Farzaneh Heidari

PhD Candidate · Interpretability & Tensor Networks

About Me

I am a PhD candidate at Mila and Université de Montréal, currently visiting RIKEN AIP in Tokyo. My research explores the interpretability of deep learning models using tensor networks and concept steering in diffusion models. I’m also gradually learning about quantum circuit interpretability with curiosity, as it intersects with some of my interests.

I hold a Master's in Computer Science from York University and a Bachelor's in Electrical Engineering from Sharif University of Technology. Outside of research, I enjoy bouldering, playing chess, reading literature, and walking in nature. I’m also part of a sustainability committee, learning about different aspects of sustainability, from environmental to social and institutional, and thinking about how they relate to research practice.

I also co-organize a Tensor Network Reading Group, where we discuss foundational and advanced papers at the intersection of tensors, quantum theory, and machine learning.

LinkedIn · GitHub · 📚 · 📷

Research Projects

Multilinear Concept Steering in Diffusion Models

Replace and control concepts in generated images using multilinear projections in the noise space, preserving all other features.

Tensor Networks for Graph Explainability

Use tensor decompositions to represent and explain small graph-structured data and probe information flow and entanglement.

Tensor Networks for Expressiveness of Graph Models

Compare the expressive power of various graph neural network architectures using tensor network representations and decompositions.