AKBARALI VAKHITOV

About Me

I am an engineering graduate of Inha University, Incheon, South Korea. I have acquired ML(Supervised Learning/Unsupervised Learning), Deep Learning(NN, CNN, RNN, Transformers, LSTM, Attention...) C++, Python, Linux, Application Deployment skills, and more. Currently, as a research intern at CVHAI Lab (Hanyang University), I focus on manifold mixup and latent space interpolation techniques to enhance 3D UNet performance for CT scans. I strive to bridge theoretical advancements with real-world applications, confident that AI can drive impactful innovations in healthcare and beyond

Education

Inha University, Integrated System Engineering

• 03/2020 – 08/2024
• Graduated with Honors
• GPA: 3.65/4.0

Personal Projects

1. Multi-Class Classification of Chest Diseases

• This project is a multi-class classification system for chest diseases, specifically targeting the following classes: COVID-19 Viral Pneumonia, Bacterial Pneumonia, and Normal (No disease). Due to the scarcity of the dataset, this project leveraged the benefits of transfer learning. I utilized ResNet50, a pre-trained network on the ImageNet dataset, as the base model.
• Repo: Multi-Class Chest Diseases

2. Object Detection Finetuning using Mask R-CNN

• This project is about finetuning a pre-trained Mask R-CNN model on the Penn-Fudan Database for Pedestrian Detection and Segmentation. It contains 170 images with 345 instances of pedestrians, and it is used to illustrate how to use the new features in torchvision in order to train an object detection and instance segmentation model on a custom dataset.
• Repo: Mask R-CNN Finetuning

3. Spam Detection with LSTM

• This project demonstrates the implementation of a spam detection model using an LSTM (Long Short-Term Memory) network. The model is trained on a dataset of SMS messages to classify them as either spam or ham (non-spam).
• Repo: LSTM Spam Detector

4. Neural Network from Scratch in TensorFlow

• This project implements a simple neural network from scratch using TensorFlow. It includes the full process from initializing the network, performing forward propagation, computing loss, updating parameters, and training the model. The implementation demonstrates the core principles of neural networks and provides a foundation for further exploration and customization. The goal of this project is to implement a Neural Network model in TensorFlow using its core functionality (i.e., without the help of a high-level API like Keras). While it’s easier to get started with TensorFlow with the Keras API, it’s still worth understanding how a slightly lower-level implementation might work in TensorFlow
• Demonstrates manual forward propagation, parameter updates, training loop in raw TensorFlow.
• Repo: NN from Scratch

Research Experience & Internship

Research Intern at CVHAI Lab, Hanyang Univ.

• 10/2024 – Present, South Korea
• Project: Midpoint Interpolation to Learn Compact & Separated Representations for Medical 3D UNet Segmentation of CT Scans
• Lab Link: CVHAI Lab
• Achievements/Tasks:
  • Integrating manifold mixup at latent space for better class representation.
  • Exploring 3D UNet architectures for improved CT scan segmentation.