Deep Learning

In this paper, we propose a novel loss function, named Residual-based Adaptive Huber Loss (RAHL), which adds a learnable residual to the delta hyperparameter from Huber Loss, enabling a model to adapt based on the distribution of errors in the data.

This project aims to investigate and propose methods to reduce the computational and human burden of deep learning in order to enable the development of more powerful Autonomous Computational Intelligences (ACIs).

Do you want to know more about using unknown samples to tighten the classification boundaries? Check our paper “Tightening Classification Boundaries in Open Set Domain Adaptation through Unknown Exploitation”, just accepted for publication in the SIBGRAPI - Conference on Graphics, Patterns and Images.

In this paper, we addressed the open-set domain adaptation (OSDA) problem. We propose three new approaches to improve OSDA methods by using unknown samples to tighten the classification boundaries of the model.

In this paper, we addressed the multi-domain learning problem. We propose to prune a single model for multiple domains, making it more compact and efficient. To do so, we encourage multiple domains to share convolutional filters, enabling us to prune weights not used by any of the domains at test time.

Do you want to know more about pruning models for multiple domains according to a user budget? Check our paper “Budget-Aware Pruning for Multi-Domain Learning”, just accepted for publication in the International Conference on Image Analysis and Processing.

In this work, we introduce a novel the Biome Cerrado Dataset (CerraData), a large database with a total of 2.5 million tiles of 256x256 pixels, obtained from 150 scenes made by the Wide Panchromatic and Multispectral Camera~(WPM) of the China-Brazil Earth Resources-4A~(CBERS-4A) satellite.

Are you looking for a robust dataset for studying the Brazilian Cerrado vegetation? Check our paper “A High-Spatial Resolution Dataset and Few-shot Deep Learning Benchmark for Image Classification”, just accepted for publication in the SIBGRAPI - Conference on Graphics, Patterns and Images.

In this work, we propose three new approaches in the context of DML. We are particularly interested in NNGK due to its robustness and simplicity. As such, we introduce variants that take advantage of Mixup to cope with metric learning in incomplete supervision scenarios.

Do you want to know more about metric learning in incomplete supervision scenarios? Check our paper “Mixup-based Deep Metric Learning Approaches for Incomplete Supervision”, just accepted for publication in the IEEE International Conference on Image Processing.