In our research, we focused on optimizing the BERT model, a cornerstone in modern NLP applications known for its multi-headed attention mechanism. While BERT's efficacy in various NLP tasks is undisputed, its substantial computational demands limit its broader applicability. To address this, we employed "Knowledge Distillation" to scale down the model. Our approach involved reducing the number of layers, attention heads, and other hyperparameters in the BERT-base-uncased model. This strategy successfully decreased the model size and training time, with only a negligible impact on fine-tuning accuracy for downstream tasks. Our work contributes to making advanced NLP models more accessible and feasible for a wider range of applications.

In my project, I delved into the realm of Deep Learning, a subset of Machine Learning and a crucial aspect of Artificial Intelligence, to explore its applications in computer vision. The primary goal was to gain practical experience in implementing deep learning for image and video processing. To achieve this, I developed an Android application capable of real-time object detection as well as recognizing objects in images. The backbone of this project was YOLO V4, chosen for its efficiency as an Object Detection Algorithm (ODA). The model was trained using the MS COCO dataset, enabling the application to identify 80 different object classes. This project not only provided valuable insights into the capabilities of deep learning in computer vision but also showcased the integration of sophisticated machine learning models into user-friendly mobile applications.

During my software engineering course, I developed this website along with my project partner which basically is a dedicated social media platform for the teachers and students who can communicate with each other, open room to have study sessions, discussions etc. The accounts can only opened with nsu.edu mail. During the entire development, the agile method was thoroughly followed. From time to time feedback was taken from NSU students and teachers and implemented accordingly. The report mentions every aspect of the project and link to the repository is given below.

I developed a voice command application for Android utilizing Android Studio and Java. This application integrates Google's Voice API to facilitate voice recognition and processing. Users can interact with the app through voice commands, making it user-friendly and accessible. The integration of Google's Voice API ensures high accuracy in voice recognition, enhancing the overall user experience. This project showcases my skills in mobile application development and my ability to incorporate advanced APIs for improved functionality. The report mentions every aspect of the project and link to the repository is given below.

I designed and built a rain detector using Arduino and electronic circuits. This device is capable of detecting rainfall and can be classified as an Internet of Things (IoT) project if it features network connectivity. By integrating components for data collection and communication, the rain detector can send real-time weather updates or alerts, making it an innovative and practical tool for weather monitoring and related applications.

The first proper project in my undergrad life was programming this wonderful 2D arcade game called Obstacles. In this game a aircraft in space shoots stars and collect points dodging obstacles. It took me night and days during the covid pandemic to complete it. I am really proud of this. The report of this game is attached, also the git repository is mentioned below. Just click the logo.