Rahim Entezari

I am a PhD studet at Institute for Technical Informatics, TU Graz and Complexity Science Hub, Vienna. I try to understand how/why deep learning works.

Before that, I did my master in AI at IUST and bachelor in computer engineering at AUT, Iran.

I am excited to join Stability as a Research Scientist.

Email  /  CV  /  Google Scholar  /  Twitter  /  Github

profile photo


I'm interested in deep learning, optimization/generalization and sparsity. Much of my research is about understaning deep learning phenomena.

The Role of Pre-training Data in Transfer Learning
Rahim Entezari*, Mitchell Wortsman, Olga Saukh Hanie Sedghi, Ludwig Schmidt,


Despite the success of transfer learning paradigm for various downstream tasks, still a question remains: what data and method should be used for pre-training? We study the effect of the pretraining data distribution on transfer learning in the context of image classification, investigating to what extent the choice of pre-training datasets impacts the downstream task performance.

REPAIR: REnormalizing Permuted Activations for Interpolation Repair
Keller Jordan*, Hanie Sedghi, Olga Saukh, Rahim Entezari, Behnam Neyshabur

Accpeted at ICLR, 2023


In this paper we look into the conjecture of Entezari et al. (2021) which states that if the permutation invariance of neural networks is taken into account, then there is likely no loss barrier to the linear interpolation between SGD solutions. First, we observe that neuron alignment methods alone are insufficient to establish low-barrier linear connectivity between SGD solutions due to a phenomenon we call variance collapse: interpolated deep networks suffer a collapse in the variance of their activations, causing poor performance. Next, we propose REPAIR (REnormalizing Permuted Activations for Interpolation Repair) which mitigates variance collapse by rescaling the preactivations of such interpolated networks.

How well do contrastively trained models transfer?
M. Moein Shariatnia*, Rahim Entezari*, Mitchell Wortsman*, Olga Saukh Ludwig Schmidt,
ICML Pre-training: Perspectives, Pitfalls, and Paths Forward, 2022

There are two prevailing methods for pre-training on large datasets to learn transferable representations: 1) supervised pre-training on large but weakly-labeled datasets; 2) contrastive training on image only and on image-text pairs. While supervised pre-training learns good representations that can be transferred to a wide range of tasks, contrastively trained models such as CLIP have demonstrated unprecedented zero-shot transfer. In this work we compare the transferability of the two aforementioned methods to multiple downstream tasks.

Studying the impact of magnitude pruning on contrastive learning methods
Francesco Corti*, Rahim Entezari*, Sara Hooker, Davide Bacciu, Olga Saukh
ICML Hardware Aware Efficient Training workshop, 2022

We study the impact of different pruning techniques on the representation learned by deep neural networks trained with contrastive loss functions. Our work finds that at high sparsity levels, contrastive learning results in a higher number of misclassified examples relative to models trained with traditional cross-entropy loss.

The Role of Permutation Invariance in Linear Mode Connectivity of Neural Networks
Rahim Entezari, Hanie Sedghi, Olga Saukh, Behnam Neyshabur
Accpeted at ICLR, 2022
Twitter / arXiv

We conjecture that if the permutation invariance of neural networks is taken into account, SGD solutions will likely have no barrier in the linear interpolation between them.

Understanding the effect of sparsity on neural networks robustness
Lukas Timple*, Rahim Entezari*, Hanie Sedghi, Behnam Neyshabur, Olga Saukh
ICML Overparameterization workshop, 2021

We show that, up to a certain sparsity achieved by increasing network width and depth while keeping the network capacity fixed, sparsified networks consistently match and often outperform their initially dense versions.

Class-dependent Pruning of Deep Neural Networks
Rahim Entezari, Olga Saukh
IEEE Second Workshop on Machine Learning on Edge in Sensor Systems (SenSys-ML), 2020

we propose an iterative deep model compression technique, which keeps the number of false negatives of the compressed model close to the one of the original model at the price of increasing the number of false positives if necessary.

Deep and Efficient Impact Models for Edge Characterization and Control of Energy Events
Grigore Stamatescu, Rahim Entezari, Kay Roemer, Olga Saukh
IEEE 25th International Conference on Parallel and Distributed Systems (ICPADS), 2019

we present a hierarchical energy system architecture with embedded control for network control in microgrids.

Avid: Adversarial visual irregularity detection
Mohammad Sabokrou, Masoud Pourreza, Mohsen Fayyaz, Rahim Entezari, Mahmood Fathy, Juergen Gall, Ehsan Adeli
Asian Conference on Computer Vision (ACCV), 2018

We propose an end-to-end deep network for detection and fine localization of irregularities in videos (and images).

Thanks Jon!