Hi, my name is Gerard Caravaca
I'm the Artificial Intelligence Engineer.

This thesis presents a deep learning-based system for detecting transportation modes using smartphone sensors in Barcelona, focusing on algorithm development, data preprocessing, and real-time pattern prediction for urban planning. It introduces a hierarchical model combining CNNs and LSTMs for efficient and battery-friendly detection, supported by an Android application for data collection, while addressing ethical and privacy considerations. (Paper in review process)

This system developed for the Hospital Clínic de Barcelona is part of my final degree thesis. It is a system that aims to compare cancer data in different European countries with pollution data. The ultimate goal of the system is to allow the hospital's research group to find similarities between the increase of cancer cases and the increase of polluting gases in Europe. This thesis is part of the work presented in the paper: OA13.04 Prevalence of Molecular Alterations in NSCLC and Estimated Indoor Radon in Europe: RADON EUROPE Study.

CarPlateDetection is a project focused on the detection of license plates in images using a fine-tuned YOLO (You Only Look Once) model. This project aims to provide a high-performance solution for license plate recognition tasks by leveraging a dataset specifically tailored for this purpose.

Critical analysis of the work developed by Ismkhan, H in the paper: I-k-means-+: An iterative clustering algorithm based on an enhanced version of the k-means. Including implementation and testing.

Exploratory analysis of DL segmentation in fashion using orcnet models and SAM.

Analysis and implementation of the Twin Delayed Deep Deterministic Policy Gradients (TD3)( Fujimoto et al., 2018) RL algorithm.

Evaluation and summarization of some Transfer Learning state of the art papers applied to Reinforcement Learning.

Implementation and evaluation of the PRISM algorithm introduced by Jadzia Cendrowska as an approach for the induction of modular rules from data.

SemEval (Semantic Evaluation Exercises) are a series of workshops which have the main aim of the evaluation and comparison of semantic analysis systems. The data and corpora provided by them have become a ’de facto’ set of bench-marks for the NLP comunity. STS is also known as paraphrases detection. A pair of texts is a paraphrase when both texts describe the same meaning with different words. This project is a mock participation in the 2012 SemEval for phrase similarity.

The aim of this project is to create a complete method for training artificial neural networks using genetic algorithms (GAs) and evolutionary algorithms (EAs) as alternatives to the classical backpropagation approach. In this project, me and my classmate explored the use of GAs and EAs to optimize ANNs in various ways, including the evolution of connection weights, architectures, hyperparameters, activation functions, and learning rules.