deep learning

Projet personnel de classification des maladies des feuilles d'arbre. Article technique disponible sur mon blog : https://boulayc.gitlab.io/blog/

A customized Pix2Pix implementation for inpainting the headless region of seated Buddha statues found in Anuradhapura, Sri Lanka.

This is a combination of a segmentation model and a detection model trained on seated Buddha Statue dataset collected from Google Images based on YOLO v8. This can be used to segment seated Buddha Statue objects from images, and then add a mask to the statue head. This is expected to be used as inputs to an inpainting work based on GAN.

This is a detection model trained on seated Buddha Statue dataset collected from Google Images. The model can be used to detect seated Buddha Statue objects from images under two classes as; Statue head and Statue body. The model is based on YOLO v8

Research, code, algorithms and all related to artificial intelligence

FakET: Simulating Cryo-Electron Tomograms with Neural Style Transfer

Developed a project utilizing Generative Adversarial Networks (GANs) to convert grayscale images to RGB color images. Leveraged deep learning techniques to train the GAN model on a dataset of grayscale and corresponding color images, achieving realistic colorization results. This project demonstrated proficiency in image-to-image translation and advanced deep learning methodologies within the realm of computer vision.

Real-time Gender and Age Recognition from Audio

Controversy quantification of topics on twitter, based on user probability to participate in a controversy topic, using GNN and NLP models.

The aim of this project is to provide an exploratory analysis of Domain Adaptation (DA) techniques in the context of PHM for Bearings fault prognosis, focusing on Health Index (HI) estimation and Remaining Useful Life (RUL) prediction. The adopted dataset is the PRONOSTIA/FEMTO-ST bearings dataset.

A project focused on weather classification using advanced deep learning techniques, specifically leveraging TensorFlow and a custom Convolutional Neural Network (CNN). The project involved the integration of four diverse weather datasets, namely ACDC, MWD, UAVid, and Syndrone, covering various weather conditions, including clear sky, cloudy, rainy, and sunny weather. Developed a custom CNN architecture using TensorFlow's Keras API, incorporating convolutional layers for feature extraction and dense layers for classification.

Final project of course ML junior by Skillbox. Goal: Build model to predict client's loan default. Metrics: ROC-AUC

Master thesis - Deep learning-based pairwise alignment of protein sequences using transformer architecture.

Predicting pathogenic potentials of short DNA reads with reverse-complement deep neural networks.

Prediction of age from X-Ray images of hand bones using deep learning models. We used 3 models: Shallow, ResNet50, and InceptionV4. The best result achieved was with a mean absolute error of 10 months using InceptionV4. The preprocessing of data included computer vision techniques like CLAHE filter and reducing channels, and also creativities such as using the Google MediaPipe library to detect hands and crop on them.

A Machine Learning approaches to identify genetic variants associated with disease