Neuroevolution of Self-Attention Over Proto-Objects
Evolves self-attention mechanisms over proto-objects, enabling agents to learn compact, interpretable visual representations for control tasks.
Computer Science · Artificial Intelligence
Rafael Coimbra Pinto
Professor and researcher
I investigate systems that learn continuously: machine learning, neural networks, reinforcement learning, evolutionary computation, and AI for games.
Profile
Applied research in continual learning
PhD in Computer Science from the Federal University of Rio Grande do Sul (UFRGS) and professor at the Federal Institute of Education, Science and Technology of Rio Grande do Sul (IFRS), Canoas Campus.
My career connects artificial intelligence research, student education, and software development experience. My current work includes the neuroevolution of attention mechanisms, incremental models, and agents capable of learning from streams of experience.
Research output
Publications
Complete list, from newest to oldest. Use the search field to find a topic, author, venue, or year.
20 publications
PReLU: Yet Another Single-Layer Solution to the XOR Problem
Investigates a single-layer solution to the XOR problem based on parametric rectified linear units, revisiting a classic limitation of neural networks.
Fast Reinforcement Learning with Incremental Gaussian Mixture Models
Combines reinforcement learning with incremental Gaussian mixture models for fast, data-efficient function approximation.
Model-Free Episodic Control with Online State Aggregation
Combines model-free episodic control with online state aggregation, reducing memory requirements and generalizing across similar experiences.
Beating Bomberman with Artificial Intelligence
Explores artificial intelligence techniques to build competitive Bomberman agents and evaluate strategies in a dynamic environment.
Continuous Reinforcement Learning with Incremental Gaussian Mixture Models
Doctoral thesis on continuous reinforcement learning with incremental Gaussian mixture models for online learning and control.
Scalable and Incremental Learning of Gaussian Mixture Models
Presents a scalable, incremental approach for learning Gaussian mixtures directly from data streams without storing the entire history.
A Fast Incremental Gaussian Mixture Model
Introduces a fast, incremental, single-pass Gaussian mixture model that adapts its structure as new data arrives.
Autocorrelation and Partial Autocorrelation Functions to Improve Neural Networks Models on Univariate Time Series Forecasting
Uses autocorrelation and partial autocorrelation functions to select relevant inputs and improve neural networks for univariate time-series forecasting.
Using a Gaussian Mixture Neural Network for Incremental Learning and Robotics
Applies a Gaussian mixture neural network to incremental learning and demonstrates its use in robotics problems.
One-Shot Learning in the Road Sign Problem
Evaluates one-shot learning for road-sign recognition with an incremental neural model.
Learning Abstract Behaviors with the Hierarchical Incremental Gaussian Mixture Network
Investigates a hierarchical incremental Gaussian mixture network for acquiring abstract representations and behaviors.
IGMN: An Incremental Gaussian Mixture Network that Learns Instantaneously from Data Flows
Describes an incremental Gaussian mixture network that learns from data streams in real time and adjusts its complexity automatically.
Echo State Incremental Gaussian Mixture Network for Spatio-Temporal Pattern Processing
Combines echo state networks and incremental Gaussian mixtures to process spatiotemporal patterns in sequences.
Recursive Incremental Gaussian Mixture Network for Spatio-Temporal Pattern Processing
Proposes a recursive incremental architecture based on Gaussian mixtures to capture temporal dependencies in sequential patterns.
Applying the Incremental Gaussian Neural Network to Concept Formation and Robotic Tasks
Studies incremental concept formation and its application to robotic tasks using a Gaussian neural network.
Online Incremental One-Shot Learning of Temporal Sequences
Master’s thesis on online, incremental, one-shot learning for modeling and predicting temporal sequences.
Um Estudo de Redes Neurais Não-Supervisionadas Temporais
A study of unsupervised neural models for representing and learning temporal relationships.
LoopSOM: A Robust SOM Variant Using Self-Organizing Temporal Feedback Connections
Presents a robust self-organizing map variant that incorporates self-organizing temporal feedback connections.
A Neocortex Inspired Hierarchical Spatio-Temporal Pattern Recognition System
Proposes a neocortex-inspired hierarchical system for recognizing spatiotemporal patterns by combining layered organization and temporal memory.
No publications match your search. Try another term.
Contact
Research, supervision, and collaboration
Federal Institute of Education, Science and Technology of Rio Grande do Sul — Canoas Campus.
rafael.pinto@canoas.ifrs.edu.br