Moein Sorkhei

Studies

• • Master's level: KTH Royal Institute of Technology

  • Bachelor's level: University of Tehran

Research Interests

• • Deep learning applied to computer vision, especially medical in medical imaging tasks

  • Uncertainty in deep learning algorithms

  • Explainatiablity and interpretability of deep learning algorithms

Awards and Honors

• • KTH One-Year Scholarship (certificate) [2019] - Awarded to 15 out of 160 applicants with exceptional academic records

  • Ranked 4th among 30 IT student at University of Tehran [2018]

  • Ranked 658th among more than 300,000 participants (i.e, top 0.21%) in the Nationwide University Entrance Exam [2013]

Projects

Here is a list of the related projects that I did. For a more detailed description, please refer to my CV.

• • [Re] Unsupervised Representation Learning for Multivariate Time Series

As part of the Advanced Deep Learning course and the NeurIPS 2019 Reproducibility Challenge, my teammates and I have re-implemented this paper. Our submission is selected to be published in the ReScience-C Journal — Advanced Deep Learning Course, Aug. 2019.

• • Thoracic Disease Classification and Localization

In this project, deep-learning-based algorithms are implemented for the classification and localization of thoracic diseases in X-ray images by fine-tuning the parameters of ImageNet pre-trained models. Different attention mechanisms are also employed which improve the performance in some cases. The report could be found here — Project Course in Data Science, Aug. 2019.

• • Generative Modeling Using Normalizing Flows:

In this project, a recent flow-based generative model, Glow, is used for the task of generating images on different datasets in unconditional or conditional scenarios. The goal is to explicitly use the useful representation of data learned by flow-based models to generate novel images. The project is being done at the moment, but recent advances could be found here and a very preliminary report could be viewed here — Advanced Individual Course in Computer Science, Aug. 2019.

• • Essays on Deep Learning Topics

Scientific summarization of related papers in different areas of deep learning in the form of essays: an essay on uncertainty estimation of neural networks (Bayesian backpropagation), an essay on deep generative models, and an essay on understanding deep neural networks (observing the effect of batch size on generalization of deep neural networks) — Advanced Deep Learning Course, Aug. 2019.

• • Project on Topic Modeling

Implementation and analysis of the Latent Dirichlet Allocation algorithm for topic modeling in different text datasets and its comparison to other algorithms used in the field. The report could be seen here — Advanced Machine Learning Course, Oct. 2018.

• • Assignments in Advanced Machine Learning Topics

Theoretical assignment on topics including Bayesian linear regression, Gaussian processes, linear representation learning, and model selection. Please refer to the assignment and derivations for more details.

Theoretical assignment on diverse topics in directed graphical models including d-separation, expectation maximization, and variational inference on mixture of hidden Markov models. Please refer to the assignment and derivations for more details — Advanced Machine Learning Course, Oct. 2018.

• • Assignments in Probabilistic Graphical Models Topics

Assignments on diverse topics in directed and undirected graphical models including the message passing algorithm, evaluation of Bayesian networks using scores such as K2 and BIC, applying conditional random fields to handwritten image data for optical character recognition.

Assignment on combining variational inference with sequential Monte Carlo methods with importance sampling and observing their improvements on the evidence lower bound (ELBO) — Probabilistic Graphical Models Course, Mar. 2019.

• • Projects on Machine Learning Topics

Implementing and analyzing the functionalities of decision trees, SVMs with different kernel functions, the NB classifier with boosting for classification, and PCA for dimensionality reduction — Machine Learning Course, Aug. 2018.

• • Paper Exposition on Approximation Algorithms

An exposition of a randomized approximation algorithm used for K-means clustering. The report could be found here — Advanced Algorithms Course, Aug. 2018.

• • Music Recognition

Implementation of a simple music recognizer based on the fingerprinting algorithm of the music frequencies. The report could be found here — Speech Technology Course, Mar. 2019.

Related Courses

Here I have listed the most relevant courses I have passed throughout my master's study along with their grades (A-F scale).

• • Advanced Deep Learning: A

  • Machine Learning: A

  • Advanced Machine Learning: A

  • Probabilistic Graphical Models: A

  • Advanced Algorithms: A

For a complete list of courses please refer to my transcript of records.

Teaching Experience

• • TA in Artificial Neural Networks and Deep Architectures, EECS Department, KTH Royal Institute of Technology

  • TA in Advanced Algorithms, EECS Department, KTH Royal Institute of Technology

  • TA in Artificial Intelligence, ECE Department, University of Tehran

  • TA in Data Structures and Algorithms, ECE Department, University of Tehran

  • TA in Design and Analysis of Algorithms, ECE Department, University of Tehran