SOURAJIT SAHA

Office: Room ****, Level *, Dept. Of CSE

Independent University Bangladesh, Dhaka 1229

B adk2b6@r.postjobfree.com

ccse.iub.edu.bd/sourajit

Education

2021–

Ongoing

University of Maryland Baltimore County, USA,

Computer Science, PhD.

2013–2017 BRAC University, Bangladesh,

Computer Science, Bachelor of Science.

Experience

Jan 2021–

Ongoing

Graduate Teaching Assistant, Department of Computer Science & Electrical Engineer- ing, University of Maryland Baltimore County, USA. Nov

2019–July

2020

Research and Development Officer, School of Engineering & Computer Science, IUB.

{ Writing & Evaluating Research Proposals.

{ Developing a Neural-Net Training Scheduling Platform on GPU resources of CCSE Super Computer. July

2018–July

2020

Research Assistant, Center For Cognitive Skill Enhancement (CCSE), IUB.

{ Conducting Research and Experiment on Deep Learning and it’s application on Computer Vision.

{ Published in conferences at USA, India, Bangladesh.

{ Taught CS 433 : Deep Learning & Computer Vision, CS 439 : Mathematics for Deep Learning. April

2018–Dec

2018

Adjunct Junior Lecturer, Independent University Bangladesh. Courses Taught at School of Engineering & Computer Science

{ CSE 203 : Data Structure (Summer 2018).

{ CSE 211 : Algorithms (Autumn 2018).

Jan

2017–April

2017

Trainer, Sports Programming, BRAC University.

Department of Computer Science & Engineering

{ Combinatorics, Dynamic Programming, Graph Theory. Publications

JJJ Google Scholar Citation = 37, h-index = 4, Number of Reads on Research Gate = 2700. Sep 2020 A comprehensive set of novel residual blocks for deep learning architectures for diagnosis of retinal diseases from optical coherence tomography images. [Book Chapter on Springer].

{ Published as a book chapter in Deep Learning Applications Volume 2, page: 25-48. Sep 2019 Optic-Net: A Novel Convolutional Neural Network for Diagnosis of Retinal Diseases from Optical Tomography Images. [IEEE Xplore].

{ Published and Presented at 18th IEEE International Conference on Machine Learning and Applications

- ICMLA 2019, FL, USA.

{ Proposed Computationally Cheaper Residual Building Blocks for CNN with a Novel Signal Attenuation Mechanism to Prevent Gradient Degradation.

{ Achieved State-of-the-art Performance on Multiple OCT Datasets.

{ Outperformed Human Diagnosticians in Real Time.

Feb 2019 Total Recall: Understanding Traffic Signs Using Deep Convolutional Neural Network . [IEEE Xplore].

{ Published and Presented at 21st IEEE International Conference of Computer and Information Technol- ogy - ICCIT 2018, Dhaka, Bangladesh.

{ Proposed a novel Dilated Skip Connection Mechanism within nested Residual Building Block for CNN.

{ Achieved State-of-the-art Performance on Multiple Traffic Sign Datasets. 1/3

Dec 2018 A Lightning fast approach to classify Bangla Handwritten Characters and Numerals using newly structured Deep Neural Network. [ELSEVIER - Procedia Computer Science].

{ Published and Presented at International Conference on Computational Intelligence and Data Science 2018, Delhi, India.

{ Proposed a novel CNN Block called Divide and Merge Mapping (DAMM) to decompose the Neural Network Learning Function.

{ Devised an Exhaustive Optimal Path Selection Search to downsize the Neural Network Complexity.

{ Achieved State-of-the-art performance on the Largest Dataset on Isolated Bengali Handwritten Characters. Nov 2018 Detecting Sex From Handwritten Examples. [IEEE Xplorer].

{ Published and Presented at 2018 IEEE International Conference On System, Computation, Automation and Networking, Pondicherry, India.

{ Proposed an Algorithm to train Neural Networks on Encoded Data.

{ Propounded an CNN-RNN ventured pipeline to train Image Classifiers. Sep 2018 An Efficient Traffic Sign Recognition Approach Using a Novel Deep Neural Network Selection Architecture. [Undergraduate Thesis] [Springer - Advances in Intelligent Systems and Com- puting].

{ Published and Presented at International Conference on Emerging Technologies in Data Mining and Information Security 2018, Kolkata, India.

{ Devised an algorithm to train cascade of CNNs for Image Classification.

{ Unprecedentedly Surpassed Class Imbalance Bias on GTSRB Dataset without Over-fitting. Current Projects

May 2019–

Ongoing

Rethinking Activation Functions In Neural Networks & Down-sampling Schemes In CNN, VIRG, CCSE - IUB.

August 2019–

Ongoing

Studying Loss Functions To Address Class Imbalance Problem & Their Consequences On Optimization Performance In Semantic Segmentation Models, VIRG, CCSE - IUB. Oct 2019–

Ongoing

Attaining Super-resolution with Visual Attention & Deep Learning, VIRG, CCSE - IUB. Nov 2019–

Ongoing

Collecting & Annotating Object Detection & Semantic Segmentation Dataset On The Streets Of Dhaka, VIRG, CCSE - IUB, Sponsored by Uber BD.

Research Interest

Future Deep Learning, Computer Vision.

{ Addressing the pressing issues in Object Detection, Semantic Segmentation and Instance Segmentation.

{ CNN Architecture, Pose Estimation, Object Tracking, Traffic Monitoring, Medical Image Analysis, Saliency Prediction, Super-Resolution, Pattern Recognition, Adversarial Attacks, Visual Attention.

{ Studying the viability of Interpretative Convolutional Neural Networks w.r.t. cognizant functionalities.

{ Learning Knowledge Representation from Images.

Deep Learning Seminars I Spoke At

2018–2019 Bengali.AI, BRAC University, Tech Academy Bangladesh, Independent University Bangladesh. Skills

Languages Python, C++, Java, Shell Script

Frameworks PyTorch, Detectron2, Tensorflow, CUDA programming OS Ubuntu, Windows, MAC OS

Miscellaneous LaTeX, Deep Learning, Data Structure & Algorithms, Linear Algebra, Optimization Theory Achievements & Awards

Fellowship Research Fellowship Award : Graduate School Division, Independent University Bangladesh, Granted BDT 500,000 (USD 6,000) : Spring 2019 - Autumn 2019. Scholarship Dean’s Award & Vice Chancellor’s Award For Academic Excellence at BRAC University. 2014–2017 Champion, Intra University Programming Contest, BRAC University. 2/3

2015 Ranked 22 in Shahjalal University of Science & Tech - Inter University Programming Contest. 2017 Ranked 15 in MBSTU Programming Contest, 13 in City University IUPC. References

Mahady Hasan, PhD,

Associate Professor & Head of the Dept.,

Dept. Of CSE, Independent University Bangladesh,

adk2b6@r.postjobfree.com.

Amitabha Chakrabarty, PhD,

Associate Professor,

Dept. Of CSE, BRAC University,

adk2b6@r.postjobfree.com.

3/3

Contact this candidate