Banca de DEFESA: ANTÔNIO OLIVEIRA FILHO

Uma banca de DEFESA de DOUTORADO foi cadastrada pelo programa.
STUDENT : ANTÔNIO OLIVEIRA FILHO
DATE: 03/09/2026
TIME: 10:30
LOCAL: Online
TITLE:

Phoeni6: a Systematic Approach for Evaluating the Energy Consumption of Neural Networks

KEY WORDS:

Deep Neural Networks; Energy Consumption; Fair Comparison; Result Reproducibility; Experiment Automation; Containerization; Orchestration; Phoeni6.

PAGES: 102
BIG AREA: Ciências Exatas e da Terra
AREA: Ciência da Computação
SUMMARY:

The growing computational demand of neural networks has intensified concerns regard-
ing their energy consumption and environmental impact. Despite significant advances
in accuracy and performance, systematic methods for assessing energy efficiency remain limited. This thesis presents Phoeni6, a framework designed to support the principles of fair comparison and result reproducibility in the energy evaluation of neural networks. Phoeni6 integrates a modular and containerized architecture, a data management system based on NoSQL databases, and a methodology that automates the configuration, execution, and analysis of experiments across different hardware environments. By structuring evaluations through standardized data models and orchestrating tasks via containerized applications, the framework ensures portability, transparency, and scalability. To validate the proposal, experiments were conducted with different neural networks across diverse hardware scenarios, ranging from classical architectures to modern high-performance models. The experiments encompassed multiple input data formats, model compression techniques, and execution strategies, allowing for a detailed measurement of energy consumption under various conditions. The results highlight Phoeni6’s ability to collect, organize, and analyze large volumes of energy-related data, supporting standardized comparisons between models and devices. Furthermore, they demonstrate that the framework enables the correlation of energy metrics with performance indicators such as execution time and accuracy, thereby enhancing the understanding of trade-offs involved in applying neural networks in different contexts. This research contributes to the advancement of sustainable artificial intelligence practices by offering a systematic and reproducible approach to energy-focused experimentation. Thus, Phoeni6 establishes itself as a powerful tool to support both academic investigations and industrial applications concerned with the energy efficiency and environmental impact of machine learning models.

COMMITTEE MEMBERS:
Presidente - 1673543 - SAMUEL XAVIER DE SOUZA
Externo à Instituição - CARLOS ALBERTO VALDERRAMA SAKUYAMA - UMONS
Externo à Instituição - JOÃO ALEXANDRE BAPTISTA VIEIRA SARAIVA - UMinho
Externo à Instituição - FRANCISCO CHAGAS DE LIMA JUNIOR - UERN
Externo à Instituição - LUCAS MELLO SCHNORR - UFRGS