Education
University of Waterloo
Master's in Computer Science
Sep. 2021 - Sep. 2023
Supervisors: Prof. Yuri Boykov
and Prof. Olga Veksler
Related Coursework: Deep Learning, Computational Vision, Reinforcement Learning
Thesis: Occlusion-ordered Semantic Instance Segmentation
Sharif University of Technology
BSc in Computer Engineeiring
Sep. 2016 - Feb. 2021
Supervisor: Prof. MH Rohban
Related Coursework: Probability & Statistics, Linear Algebra, Artificial Intelligence, Signals & Systems, Modern Information Retrieval, Algorithmic Game Theory, Numerical Methods, Design of Algorithms, Computer Simulation, Database Design, Data Structures & Algorithms
Publications
Is critical input information encoded in specific sparse pathways within the neural network? In this work, we discuss the problem of identifying these critical pathways and subsequently leverage them for interpreting the network’s response to an input. We demonstrate that sparse pathways derived from pruning do not necessarily encode critical input information. …
Ashkan Khakzar, Soroosh Baselizadeh, Saurabh Khanduja, Christian Rupprecht, Seong Tae Kim, and Nassir Navab
IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) 2021 [arXiv][CVF OpenAccess]
Here, we propose to use the “distillation” of features at various layers of an expert network, which is pre-trained on ImageNet, into a simpler cloner network to tackle issues in unsupervised representation learning for anomaly detection/localization. We detect and localize anomalies using the discrepancy between the expert and cloner networks’ intermediate activation values given an input sample. …
MohammadReza Salehi, Niousha Sadjadi*, Soroosh Baselizadeh*, M. H. Rohban, and Hamid R. Rabiee
IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) 2021 [arXiv][CVF OpenAccess]
The black-box nature of CNNs has sparked many recent works to explain the prediction via input feature attribution methods. However, input these methods merely identify the importance of input regions for the prediction and lack semantic interpretation of model behavior. Here, we first identify the semantics associated with internal units (feature maps) of the network. We proceed to investigate the following questions; …
A. Khakzar, S. Musatian, J. Buchberger, I. Valeriano Quiroz, N. Pinger, S. Baselizadeh, S. T. Kim, Nassir Navab
International Conference on Medical Image Computing and Computer-Assisted Intervention (MICCAI) 2021 [arXiv] [Springer (DOI)] [SharedIt]
Rethinking Positive Aggregation and Propagation of Gradients in Gradient-based Saliency Methods
A popular family of saliency methods utilize gradient information. In this work, we empirically show that two approaches for handling the gradient information, namely positive aggregation, and positive propagation, (e.g. in methods like GradCAM++ and FullGrad) break these methods. Though these methods reflect visually salient information in the input, they do not explain the model prediction anymore …
Ashkan Khakzar, Soroosh Baselizadeh, Nassir Navab
International Conference on Machine Learning (ICML) 2020 - Workshop on Human Interpretability in ML (WHI) [arXiv][Full Proceedings of the Venue]
Experiences
Machine Learning Researcher and Engineer
Electronic Arts (EA)
Sep. 2021 - Present
Advanced Technology Group (ATG) @ EA Sports, Vancouver, Canada
- Working on novel computer vision and ML solutions to enhance gaming experiences across EA Sports titles.
Remote Research Assistant
CAMP @ TUM, Germany
July 2019 - March 2021
https://www.cs.cit.tum.de/camp/start/Computer Aided Medical Procedueres and Augmented Reality (CAMP) @ Technical University of Munich, Germany
- Collaborated in a multi-organizational and international team from the UK (Oxford) and Germany (TUM)
- Designed and implemented a new algorithm for explaining deep image classification models using critical pathways in neural networks, improving the state-of-the-art on CIFAR, BirdSnap by 10%
- Implemented sanity checks methods from scratch using PyTorch, numpy, sklearn, and scipy, including layer-by-layer model randomization and input randomization
- Formulated the critical pathways in deep models as a linear approximation, proving the efficacy of the method
- Implemented, and analyzed weaknesses of 5+ gradient-based image saliency methods using PyTorch
- Developed parts of a study on interpreting the COVID-19 diagnosis deep models and their features’ semantics
- Wrote significant parts of publications at CVPR, ICML, and MICCAI
- Used Git continuously for version control and coordination
Research Assistant
Robust and Interpretable ML (RIML) Lab
March 2019 - March. 2021
Robust and Interpretable Machine Learning (RIML) Lab @ Sharif University of Technology, Iran
- Developed a new method for image anomaly detection & localization based on knowledge distillation & transfer learning improving the state-of-the-art up to 20% on 7 industrial and medical datasets
- Implemented several interpretability methods of deep neural networks including Grad-CAM, Integrated Gradients, Vanilla Gradients in PyTorch
- Designed, and implemented experiments for evaluation/ablation using numpy, PyTorch, and matplotlib
- Wrote most parts of a paper published at CVPR
- Presented findings using Jupyter and Tensorboard in meetings, and performed version control on Git
Natural Language Processing (NLP) Intern
- Crawled, and scrapped user comments data from Twitter and Persian news websites using Scrappy
- Cleaned, normalized, and pre-processed the collected text data in Python
- Performed primary data analysis and visualization using Pandas, and matplotlib
- Labelled, and partitioned train/test data and created a dataset with near 100k size for offensive vs. non-offensive Persian comments based on the preliminary data analysis
- Implemented Transformer models based on BERT using Wordpiece tokenizer in PyTorch for offensive vs. non-offensive classification of Persian comments
- Tested, and benchmarked different versions of the model using PyTorch, Seaborn and matplotlib
- Documented, and presented the findings in meetings and to the R&D department
Honors & Awards
-
Reviewer at CVPR2024, AAAI2025, CVPR2025
-
David R. Cheriton Graduate Scholarship, Department of Computer Science, University of Waterloo [2021-2023]
-
International Master’s Award of Excellence (IMAE), University of Waterloo [2021-2023]
-
Undergraduate Excellence Award from the Technical University of Munich, Germany [2019]: given annually to <= 5 international students with an excellent background by the chair for Computer Aided Medical Procedures and Augmented Reality (CAMP).
-
Ranked 3rd based on GPA out of all ~110 bachelor students of Computer Enigneering Department @ Sharif University of Technology entered at 2016.
-
Iran’s National Elites Foundation (INEF) Fellowship [2016-2021]: Recognized as scientiffic elite.
-
Iran’s National University Entrance Exam (Konkur) Top Scorer [2016]: Ranked 12th nationally (2nd in the relevant region) out of more than 180’000 (50’000 in the relevant region) participants.