Luciano Di Palma is a Ph.D. student in Machine Learning and Data Management at CEDAR, a joint research team between École Polytechnique and Inria Saclay. His main research interest is designing scalable, intelligent analytics systems with strong theoretical performance guarantees and leveraging state-of-the-art machine learning and statistical techniques.
Ph.D. in Active Learning and Data Exploration, 2017–2021
Ecole Polytechnique and Inria Saclay
M.Sc. in Data Science, 2016–2017
Université Paris-Saclay
Cycle Ingénieur Polytechnicien, 2014–2016
École Polytechnique
B.Sc. in Physics, 2011–2013
University of São Paulo
In this thesis, we propose a human-in-the-loop framework for efficient model development over large data sets. In this framework, we …
During my time at Wildlife, I was part of their Data Engineering team, which is responsible for all data-related tasks in the company. More precisely, I had the opportunity to work on several impactful projects, including the:
Design and implementation of a data pipeline for real-time processing, storage, and visualization of several metrics of interest, such as the number of app crashes, new installs, and active online users.
Installation, configuration, and benchmarking of the company’s first Spark cluster, enabling engineers to run data processing tasks and train Machine Learning models over terabytes of user data.
Analysis of the number of impressions, clicks, and installs of users across multiple games and AdNets, leading to an improvement in user targeting and an increase in the installs-to-impressions ratio.
Applied concepts of Reinforcement Learning and Neural Networks techniques to develop an Adaptive Traffic Lights controller capable of efficiently choosing good control options for vehicle flux improvement. More precisely, I have:
Surveyed the State-of-the-Art in Neural Networks and Reinforcement Learning.
Designed and trained an adaptive traffic lights controller capable of adapting the distribution of green time over different lanes with varying vehicle throughput. The controller was trained using Reinforcement Learning techniques and Neural Networks.
Empirically proved that our adaptive controller is more efficient than fixed-time controllers in managing vehicle traffic, achieving a 25% reduction on the average number of queued vehicles.