Pattern Recognition using Neuromorphic Computing

Developed a Bengali digit classifier using Spiking Neural Network (SNN). Used Leaky integrate and Fire neurons for computational unit and Rate coding as the encoding scheme. Simulated and observed the neural spiking pattern for different image inputs based on computational neuroscience and Hodgkin-Huxley model with a different type of synaptic connection with BRIAN toolkit and SLAYER framework (Undergrad Thesis)

Tools used: SLAYER Framework, BRIAN Toolkit, Python,

Cadence Automation

Layout, schematic, simulation, post-layout extraction flow automation with SKILL script and python in various Cadence tools.

Tools used: SKILL, Python, BASH, Cadence Virtuoso

4bit low power comparator using domino logic

Designed and simulate power-efficient 4bit low power comparator using domino logic with Cadence Virtuoso as VLSI lab project. Power consumption, timing analysis, etc. are checked. Used Cadence Virtuoso to simulate and make the layout of the design. DRC and LVS were also verified. (Undergrad Project)

Tools used: Cadence Virtuoso

4bit MIPS processor RTL synthesis, placement, and routing

Performed RTL synthesis of 4bit MIPS processor with Cadence Genus with various constraints. Placement and Routing (PNR) was done with Cadence Encounter. Different types of violations like DRC, setup and hold violation, max fanout violation, max transition violations were cleared as far as possible. (Undergrad Project)

Tools used: Cadence Genus, Cadence Encounter, ModelSim

4-2 Adder Compressor

Designed a power-efficient adder that can operate 4 additions simultaneously as VLSI project. The timing analysis, frequency response and power analysis were observed and matched with the original paper. Worked on 90nm technology so the power analysis was not accurately matched. Used Cadence Virtuoso/ Spectre to simulate this design. (Undergrad Project)

Tools used: Cadence Virtuoso

Unsupervised Anomaly Detection with Adversarial Defense

Participated in IEEE Signal Processing Cup 2020 with the title named “Abnormalities detection in the behavioral of the ground and aerial systems based on embedded sensor data in real-time” and achieved 8th position in the world ranking. Algorithm robustness was also studied by employing an Adversarial Attack and defense mechanism.

Tools used: Tensorflow, Keras, MATLAB, Python

Bangla Numeral Detection

Participated in Computer Vision Challenge on Bengali Handwritten Digit Recognition by Bengali.ai community and stood 7th. Used ensemble learning and custom CNN model based on VGG16. This is the first-ever machine learning contest in Bangladesh held in Kaggle based on a recently developed large dataset namely NumtaDB consists of 85000+ specimens.

Tools used: Keras, Tensorflow, MATLAB, Python

Video Activity Recognition

Participated in IEEE Video and Image Processing Cup 2019 (VIP Cup 2019) where an algorithm was developed for office activity recognition like writing, reading, working on pc, etc. from the video. For this a custom python, OpenCV wrapper was developed with C++ and CUDA toolkit for GPU accelerated optical flow detection using Dual TVL1 algorithm.

Tools used: OpenCV C++/Python, Keras, Tensorflow, Sonet, C++, Python

Camera Forensic

Participated in IEEE Signal Processing Cup 2018 (SP Cup 2018) with the title named “Forensic Camera Model Detection using Machine Learning” where a technique was developed to detect camera models from blind images using machine learning by analyzing and extract image features like noise patterns and demosaicing algorithm. Tensorflow/Keras (Python) library and MATLAB were used to solve the problem.

Tools used: Keras, Tensorflow, MATLAB, Python

IoT Project

Developed a home appliances power consumption monitoring system using Raspberry Pi, Arduino, and Adafruit.IO IoT framework with a web panel for controlling and monitoring appliances. (Undergrad Project)

Tools used: Adafruit.IO framework, Python

CV Sorting Software

Developed a CV sorting and interview management system with Java Spring Boot for my employer, Neural Semiconductor Limited as a part of automation training.

Tools used: Java Spring Boot, java

Android and web development

Developed a web API for the doctor-patient management system and Android app named BIOCTOR combined with IoT.

Tool Used: Android Studio, Laravel, Wacom Touchpad API, Java, Javascript, PHP