Mehran Shakerinava
Mehran Shakerinava
Email: mehran (at) neograd.ai / mehranshakerinava (at) gmail.com
Building Neograd, a new AI lab in Montreal
PhD in Computer Science from McGill and Mila, 2021–2026
CV / X
I am building Neograd, an applied AI research lab developing general-purpose AI systems that
process streams of sensory information and respond in real time.
I completed my PhD in Computer Science at McGill University and Mila in August 2026, supervised by Prof. Siamak Ravanbakhsh.
My doctoral research lay at the intersection of machine learning, optimization, and geometry, with a focus on theoretical foundations
and practical applications of symmetry in learning algorithms.
Prior to McGill, I received a Bachelor's degree in computer engineering from Sharif University of Technology in Iran.
If you'd like to discuss research, potential collaboration, or just connect, feel free to email me.
Research Interests
- Design and analysis of machine learning algorithms
- Symmetry and geometry in machine learning and optimization
- Theoretical foundations of machine learning
- Utility theory and reinforcement learning
Publications
-
The Expressive Limits of Diagonal SSMs for State-Tracking
Mehran Shakerinava*, Behnoush Khavari*, Siamak Ravanbakhsh, Sarath Chandar
ICLR 2026
[openreview]
[arXiv]
-
Beyond Scalar Rewards: An Axiomatic Framework for Lexicographic MDPs
Mehran Shakerinava, Siamak Ravanbakhsh, Adam Oberman
NeurIPS 2025 (Spotlight)
[arXiv]
-
Parity Requires Unified Input Dependence and Negative Eigenvalues in SSMs
Behnoush Khavari*, Mehran Shakerinava*, Jayesh Khullar*, Jerry Huang*, François Rivest, Siamak Ravanbakhsh, Sarath Chandar
ICML 2025 Workshop on Methods and Opportunities at Small Scale
[paper]
[arXiv]
[code]
-
Weight-Sharing Regularization
Mehran Shakerinava*, Motahareh Sohrabi*, Siamak Ravanbakhsh, Simon Lacoste-Julien
AISTATS 2024
[paper]
[arXiv]
[code]
-
Structuring Representations Using Geometric Invariants
Mehran Shakerinava*, Arnab K. Mondal*, Siamak Ravanbakhsh
NeurIPS 2022
[paper]
-
Utility Theory for Sequential Decision Making
Mehran Shakerinava, Siamak Ravanbakhsh
ICML 2022
[paper]
[arXiv]
-
Equivariant Networks for Pixelized Spheres
Mehran Shakerinava, Siamak Ravanbakhsh
ICML 2021
[paper]
[arXiv]
[code]
Older Publications
-
Harnessing Pairwise-Correlating Data Prefetching With Runahead Metadata
Fatemeh Golshan, Mohammad Bakhshalipour, Mehran Shakerinava, Ali Ansari, Pejman Lotfi-Kamran, Hamid Sarbazi-Azad
Computer Architecture Letters (CAL) 2020
[paper]
-
Bingo Spatial Data Prefetcher
Mohammad Bakhshalipour, Mehran Shakerinava, Pejman Lotfi-Kamran, Hamid Sarbazi-Azad
International Symposium on High-Performance Computer Architecture (HPCA) 2019
[paper]
[code]
Preprints
-
Newman's Theorem via Carathéodory
Yaqiao Li, Ali Mohammad Lavasani, Mehran Shakerinava
arXiv preprint 2024
[arXiv]
This page has been accessed
times since Dec, 2024.