations to compare the effects of repeated decisions alongside conventional error measures on static data.⁠[2]

Studying representation in text and images

She co-authored an EMNLP 2023 paper examining whether critiques and voting among model responses could improve demographic representation in language-model output. At the 2024 ACM Conference on Fairness, Accountability, and Transparency, she and colleagues reported a measure of diversity in images of people. They calibrated it against human judgments to help retrieval systems surface differences such as cultural attire and disability.⁠[5][3]

Discussing character consistency in Nano Banana

She appeared on Training Data with Google DeepMind product manager Nicole BrichtovaNicole BrichtovaNicole Brichtova, whose name is also written Nicole Brichtová in Czech, is a product leader at Google DeepMind. She studied international economics at Georgetown, earned a Master of Management Studies at Duke and went on to lead work on image generation and editing in Gemini.Open the full entry. The team had previously worked on the Imagen family of image generators. In testing editing, they used photos of people they knew, whose facial differences were easier to notice than those of strangers. Srinivasan described users combining different video tools with an image model to reuse a character across scenes, and explained why the team relied on people to judge subtle, subjective visual differences. Her parents, aunts and uncles, and friends had all tried chatbots. What especially interested her was the emotional response she saw when they encountered visual work.⁠[1][6]

Related Organizations, Websites and Public Accounts

Google Research: Software engineer. Official website: https://research.google/ (opens in a new window).⁠[2]

Google DeepMind: Research engineer. Official website: https://deepmind.google/ (opens in a new window).⁠[1]

Sources6 sources

  1. How Google’s Nano Banana Achieved Breakthrough Character Consistency (opens in a new window)sequoiacap.com↑
  2. ML-fairness-gym: A Tool for Exploring Long-Term Impacts of Machine Learning Systems (opens in a new window)research.google↑
  3. Generalized People Diversity: Learning a Human Perception-Aligned Diversity Representation for People Images (opens in a new window)facctconference.org↑
  4. Hansa Srinivasan (opens in a new window)superurop.mit.edu↑
  5. Improving Diversity of Demographic Representation in Large Language Models via Collective-Critiques and Self-Voting (opens in a new window)aclanthology.org↑
  6. How Google’s Nano Banana Achieved Breakthrough Character Consistency (opens in a new window)youtube.com↑