Tamar Rott Shaham

I am a postdoctoral fellow with Antonio Torralba at CSAIL, MIT. I did my PhD at the Electrical & Computer Engineering faculty of the Technion where I worked with Tomer Michaeli.

My research interests focus on understanding, controlling, and enhancing AI models. I develop interpretability tools that automatically discover and explain the internal operations of machine-learning models and use gained insights to control model behavior, enhance performance, and prevent undesired outcomes.

Office:      45-733D

Office hours: I'm hosting pro bono office hours, here are more details.

Email  /  Google Scholar  /  Twitter  /  Github

profile photo

News

[April 2024] We introduce MAIA, a Multimodal Automated Interpretability Agent that solves interpretability tasks by iteratively designing experiments on other AI systems.

[Feb 2024] Why does fine-tuning LLMs work so well? Our ICLR'24 paper reveals it's not about introducing new mechanisms but enhancing the existing ones!

[Jan 2024] Check out our new paper: A Vision Check-up for Language Models, to be presented at CVPR'24!

[Dec 2023] MIT News covers our recent work on Automated Interpretability Agents (AIAs)

[Sep 2023] FIND a new benchmark for evaluating automated interpretability methods, was accepted to NeurIPS!

[June 2023] Internal Diverse Image Completion is presented by Noa at CVPR

[Dec 2022] SinGAN follow up paper BlendGAN for smooth image blending across time and space

[Sep 2022] I joined Antonio Torralba's lab as a postdoc

[Oct 2021] Our paper GANs Spatial Control via Inference-Time Adaptive Normalization was accepted to WACV

[Sep 2021] Our works on single audio generative model and deep self-dissimilarities were accepted to NeurIPS

[Jun 2021] Check out our new paper about learning a generative model from a single short audio source

[Mar 2021] Our ASAPNet paper was accepted to CVPR

[Oct 2020] I participated in the IMVC2020's GANs panel

[Aug 2020] We are organizing the Deep Internal Learning (DIL) workshop in conjunction with ECCV 2020 (check out my joint keynote with Tomer)

[Jan 2020] I received the Adobe Research Fellowship

[Jan 2020] I gave a talk about SinGAN at the Israeli Computer Vision day

[Nov 2019] SinGAN won ICCV’19 Best Paper Award (Marr Prize)!

[Aug 2019] I participated in the Google Student Retreat at London, for Women Techmakers Scholars (now called Generation Google Scholarship), and met an amazing group of women from all over Europe, the Middle East and Africa

[July 2019] I am interning at Adobe Research Seattle for summer 2019, working with Eli Shechtman, Michaël Gharbi, and Richard Zhang


Publications

A Multimodal Automated Interpretability Agent
Tamar Rott Shaham*, Sarah Schwettmann*, Franklin Wang, Franklin Wang, Achyuta Rajaram, Evan Hernandez, Jacob Andreas, Antonio Torralba,
paper / web / code / experiment browser
Fine-Tuning Enhances Existing Mechanisms: A Case Study on Entity Tracking
Nikhil Prakash, Tamar Rott Shaham, Tal Haklay, Yonatan Belinkov, David Bau
ICLR, 2024
paper / web / code

A Vision Check-up for Language Models
Pratyusha Sharma*, Tamar Rott Shaham*, Manel Baradad, Stephanie Fu, Adrián Rodríguez-Muñoz, Shivam Duggal, Phillip Isola, Antonio Torralba
CVPR, 2024
Highlight paper
paper / web

FIND: A Function Interpretation Benchmark for Evaluating Interpretability Methods
Sarah Schwettmann*, Tamar Rott Shaham*, Joanna Materzynska, Neil Chowdhury, Shuang Li, Jacob Andreas, David Bau, Antonio Torralba,
NeurIPS, 2023
paper / web / code
Discovering Variable Binding Circuitry with Desiderata
Xander Davies*, Max Nadeau*, Nikhil Prakash*, Tamar Rott Shaham, David Bau,
ICML Workshop Deployable Generative AI, 2023
paper / web / code
Automation in Interior Space Planning: Utilizing Conditional Generative Adversarial Network Models to Create Furniture Layouts
Hanan Tanasra, Tamar Rott Shaham, Tomer Michaeli Guy Austern, Shany Barath,
Buildings, 2023
paper
Internal Diverse Image Completion
Noa Alkobi, Tamar Rott Shaham, Tomer Michaeli
CVPR, Generative Models for Computer Vision Workshop, 2023
paper
BlendGAN: Learning and Blending the Internal Distributions of Single Images by Spatial Image-Identity Conditioning
Idan Kligvasser, Tamar Rott Shaham, Noa Alkobi, Tomer Michaeli
Arxiv, 2022
paper
GANs Spatial Control via Inference-Time Adaptive Normalization
Karin Jakoel*, Liron Efraim*, Tamar Rott Shaham
WACV, 2022
paper / video / supplementals
Catch-A-Waveform: Learning to Generate Audio from a Single Short Example
Gal Greshler, Tamar Rott Shaham, Tomer Michaeli
NeurIPS, 2021
paper / web / code / supplementals
Deep Self-Dissimilarities as Powerful Visual Fingerprints
Idan Kligvasser, Tamar Rott Shaham, Yuval Bahat, Tomer Michaeli
NeurIPS, 2021
Spotlight presentation
paper / supplementals
Spatially-Adaptive Pixelwise Networks for Fast Image Translation
Tamar Rott Shaham, Michaël Gharbi, Richard Zhang, Eli Shechtman, Tomer Michaeli
CVPR, 2021
project page / arXiv
SinGAN: Learning a generative model from a single natural image
Tamar Rott Shaham, Tali Dekel, Tomer Michaeli
ICCV, 2019 
Best Paper Award (Marr Prize)
project page / arXiv / CVF / supp / code / ICCV talk / Israel Vision Day talk (recommended)
Deformation Aware image Compression
Tamar Rott Shaham, Tomer Michaeli
CVPR, 2018 
Spotlight presentation
project page / paper / code / spotlight
xUnit: Learning a Spatial Activation Function for Efficient Image Restoration
Idan Kligvasser Tamar Rott Shaham, Tomer Michaeli
CVPR, 2018 
Spotlight presentation
paper / code / spotlight (by Idan)

Visualizing Image Priors
Tamar Rott Shaham, Tomer Michaeli
ECCV, 2016 
project page / paper / poster
Edge Preserving Multi-Modal Registration Based On Gradient Intensity Self-Similarity
Tamar Rott Shaham, Dorin Shriki, Tamir Bendory
IEEEI, 2014 
paper

Pro bono office hour

Inspired by Krishna Murthy and Wei-Chiu Ma, I dedicate 1-2 hours each week to providing guidance and mentorship to students from underrepresented groups, or to anyone who needs it. Specifically, if you're searching for a postdoc position, I'd be happy to share insights from my experience. Please fill out this form to contact me.


Wonder about the meaning of ?



This website is based on Jon Barron's template (source code)