Jogi Suda Neto, MSc

Jogi Suda Neto, MSc

Co-founder @ QuaTI and Machine Learning Engineer

I’m a Computer Scientist with a MSc at the São Paulo State University (UNESP), where I studied Quantum Machine Learning (QML). Professionally, I also work as a Machine Learning R&D Engineer at the CK-12 Foundation.

Academic bio & interests

As an undergrad, I first started working with classical ML for keratoconus subclinical detection. In my masters, I moved on to QML, still motivated by the medical field - I showed how quantum models can be used to detect Parkinson’s disease through the speech. After this, I progressively grew interest in the idea of group-equivariant/invariant QML. If you wonder what that is, think of the classical case of CNNs: we know that kernel parameter sharing naturally causes translation equivariance, while pooling gives invariance. We now have the same for parameterized quantum circuits: this area is being known as Geometric QML (GQML)… and it is very exciting! After all, which kinds of symmetries may be naturally incorporated into quantum models, showcasing a better inductive bias over classical architectures? Although we don’t have an exact answer, there is an exciting, and potential application for this: Particle Physics!

I am very honored to have been advised by both Prof. Rodrigo Capobianco Guido and Prof. Felipe Fanchini, both of which have greatly helped me grow in my path.

Awards

★ Deep Knowledge - Awarded for the contributions on Quantum Natural Language Processing - NTT Data

★ Meaningful Innovation - Awarded for motivating the Quantum Computing initiative - NTT Data.

🏅 Gold medal - Brazilian Astronomy Olympiad - Brazilian Astronomy Olympiad 2015

🥉 Bronze medal - Brazilian Physics Olympiad - 2015

🥉 Bronze medal - Brazilian Physics Olympiad - 2014

🥈 Silver medal - Brazilian Astronomy Olympiad - 2014

🏆 Selected to the final stage to represent the Brazilian Team in the International Olympiad of Astronomy and Astrophysics - 2015

GitHub projects

  • Quantum Sentence Transformer: This project aims to shed light on Quantum Natural Language Processing. The model leverages parametrized circuits and hybrid classical-quantum transfer learning to come up with meaningful sentence representations. All code uses Pennylane and Sentence Transformers.
  • Discrete Path Transform: code for the DPT, developed jointly with the Signal Processing lab at UNESP.
  • ParaconsistentLIB - paraconsistent logic-based feature engineering.
  • pyGAMESS-DS - automated retrieval of main molecular informations from GAMESS-US, like geometry-optimized coordinates, Gibbs energy corrections, net charges, etc.