alessiococchieri

human-values-detection

This GitHub repository presents our solution to Touché 2023 Task 4: Human Value Detection, a multilabel text classification task. We fine-tuned transformer architectures like BERT, RoBERTa, and XLNet to classify whether or not a given argument draws on a human value category.

BDA-project-sparkify

This Git repo showcases my analysis of Sparkify dataset with PySpark on Apache Spark cluster mode and JupyterLab on Docker. The goal was to identify at-risk customers and develop retention strategies. The analysis tested multiple machine learning models and uncovered insights into customer behavior and churn patterns.

blind-source-separation

This project focuses on blind source separation, separating mixed images into components using a deep learning network. The network takes in the sum of the images and outputs predicted components, evaluated using mean squared error. No preprocessing is allowed, demonstrating the effectiveness of blind source separation.

toxic-comment-classification

This repo contains code for toxic comment classification using deep learning models based on recurrent neural networks and transformers like BERT. The goal is to detect and classify toxic comments in online conversations using Jigsaw's Toxic Comment Classification dataset.

alessiococchieri

connecting-rods-inspector

Software system aimed at visual inspection of motorcycle connecting rods. The system should be able to analyze the dimensions of two different types of connecting rods to allow a vision-guided robot to pick and sort rods based on their type and dimensions.

home-price-prediction

Machine learning model in order to correctly predict the prices of Banglore houses based on different features.

POS-tagging

This GitHub repository contains code and results for a Part-Of-Speech (POS) tagging task addressed using recurrent neural networks and making use of Glove's embeddings. Starting from a baseline model, then we explored three possible variations in order to improve upon the baseline model's performance.