José Lucas De Melo Costa
1, rue Joliot Curie, Gif-sur-Yvette, 91190, France
✉️ jose-lucas.de-melo-costa @student-cs.fr
📱 +33 07 66 17 00 82
Education
2021 - 2023
Engineer's degree (Double degree) - Major in artificial intelligence
CentraleSupélec, University of Paris-Saclay - Paris, France
2018 - 2023
Computer engineering
University of São Paulo - São Paulo, Brazil
Experience
Machine learning intern
April 2023 - September 2023
French Atomic Energy Commission (CEA) - Saclay
  • Project: Study of the embeddability of attention-based networks
  • Task: Search and explore compression methods for visual transformers (ViT)
  • Action: Quantization and knowledge distillation of tiny ViT
  • Result: New quantization scheme, integer-only ViT; Work presented at 2024 HiPEAC workshop
Technical skills:   python, attention-based networks, model training
Research intern
June 2022 - Aug. 2022
University of Paris (Université Paris-Cité) - Paris
  • Situation: Unsupervised anomaly detection on medical time series data
  • Task: Create a method to detect anomaly on multivariate data
  • Action: Proposal of a new loss function for the latent space using autoencoders
  • Result: New graph-based method time series embedding; research skills acquisition
Technical skills:   python, autoencoders, custom losses on pytorch, optimization
Software Engineering intern
Jan. 2021 - June 2021
Vigilantes do sono, "Vigilantes do Sono" - Brazil
  • Situation: Startup that uses AI to help treat insomnia
  • Task: Right hand of the CTO, full stack developer
  • Action: Data processing, application development
  • Result: Launch of the mobile application, 35 thousand users
Technical skills:   typescript, react, nodeJS, react native
Research intern
April 2020 - Dec. 2020
Petrobrás 🔗, University de São Paulo - São Paulo, Brazil
  • Situation: Project in signal processing and machine learning; Working with Petrobrás’ research team
  • Task: Solution to mooring line failures (anchoring) of offshore production stations
  • Action: Machine learning pipeline to classify the status of stations
  • Result: New natural frequency-based approach for the problem; international conference 🔗
Technical skills:   python, signal processing (wavelets, STFT), machine learning algorithms (SVM, LSTM)
Teacher assistant
Feb. 2019 - Dec. 2019
University de São Paulo - São Paulo, Brazil
  • Situation: Teaching assistant in "Algorithms and data structures" and "Systems programming"
  • Task: To help students through these disciplines
  • Action: Exercise solving course + homework correction
  • Result: Communication skills, ability to explain
Technical skills:   python, C, algorithms, data structures, problem solving
Publications
Conferences:
  1. Costa, José Lucas De Melo et al. "FPSO Mooring Line Integrity Supervising System Based on Motion Data and Natural Frequency Estimation." Proceedings of the ASME 2021 40th International Conference on Ocean, Offshore and Arctic Engineering. Volume 1: Offshore Technology. June 21–30, 2021. ASME. https://doi.org/10.1115/OMAE2021-62991
  2. Costa, José Lucas De Melo et al. "Spectral Graph-Based Networks for Mooring Line Failure Detection on Fpso." Under review, 2024.
Workshops and presentations:
  1. Costa, José Lucas De Melo, Cyril Moineau, Thibault Allenet, Inna Kucher. "Centered Kernel Alignment for Efficient Vision Transformer Quantization." Oral presentation at 6th Workshop on Accelerated Machine Learning (AccML), co-located with the HiPEAC 2024 Conference.
  2. Costa, José Lucas De Melo, et al. "Learning And Understanding Strategies On Zero-Sum Games." Oral presentation at Junior Multidisciplinary Congress, Université Paris-Saclay, 2022.
  3. Costa, José Lucas De Melo, et al. "Detection of Failures in the Anchoring System of Floating Platforms by Natural Frequency of Oscillation." Presentation at the 28th International Symposium on Scientific and Technological Initiation, University of São Paulo (SIICUSP), 2020.
Projects
Neuro Spectral Supervisor (NeSS) - University of São Paulo
Sept. 2023 - Jan. 2024
Spatial-temporal graph neural network applied to mooring line failure detection of oil platforms
Hopular Applied to Fraud Detection - LUSIS, CentraleSupélec
Sept. 2022 - April 2023
Project to identify fraud in credit card transactions using modern Hopfield networks.
Chess & Learning - CentraleSupélec
Jan. 2021 - June 2021
Study the explicability of machine learning algorithms on zero-sum games. The work was presented at the ENS-Saclay multidisciplinary junior congress🔗