Research projects

Deep Learning is an essential field of Machine Learning based on computer algorithms for data representations at multiple abstraction levels. These representations involve a hierarchy of features or concepts where higher-level representations are defined from lower-level ones. The same lower-level representations help define higher-level ones. Deep learning uses artificial neural networks with many layers and large datasets to training algorithms on how to solve perceptual problems, such as detecting recognizable concepts in data, translating or understanding natural languages, interpreting information from input data, and more. Deep Artificial Neural Networks have recently won numerous contests in Pattern Recognition and Machine Learning.

The Deep Artificial Neural Networks literature and applications actually represent the best recognition performance in different Pattern Recognition fields, including computer vision, image recognition, speech recognition, natural language processing, text processing, information retrieval, and multimodal information processing. Despite the significant progress and results, many challenges remain, aspects like large-scale parameter optimization, modeling of temporal data with long-term dependencies, generative modeling, efficient Bayesian inference, multimodal data and models, and learning representations needs more contributions.

Our overarching goal is to gather the recent advances, ongoing developments, and the road that lies ahead of the field of Deep Learning in Artificial Neural Networks. The suggested research areas are (but not restricted to):

  • Deep Networks Foundations: Initialization, Optimization, and Regularization
  • Deep Learning for Tabular Data and Time Series
  • Convolutional Neural Networks: ResNets, DenseNets, EfficientNets, etc
  • Computer Vision: Detection, Segmentation, and Advanced Tasks
  • Recurrent Neural Networks: LSTM, GRU, QRNN, etc
  • Sequence-To-Sequence, Attention, Self-Attention, and Transformers
  • Natural Language Processing and Language Models
  • Generative Models: Generative Adversarial Networks, etc
  • Self-Supervision, Semi-Supervision, Graph Neural Networks, and Advanced Topics
  • Neural Architecture Search and AutoML
  • Meta-Learning, One-Shot, and Few-Shot Learning
  • Limitations and Challenges: Out-of Distribution Detection and Adversarial Robustness

Previous projects (2022.2)

  • Abordagem de Aprendizagem de Maquina para classificar por assunto as solicitações de usuários do aplicativo TJPE Atende
  • Aplicando Visual Transformers em Literatura Japonesa Classica
  • Detecção inteligente de ataques DDoS utilizando técnicas de aprendizado profundo
  • Técnicas de interpretabiliadade aplicadas à predição de séries temporais de Nowcasting
  • Tuning “Color-aware two-branch DCNN for efficient plant disease classification” to embedded systems
  • Composite Feature Selection Using Deep Ensembles
  • Rastreamento em Multiplas C ́ ameras
  • Prediction of a Ship’s Parametric Balance Using LSTM and Optimized Input Vector Space
  • Modelo de Regressao Embarcado para Calculo do Ângulo Solar Zenite
  • Detecção de Áreas em Imagens de Satélite Utilizando Segmentação Semântica
  • Séries Temporais com LSTM para dados sociais: uma análise empírica e descritiva
  • Avaliação de desempenho de uma rede neural Long Short Term Memory para previsão de preço do Bitcoin
  • Um estudo de caso sobre o uso de redes neurais com privacidade diferencial no dataset FER-2013
  • Physics-Informed Neural Network for Reliability Assessment of Multi-State Systems
  • Leveraging PETSCIIBOTS with Generative Models
  • Uma abordagem de meta-learning para inicializar o metodo FEDformer
  • Redução de Disparidade de Robustez em Deep Learning

Previous projects (2022.1)

  • Pose-oriented personal protective equipment detection
  • Geração de Arte Utilizando IA: Uma Análise Qualitativa
  • Estimação Monocular de Profundidade Utilizando MobileViT
  • Data Augmentation for Zero-Reference Deep Curve Enhancement of Images in Gray Scale and for Multiple Face Detection in Dark Images
  • Análise de Modelos Tabulares como Sistemas NIDS utilizando Agregação de Pacotes
  • Improving Human Activity Recognition Through Wearable Devices With Synthetic Data
  • Handover Mechanisms Analysis Based on Deep Learning for UAV Networks
  • Transformando pinturas de rostos em imagens realistas
  • Comparação de performance entre métodos de avaliação da qualidade de imagens
  • Improving Inference Time and Model Size of a 2D-CNN Based Intrusion Detection System for Automotive Ethernet for CPU devices
  • Estratégias de Otimização do BERT para Reconhecimento de Entidades Nomeadas na Língua Portuguesa
  • A comparison of machine learning recommendation liraries using the Steam game dataset
  • Um comparativo entre variantes do modelo LSTM na previsão de séries temporais da geração de energia eólica
  • CNN-BERT Hyperparameter Tuning in hate speech detection

Previous projects (2021.2)

  • DS-PTAM: um sistema SLAM com extração de características utilizando redes profundas
  • Vision Transformer (ViT) usado para Calibração Automática de Parâmetros Intrínsecos de Câmeras
  • Identifying Inconsistent Test Descriptions Using Deep Learning
  • Unsupervised Multi-View Multi-Person 3D Pose Estimation using Reprojection Error - DOI
  • Análise da influência do espaço de cores na tarefa de Super Resolução
  • Uma abordagem utilizando aprendizagem profunda para classificação de fragmentos genômicos de vírus que infectam humanos
  • DETRAug: Adaptação de Técnica de Processamento para Object Detection
  • Entanglement block optimization for quantum generative adversarial networks
  • Audio Retrieval Using Cross-Modal BERT Rerank for Natural Language Queries
  • Avaliação do modelo Pixel2Style2Pixel (pSp) sobre imagens com baixas resoluções
  • Uma Análise Comparativa de Métodos de Aprendizagem Profunda e Rasa para Classificação de Receitas Culinárias
  • LogBERT-BiLSTM: to Detect Malicious Web Requests - DOI
  • Previsão de vazão em diferentes localidades com um único modelo Informer
  • Aplicação do Modelo TED-net à Base de Dados LoDoPaB-CT Via Fine-Tuning
  • Detecção de commits indutores de defeitos de Software usando Transformers (Tentando Bracis 2022)
  • Revisao de redução de dimensionalidade baseada em variational autoenconders para classificação não-supervisionada de dados de alta dimensionalidade
  • Seleção e Combinação de Técnicas de Representação de Características e Classificadores para a Detecção de Fake News
  • Implementação de LSTM Univariável e Multivariável em Microcontrolador

Previous projects (2020.2)

  • DenseMAP: A Denselly Connected Modularized Processing Adaptative Neural Network
  • Instance Segmentation using Mask R-CNN using Self-Attention Network as Main Object Detection Backbone
  • PictoBERT: Pictogram Prediction using Transformers - DOI
  • Hyperparameter Search Approaches for Deep Learning applied to Tabular Data: A Comparative Study - DOI
  • Aplicação Deep Learning em dados tabulares de RT-PCR para detecção mais rápida de COVID-19
  • Intrusion Detection System with TabNet
  • Multi-cue adaptive emotion recognition
  • Crop Disease Identification Using Multi-label Residual Network
  • Deep Contrastive Learning Approaches for Offline Signature Verification Problem
  • MOSNet V2: Transformer-based Objective Assessment for Voice Conversion
  • Reinforcement Learning to Soccer Skills
  • InceptionTime for Gas Classification
  • Classificação de gases mistos através de um modelo Inception Time
  • Previsão de Geração Eólica utilizando AR-Net
  • Usando a Função AM-Softmax para o Problema de Reconhecimento de Autor em Textos Manuscritos Curtos
  • Self-calibrated U-Net for Document Segmentation
  • Learning What, Where and Which to Transfer
  • Identificação de assinatura de odores de colônias de microrganismos usando InceptionTime - DOI
  • Hierarchical Attentional Temporal Convolutional Networks for Document Classification

Previous projects (2019.2)

  • KutralNet: A mobile deep model approach - DOI
  • Comparativo Entre as Redes Neurais Faster-RCNN Inception V2, RFCN Resnet101 e SSD MobilenetV2 para o problema de Detecção de Ovos do Mosquito Aedes Aegypti em imagens
  • Classificação desbalanceada com subconceitos e sincronização de séries temporais no monitoramento ́remoto de plantações
  • Multi-Label Image Recognition With Simple Graph Convolution
  • Learning on Soccer Cooperative Environment
  • Squeezed Deep 6DoF Pose Detection Using Knowledge Distillation - DOI
  • Conditional GAN for missing data imputation
  • BERT-Side – Using BERT for Distantly-Supervised Neural Relation Extraction with Side Information - DOI
  • Multiclass financial fraud detection by integrating supervised learning with adversarial autoencoders - DOI
  • Table Retrieval: Matching Articles and Web Tables
  • Personal protective equipment detection in industrial facilities using camera video streaming
  • Benchmark de Políticas de Learning Rate e Proposição de Uma Política Baseada em Decaimentos com Feedback
  • OctHU-PageScan: A Fast Fully Octave Convolutional Neural Network for image segmentation - DOI
  • Deep Fall Detection - DOI
  • Modelo de Regressão utilizando Redes Neurais Convolucionais aplicado a Sistemas de Potência
  • Fast Unsupervised Cyber-Attack Detection for Time Series with Generative Adversarial Networks - DOI
  • Uso do YOLO na detecção de objetos na RoboCup SPL
  • Direct Random Target Projection as Feedback Alignment algorithm with layerwise feedforward training for Deep Learning Architecture
  • Quality Assessment of In-the-Wild Videos: A New Approach
  • SincNet com Aprendizado Adversário para Identificação de Locutores
  • Detecting insulators in power transmission lines: A comparison study between SSD, YOLOv3, and Faster R-CNN
  • Aprendizagem profunda aplicada à Agricultura de Precisão – Estimativa do peso vivo de suínos
  • DQN-AF: Proposta de uma Estratégia de Encaminhamento Adaptativo baseada em Deep Q-Network para Redes NDN
  • Analise da Ferramenta Spleeter: Um estudo de caso com proposta de melhorias

Previous projects (2018.2)

  • Reduced Very Deep Convolutional Neural Network for Text Classification - DOI
  • Single Image Super-Resolution Using A Multi-Scale Generative Adversarial Network
  • Towards Optimizing Convolutional Neural Networks for Robotic Surgery Skill Evaluation - DOI
  • Speaker Recognition With SincNet-N in Noisy Environment
  • Deep Voice Style Transfer with CycleGAN
  • Capsule Network for Anomaly Detection
  • Dbpedia Entities Embeddings from Abstracts Wikipedia
  • Rastreamento de Poses Humanas a Partir de Imagens RGB - DOI
  • Image Generation from Text Embeddings with Multiple Attention Sources - DOI
  • Convolutional Neural and Hierarchical Attention Networks for document classification
  • WGANs to ECG Data Augmentation in Heart Beat Anomaly Detection by 2D-CNN - DOI
  • Otimizaçãoo do Consumo de Energia em Redes Ad Hoc Aloha Empregando Deep learning - DOI
  • AM-SincNet - DOI
  • Using Attention Mechanisms to Improve ULMFiT - DOI
  • Residual Gaussian Denoiser with Adversarial Loss
  • A deep learning approach to predict remaining useful life of systems based on vibration signal
  • Semantic Segmentation for People Detection from Beach Images
  • CNN and LSTM applied to the identification of urgent postings in asynchronous discussions in MOOCs
  • Deep Shading: Convolutional Neural Networks for Screen Space Shading

Previous projects (2017.2)

  • Extracting human attributes using a convolutional neural network approach
  • Learning to generate reviews and discovering sentimento
  • Unsurpevised LSTM in handwriting recognition
  • Using Depthwise Separable Convolution to Reduce Time Processing on SqueezeNets - DOI
  • CNN implementation of model VGG and ResNet in OpenCL Multiplatform
  • Multiclass anomaly detection using deep learning and one-class SVM
  • FaceNet: A Unified Embedding for Face Recognition and Clustering
  • Computing the stereo matching cost with a convolutional neural network
  • AutoLabeling - An Automatic Labeling of text corpus in Brazilian Portuguese for the Sentiment Analysis
  • A Convolutional Neural Network approach for Speech Quality Assessment
  • Learning Audio Sequence Representations for Acoustic Event Classification
  • Devise: A Deep Visual-Semantic Embedding Model
  • Question pairs duplicate detection
  • Deep Learning for Sentiment Analysis on Movie Reviews
  • A evaluation of LBP and your variants with shallow calssifiers and deep convolutional neural network
  • Reading Wikipedia to Answer Open-Domain Questions
  • Deep Learning for steganalysis of spatial domain steganography
  • A comparison between batch normalization and dropout for Natural language processing in sentiment analysis
  • SegNet: A Deep Convolutional Encoder-Decoder Architecture for Image Segmentation - DOI

Previous projects (2016.2)

  • Activation Functions for Deep Learning: A Comparative Study using Statistical Tests - DOI
  • Reconhecimento de autores e fraudes em manuscritos utilizando Deep Learning
  • Análise para Seleção de Gradientes e o Processo de “Desaprendizagem para Aprender”
  • The Impact of Datasets Complexity and Diversity on Transfer Learning over Convolutional Neural Networks - DOI
  • Breaking Hash Passwords using Deep Learning
  • Exploring CNNs for Reduced Manually Fed Training Datasets for Text Based CAPTCHA Recognition
  • Uso de diferentes modelos de cores em redes neurais convolucionais para detecção de bordas - DOI
  • RNN-LSTM para Previsão de Séries Temporais
  • Report of project in deep learning: 2DPCANet
  • Detecção de anomalias baseada em autoencoders convolucionais