user_image

Jeova Farias Sales Rocha Neto

I am an Assistant Professor of Computer Science at Bowdoin College. Before that, I was a Visitor Professor in the Computer Science Department at Haverford College. I received a PhD in Computer Engineering and a Masters in Applied Mathematics from Brown University in 2021, where I was advised by Prof. Pedro Felzenszwalb. In 2015, I earned a Masters degree in Computer Science at University of Nice-Sophia Antipolis under Prof. Marc Antonini and prior to that I completed my Bachelors degree in Computer Engineering at Federal University of Ceara (UFC) under Prof. Fátima Medeiros.

I am broadly interested in unsupervised problems in Computer Vision and Machine Learning. In my current research, I am interested in using discrete optimization, statistics and deep learning to tackle Image Segmentation and Clustering. Previously, I worked on machine learning problems involving Probabilistic Graphical Models and Semidefinite Programming. In my undergrad years, my research focus was on Synthetic Aperture Radar and Biomedical Image Segmentation using Level Sets.

EDUCATION

  • 2016 – 2021

    PhD in Computer Engineering + M. Sc. in Applied Mathematics

    Brown University (Providence, USA)

  • 2014 – 2015

    M. Sc. in Computer Science

    University of Nice-Sophia Antipolis (Nice, France)

  • 2011 – 2015

    B. Eng. in Telematics Engineering

    Federal University of Ceará (Fortaleza, Brazil)

RESEARCH INTERESTS

  • Statistical Learning for Image Segmentation.
  • Intepretable Spectral Clustering and Image Segmentation.
  • Fast Methods for Data Clustering.
  • Usupervised Deep Learning solutions for Image Segmentation and Clustering.
  • Applications level sets in Biomedical and SAR imagery.
  • Discrete Optimization using Graph-Cuts.

PROJECTS

Level Sets for Image Segmentation
Improving level set methods for image segmentation using local image statistics.

Statistical Color Model Estimation for Segmentation
Directly estimating segmentation appearence models from images using statitical techiniques, such as Non negative Matrix Factorization.

Spectral Clustering and Image Segmentation
Improving the classical spectral algorithms for real applications.

Statistical Learning on SAR Imagery
Using statistical techinques to improve SAR image understanding and segmentation.

Deep Learning for Data Clustering
New techinques for Deep Learning-bases clustering using Graph Cuts, CNNs and Transformers.

Deep Learning for Supervised and Unsupervised Image Segmentation
Creating new architecures and training startegies that leverage local and global image contents for both supervised and unsupervised segmentation.

Psychological Perpectives in Deep Image Generation
Study of how certain patterns, such as biases, emerge in modern image generation algorithms.

Segmentation from Superpixels
Utilizing statistical and graphical modeling for segmentation using superpixels.

Techiniques to Efficiently Measure Student Performnace
Finding algorithms that impronte the classical estimators for Item Respose Theory

RECENT PUBLICATIONS

[All Publications]
  1. Patch-Based Deep Unsupervised Image Segmentation using Graph Cuts
    Wasserman, Isaac, and Farias Sales Rocha Neto, Jeova
    2024
  2. Spectral Clustering of Categorical and Mixed-type Data via Extra Graph Nodes
    Soemitro, Dylan, and Farias Sales Rocha Neto, Jeova
    2024
  3. A Sparse Graph Formulation for Efficient Spectral Image Segmentation
    Palnitkar, Rahul, and Farias Sales Rocha Neto, Jeova
    2023