Wesley Maia

Wesley Maia

M.Sc Student at USP

About Me

I am a Master’s student in Mathematics, Statistics, and Computing at the University of São Paulo, advised by Sergio A. David. My research focuses on applying machine learning techniques to solve problems in Sign Language.

With a strong background in Data Science and Machine Learning, I have contributed to several projects involving the application of deep learning and natural language processing (NLP) models to solve real-world problems. Currently, I am working at Hand Talk, the world’s largest platform for automatic translation in Sign Languages, where I lead efforts to enhance accessibility through AI-driven solutions.

Academic Background:

  • Master of Science (MS) in Mathematics, Statistics, and Computing, University of São Paulo, 2022 - Present
  • Master of Business Administration (MBA) in Data Science & Analytics, University of São Paulo, 2021 - 2022
  • Bachelor of Science (BS) in Physics, University of São Paulo, 2013 - 2016

Professional Experience:

  • Data Scientist, Hand Talk (Jun 2021 – Present)
    Conducted research on models for translating text to sign language without glosses, developing and optimizing Sign Language Translation (SLT) models to increase translation performance. Automated 3D animation processes using generative AI and designed machine translation models to enhance language support and accuracy. Led research on novel model architectures, resulting in a 15% improvement in machine translation performance, and implemented computer vision techniques for accessibility features to expand reach to visually impaired users. Streamlined data workflows by integrating cloud-based services to ensure data integrity and efficiency, while utilizing model tracking and experimentation tools in MLOps to accelerate the AI development lifecycle.

Publications

Automatic Sign Language to Text Translation Using Mediapipe and Transformer Architectures

Wesley Ferreira Maia, António M. Lopes, Sergio David
This paper explores the automatic translation of sign language into text using a combination of MediaPipe and Transformer architectures. The proposed method leverages the keypoints extracted by MediaPipe and applies Transformer-based models to recognize sign language gestures and convert them into natural language text. The research highlights the potential of integrating computer vision techniques with deep learning models to bridge the gap between sign language and written language, ultimately enhancing accessibility for the deaf and hard-of-hearing community.
Preprint Available at: SSRN Link

Multi-level Product Category Prediction through Text Classification

Wesley Ferreira Maia, Francisco Louzada Neto, & cia This article investigates applying advanced machine learning models, specifically LSTM and BERT, for text classification to predict multiple categories in the retail sector. The study demonstrates how applying data augmentation techniques and the focal loss function can significantly enhance accuracy in classifying products into multiple categories using a robust Brazilian retail dataset. The LSTM model, enriched with Brazilian word embedding, and BERT, known for its effectiveness in understanding complex contexts, were adapted and optimized for this specific task.
The results showed that the BERT model, with an F1 Macro Score of up to 99% for segments, 96% for categories and subcategories, and 93% for name products, outperformed LSTM in more detailed categories. However, LSTM also achieved high performance, especially after applying data augmentation and focal loss techniques. These results underscore the effectiveness of NLP techniques in retail and highlight the importance of the careful selection of modeling and preprocessing strategies. This work contributes significantly to the field of NLP in retail, providing valuable insights for future research and practical applications.
Preprint Available at: arxiv Link

Machine Learning Applied to Estate Pricing for Residential Rentals in Dynamic Urban Markets - the Case of São Paulo City

Wesley Ferreira Maia, Sergio A. David
This study delves into real estate rental pricing in São Paulo using advanced machine learning techniques, geospatial analysis, and NLP. The research involved a rigorous preprocessing of 35,486 instances, integrating textual and geographic data for comprehensive analysis. By comparing various regression models, including a novel Blending model, the study achieved a notable RMSLE of 0.2923. These findings offer landlords and tenants a data-driven decision-making tool for real estate pricing and contribute to informed decisions in São Paulo’s dynamic rental market. Available at: Engineering Analysis with Boundary Elements

Academic Service and Teaching

I served as a Physics, Mathematics, and Computer Science teacher at Colégio Estrutural, where I designed and taught lessons for high school students, developed comprehensive curriculums, and introduced students to programming through hands-on projects.

I also worked as a Teaching Assistant at the Institute of Physics, University of São Paulo (IFUSP), supporting professors in the Physics department, conducting one-on-one tutoring, and organizing group study sessions to assist students in understanding complex topics like Electricity and Magnetism.