Vahid Zehtab

De omnibus dubitandum est.

About Me

Hi there! I'm Vahid, a Machine Learning Researcher & Engineer based in Canada with over five years of industry and academic experience under my belt. I have a knack for solving problems through a mathematical lens, particularly in the realm of deep learning and computer vision. This is a brief summary of my academic and professional background.

Most recently, I've been exploring consultation roles, on the lookout for the ideal full-time position that aligns with my interests in Machine Learning R&D. If you're interested in discussing collaborations, reach out and let's build the future together!

Checkout my work & Connect: GithubLinkedInGoogle ScholarEmail


BSc. in Computer Engineering

Sharif University of Technology

Sep. 2017 - Jun. 2022 • Email

  • CGPA: 18.64/20 (3.8/4) — 141/140 credits
  • Capstone Project (Thesis): Anomaly Detection via Explicit Density Estimation
  • Selected Courses (All courses): Digital Image Processing (Graduate), Stochastic Processes, Artificial Intelligence, Linear Algebra, Signals & Systems, Probability and Statistics, Design of Algorithms, Data Structures and Algorithms
  • Selected Teaching Assistantships (All TAships): Head TA of Artificial Intelligence, supervised graduate TAs in Machine Learning Privacy (Graduate), tutored deep learning workshops in Machine Learning for Bioinformatics (Graduate), in addition to TAing and tutoring recitation classes for undergraduate courses such Bioinformatics, Probability theory and Statistics, Linear Algebra, and Data structures and Algorithms.

  • Selected Extracurricular Activities (All activities): Led the scientific team for Iran's largest data science contest on recommender systems (Sharif’s Datadays 2021), Designed interactive AI courses for highschoolers and undergrad freshmen for Sharif University of Technology’s RASTA and MIL events.

Professional Experience
Samsung AI Center (Toronto)

Machine Learning Research Intern
Supervised by: M.S. Brown, M.A. Brubaker, D.B. Lindell

May 2023 - Dec. 2023 • Toronto, ON, Canada

  • Assisted the camera quality team in noise modeling and denoising of RAW quad Bayer sensor outputs.
  • Designed and implemented a neural framework for efficient representation of color transformations and inverse vision processing, surpassing the SOTA by over 100x in terms of compression efficacy.

Advanced Technology Lab (ZLAB), Fanap Co.

Machine Learning Engineer

Jun. 2020 - Mar. 2021 • Tehran, Iran

  • Developed a novel satellite and aerial imagery super-resolution residual model to enhance the on-device perception of urban maps in Iran's largest navigation system by upto 8x in resolution.
  • Developed a computer vision composable deep learning prototyping framework that enables direct Neural Architecture Search (NAS) over pure PyTorch without the need of implementing extra logic.


Chaos Engineering Intern

Jul. - Oct. 2019 • Tehran, Iran

  • Implemented a Kubernetes cluster node failure simulator and designed a cluster monitoring system for internal stress-testing of the underlying infrastructure.
  • Designed and implemented a decentralized load-tester that learns and mimics true user behaviour and can scale to simulate up to millions of requests per second, for internal use.


Computer Vision Engineering Intern

Jun. 2018 - Jun. 2019 • Tehran, Iran

  • Co-developed an efficiently accurate OCR engine in C++, based on Google's Tesseract OCR Engine, for extracting information from Iranian debit card scans.
Research Experience
Vector Institute
Affiliated with UofT's Machine Learning & Computational Healthcare Lab

Graduate researcher supervised by R.G. Krishnan

Oct. 2022 - Dec. 2023 • Toronto, ON, Canada

  • Co-developed a fully differentiable Bayesian algorithm for causal structure discovery with Autoregressive Normalizing Flows extending the identiability of LSNMs to beat the SOTA in real-world datasets.
  • Developed DyPy & Lightning-Toolbox for fast implementation of complex deep learning research experiments, cutting the amount of raw code in half on average.

Sharif University of Technology

Undergraduate researcher supervised by M.H. Rohban

Dec. 2019 - Jul. 2022 • Tehran, Iran

  • Assessed the potential of repurposed adversarial training of robust reconstructive models for visual anomaly detection tasks.
  • Developed a novel deep training procedure for anomaly detection using explicit density estimation, on par with SOTA Autoencoder-based methods.
  • Studied energy-based and other explicit deep density estimation models and developed TorchDE, a unified pytorch library for deep density estimation.

École Polytechnique Fédérale de Lausanne (EPFL)

Collaboration with PhD students at LTS4 & MLO labs

Feb. 2021 - Sep. 2021 • Lausanne, Switzerland (Remote)

  • Studied the preliminary theoretical foundations of deep learning and computed the directional inductive bias of Vision Transformers revealing their inert differences to traditional convolution-based solutions.

École Polytechnique Fédérale de Lausanne (EPFL)

Remote Research Intern at VITA lab, Supervised by A. Alahi

Jul. 2020 - Dec. 2020 • Lausanne, Switzerland (Remote)

  • Assessed the state-of-the-art vehicle trajectory prediction models' explainability and reliance on different data priors.
  • Explored alternatives to rasterized data representation schemes and co-developed a Transformer based model that uses generic SVG inputs to perform reason-aware vehicle trajectory prediction tasks.
Efficient Neural Network Encoding for 3D Color Lookup Tables (Under Review)
Zehtab, V., Brown, M.S., Brubaker, M.A., Lindell, D.B. *Equal contribution as first Author

Order-based Structure Learning with Normalizing Flows (Under Review)
Kamkari, H.*, Balazadeh, V.*, Zehtab, V., Krishnan, R.G. *Equal contribution as first Author

SVG-Net: An SVG-Based Trajectory Prediction Model
Bahari, M., Zehtab, V.*, Khorasani, S.*, Ayromlou, S.*, Saadatnejad, S., Alahi, A. *Equal contribution as second author

Anomaly Detection via Explicit Density Estimation (Bachelor's thesis)
Zehtab, V.*, Shariat, K.*, Rohban, M.H. *Equal contribution as first Author