I’m a computer engineer and scientist from the greenish Costa Rica. I’m interested in Deep Learning, Computer Vision, Natural Language Understanding and Biodiversity Informatics. I enjoy both academia and industry, science and engineering.
During my doctoral studies I worked mainly on automatic plant identification using computer vision and deep learning techniques, together with researchers from CIRAD and INRIA. I also worked on development of hierarchical architectures/loss functions to deal with taxonomic hierarchies of classes in the plant domain. We also pioneered work with herbarium institutions images, which have opened a new research area towards domain adaptation and overall using a new media type for plant species automatic identification and conservation.
In the industry, besides working as a Software Engineer, I have worked in conversational AI creation using NLU models, mostly language models in English and Japanese, using open-source tools like RASA, Tensorflow/Pytorch, as well as text classifiers, sequence classification and named entity recognition models.
Additionally, I have done research on generative models (VAE / GANs) for multiple image enhancement, to produce e-Commerce related images, optimized for Click Though Ratio among other things.
My current work at Microsoft is focused on explainability of business phenomena using machine learning, as well as explaning bias in populations when regressions arise. In the academy I keep doing research on automatic plant identification, but now focused on generative models and self-supervision to tackle imbalance. I advise several master students on related topics for their thesis.
My main current interests are focused on self-supervision, data imbalance, image inpaiting and reconstruction, especially using noise to map multiple reconstruction behavior.