Carlos Ocanto Dávila Email & Phone Number

Paris, Île-De-France, France

Collaborated with the Journal Le Monde on a series of four articles (including two front pages in the printed version) that studied the weight of the french political tendencies on Twitter : Segmented the french twittersphere into clusters of users using state-of-the-art Machine Learning, network analysis, and NLP methods.Developed a new initialization method, and model selection criteria for the Machine Learning model behind the platform linkage.fr, which combines topic modeling and graph analysis. This resulted in a 46% improvement of the node ARI, and a 78% increase in the model selection accuracy in the most challenging simulated scenarios, while reducing the computation time in half. Designed production-ready C++ and Python scripts currently used in Machine Learning pipelines and critical tasks in the platform.Collaborated with stakeholders from different backgrounds to implement use-cases of the Linkage technology that addressed real-world issues using Machine Learning Methods. Produced meaningful visualizations that summarized the results of the AI methods for a wide audience.Studied State-of-the-art methods for graph and text analysis. Ultimately designing a new online method to explore dynamic data streams of nodes arriving at a network.

Mougins, Provence-Alpes-Cote D'Azur, France

Created an NLP model for password security assessment using Deep/Statistical Learning methods. This method outperforms state-of-the-art reference tools: More than 10% increase in detection rate.Clustered texts using Deep Learning methods in TensorFlow/PyTorch, while leveraging features obtained from language-specific ontologies based on WordNet. Researched state-of-the-art techniques for Natural Language Processing, and password security.Published a paper at an international per-reviewed conference (ICISSP). A patent in the United States is pending approval.Designed the entire pipeline of data ingestion, preprocessing, and inference, of an innovative system for assessing password security using AI currently used by SAP's software security research team.Presented the results of the project to audiences of different backgrounds at SAP. Adapted the content to display the information in the most meaningful way possible for the specific public.

Nice Area, France

Implemented a Learning-to-Rank (LTR) architecture (ListNet) in Keras that outperformed FranceLabs' existing solution by almost 10 % (NDCG), and is currently in use by the company's products.Created an internal LTR dataset. Extracted features based on those present in the industry-reference Microsoft's LETOR4. Added new columns using traditional NLP metrics, and query-document similarities based on Fasttext embeddings.Studied the State of the Art in Learning-To-Rank Applications.

Madrid Area, Spain

Bachelor's Final Project Internship at UC3M's Intelligent Systems Laboratory:Studied the pose estimation quality of stereo-vision methods using the predictions of machine learning models. Created a detection, tracking and pose estimation application using HOG+SVM, Kalman filters and Stereo Cameras (reaching an 8% F1 improvement).Integrated diverse ROS machine learning modules to be used in an autonomous vehicle.

Master 2- Master Of Science, Data Science, Highest Honors - Mention Très Bien

http://univ-cotedazur.fr/en/idex/formations-idex/data-science/ Semester 1: Bayesian Learning Deep Learning Statistical Analysis of Graphs.

Master 1- Master Of Science, Data Science, Highest Honors - Mention Très Bien

http://univ-cotedazur.fr/en/idex/formations-idex/data-science/ Semester 1: Refresher Statistical Inference Introduction to Data Mining.

Bachelor Of Engineering - Be, Electrical, Electronics And Communications Engineering