Shayan Shekarforoush

About Me

I am currently studying the final year of my bachelor’s degree in Computer Engineering at Sharif University of Technology, Iran. My research interests include Deep Learning in particular, and Machine Learning in general, with applications in Computer Vision and Graph Learning. I am intended to better understand these methods in a principled manner by investigating both their underlying mathematics and their practical implications in real life. Though the recent proposed techniques have demonstrated remarkable performance, major challenges in terms of understanding their underlying mathematical foundations remain unresolved. My research goal is to channel such better theoretical and functional understanding of these new methods towards better practical solutions for real life problems.

Publications

Cell type identification in single-cell RNA sequencing based on gene interaction networks


Current single-cell classification methods take genes as independent features to assign cell types disregarding their interactome which has been shown to be of great importance in cellular processes. Here, we introduce single-cell Graph Convolutional Network (scGCN) which takes gene-gene interactions into account by constructing a co-expression network of a subset of genes and uses a graph deep learning approach to classify single-cells.

Elyas Heidari, Shayan Shekarforoush, Laleh Haghverdi, and Wolfgang Huber

Intelligent Systems for Molecular Biology / European Conference on Computational Biology (ISMB/ECCB 2019)

Graph Convolution Based Attention Model for Personalized Disease Prediction


Often a varied order of importance for the multi-modal heterogeneous data is considered for personalized disease diagnosis decisions. Towards this, we introduce a model which not only improves the disease prediction but also focuses on learning patient-specific order of importance for multi-modal data elements. In order to achieve this, we take advantage of LSTM-based attention mechanism and graph convolutional networks (GCNs) to design our model. GCNs learn multi-modal but class-specific features from the entire population of patients, whereas the attention mechanism optimally fuses these multi-modal features into a final decision, separately for each patient.

Anees Kazi, Shayan Shekarforoush, S.Arvind krishna, Hendrik Burwinkel, Gerome Vivar, Benedict Wiestler, Karsten Kortum, Seyed-Ahmad Ahmadi, Shadi Albarqouni, and Nassir Navab

Medical Image Computing and Computer Assisted Intervention (MICCAI 2019)

InceptionGCN: Receptive Field Aware Graph Convolutional Network for Disease Prediction


In this paper, we introduce a new spectral domain architecture for deep learning on graphs for disease prediction. The novelty lies in defining geometric Inception Modules which are capable of capturing intra and inter-graph structural heterogeneity during convolutions. We design filters with different kernel sizes to build our architecture.

Anees Kazi, Shayan Shekarforoush, S.Arvind krishna, Hendrik Burwinkel, Gerome Vivar, Karsten Kortum, Seyed-Ahmad Ahmadi, Shadi Albarqouni, and Nassir Navab

Information Processing in Medical Imaging (IPMI 2019) [PDF][Code]

Self-Attention Equipped Graph Convolutions For Disease Prediction


Multi-modal data comprising imaging (MRI, fMRI, etc.) and non-imaging (demographics, etc.) data can be collected together and used for disease prediction. A model capable of leveraging the individuality of each multi-modal data is required for better disease prediction. We propose a graph convolution based deep model which takes into account the distinctiveness of each element of the multi-modal data. We incorporate a novel Self-Attention Layer, which weights every element of the demographic data by exploring its relation to the underlying disease.

Anees Kazi, S.Arvind krishna, Shayan Shekarforoush, Karsten Kortum, Shadi Albarqouni, and Nassir Navab

IEEE International Symposium on Biomedical Imaging (ISBI 2019) [PDF]

Experience

Technical University Munich

Research Intern

July 2018 - September 2018

Remote Collaboration

September 2018 - May 2019

Research Intern

July 2019 - September 2019

In summer 2018, I joined Computer Aided Medical Procedures (CAMP), as an intern under the supervision of Prof. Dr. Nassir Navab, at Technical University of Munich (TUM). I started to study the recently emerging topic of Geometric Deep Learning. My first major involvement involvement with the team was on a research project aiming to address the problem of semi-supervised disease prediction in a multi-modal setting, when the dataset comprises of both imaging and non-imaging data. This work was published in the proceedings of the IEEE International Symposium on Biomedical Imaging (ISBI 2019), and was presented as an oral contribution.

In another research project, I investigated how changing the receptive field size in GCN alters its overall performance. Accordingly, we designed a novel method called InceptionGCN utilizing inception modules, where each module has a different kernel size. The result of this study is now published and accepted for oral presentation (with 7% acceptance rate) in the conference of Information Processing in Medical Imaging (IPMI 2019).

As continuation of my work on the first paper, I committed to a remote collaboration with CAMP on another project in which our team incorporated a Long Short-Term Memory (LSTM) based attention mechanism into GCN to make patient-specific disease predictions. This work also resulted in a paper published in the International Conference on Medical Image Computing and Computer Assisted Intervention (MICCAI 2019).

In summer 2019, I rejoined CAMP, but this time I took the initiative and identified the problem of automatically learning the receptive field size in GCNs as an open problem. I proposed a new model capable of adaptively learning the receptive field size in an end-to-end manner. Later, I realized that my model has direct applications to 3D shape analysis tasks such as mesh regression and shape correspondence.

European Molecular Biology Laboratory

Remote Collaboration

December 2018 - March 2019

I joined a project on single-cell transcriptomics under the supervision of Dr. Laleh Haghverdi, winner of an Erwin Schrödinger Prize. Most of the existing single-cell classification methods take genes as independent features to assign cell types disregarding their interactions, which has been shown to be of great importance in cellular processes. We introduced single-cell Graph Convolutional Network (scGCN), which takes gene-gene interactions into account by constructing a co-expression network of a subset of genes. The results of this work were presented as a poster in the ISMB/ECCB 2019 conference.

Sharif University of Technology

Research Assistant

April 2018 - Present

As part of my Bachelor Thesis, I worked under supervision of Dr. Soleymani on Graph Learning in Heterogenous Networks. In this project, I have proposed and implemented a novel Autoencoder approach to efficiently solve the problem of link prediction in networks with diversity in nodes.

DataDays

Scientific Staff

December 2018 - March 2019

DataDays was the First National Data Scientific Competition held by Sharif University of Technology. Its main goal was to broaden appeal for working on problems related to data. Moreover, it is a large-scale attempt to assess the level of knowledge in this field in Iran.

As a volunteer, I joined the Scientific Staff and help them to propose questions which could be solved by Machine Learning and Deep Learning methods. In addition, we developed some Perisan Jupyter Notebooks as tutorials mostly related to Data Analysis and Machine Learning. We made it accessible to all the teams in order to encourage them to raise their skill and knowledge in these fields.

The whole competition was centered around working on a dataset collected by Cafebazaar from Divar. This dataset mainly comprises online advertisements published by people in different categories.

Yektanet

Data Scientist

July 2017 - September 2017

During summer of 2017, I worked for a start-up called Yektanet, an online advertisement platform in my country. This platform is basically publishing advertisements on well-known publishers for achieving clicks and views. I was employed as data scientist and developed algorithms determining the subject of web pages using their textual data. They are now deployed for displaying relevant items of advertisement on each webpage.

Sharif Univeristy of Technology

Teaching Assistant

Some of my teaching experiences as assistant:

Spring 2019:

Fall 2018:

  • Stochastic Processes (Graduate Course)
  • Linear Algebra

Spring 2018:

  • Probability and Statistics

Fall 2017:

  • Probability and Statistics

Education

Sharif Univeristy of Technology

BSc Computer Engineering

2015 - Present

Studying Software Engineering at Computer Engineering Department.

Related Courses:

Signal Processing, Probability and Statistics, Linear Algebra, Regression Analysis, Stochastic Processes, Artificial Intelligence, Introduction to Machine Learning, Differential Geometry, Information Theory (Graduate Course).

Allameh Helli 1 Highschool

Diploma

2011-2015

I received my Diploma in Mathematics and Physics from this highschool affiliated with the National Organization for the Development of Exceptional Talents (NODET).

A Little More About Me

Alongside my interests in Science and Technology, some of my other hobbies are:

  • Rock Music
  • Ultimate Frisbee
  • Basketball
  • Playing Tennis