AI FRAMEWORK TO SELECT ROBUST CONVOLUTED NEURAL NETWORKS FOR SENSOR DATA

Louis Adrian Böhm; Odemir Bruno

Todos os Trabalhos

Trabalho

Título do Trabalho

AI FRAMEWORK TO SELECT ROBUST CONVOLUTED NEURAL NETWORKS FOR SENSOR DATA

Autores

Louis Adrian Böhm
Odemir Bruno

Modalidade

Pôster - resumo

Área temática

Sensores e biossensores

Data de Publicação

27/05/2026

País da Publicação

Brasil

Idioma da Publicação

pt-BR

Página do Trabalho

https://www.even3.com.br/anais/workshop-do-ineo-2026/1462694-ai-framework-to-select-robust-convoluted-neural-networks-for-sensor-data

ISBN

978-65-272-2467-9

Palavras-Chave

Convolutional Neural Networks, Machine Learning, Sensor Data, Image Recognition, Reinforcement Learning

Resumo

Selecting the optimal convolutional neural network (CNN) backbone is critical for task-specific imaging techniques such as sensor data processing, medical imaging, remote sensing, and others. CNNs are specialized deep learning algorithms that are inspired by the visual cortex found in animal brains. They use automated filter optimization to process grid-like visual data and extract spatial features like textures and shapes. When deploying pre-trained CNNs, they efficiently extract features from sensor images, which can then be used to train classifiers specific to a given task. However, finding the best CNN backbone relies on exhaustive empirical testing and is often labor-intensive [1]. We propose a novel backbone selection framework that repurposes reinforcement learning (RL) agents as diagnostic probes of CNN feature quality. Our method deploys RL agents that learn to manipulate images of one class such that their CNN features become less distinguishable from those of another class. The central idea is that the learning performance of the RL agent serves as a metric for the representation quality of the backbone: A backbone whose features are more easily manipulated by the RL agent fails to encode the features robustly. To validate our approach, we benchmark it against an existing study that manually evaluates different CNN backbones for the classification of lung diseases from X-ray images [2]. Our method produces a comparable ranking of backbone quality while demonstrating superiority by fully automating and generalizing the selection process, effectively eliminating the need for labor-intensive manual evaluation. Furthermore, we apply our framework to a dataset of blood screening images to propose a suitable and robust CNN backbone for detecting PCA proteins, a biomarker used in prostate cancer screening [3]. [1] D. R. Sarvamangala and Raghavendra V. Kulkarni. “Convolutional neural networks in medical image understanding: a survey”. en. In: Evolutionary Intelligence 15 (1 Mar. 2022), pp. 1–22. doi: 10.1007/s12065-020-00540-3. [2] Md Mamunur Rahaman et al. “Identification of COVID-19 samples from chest X-Ray images using deep learning: A comparison of transfer learning approaches”. en. In: Journal of X-Ray Science and Technology: Clinical Applications of Diagnosis and Therapeutics 28 (5 Sept. 2020), pp. 821–839. doi: 10.3233/xst-200715. [3] Valquiria C. Rodrigues et al. “Electrochemical and optical detection and machine learning applied to images of genosensors for diagnosis of prostate cancer with the biomarker PCA3”. en. In: Talanta 222 (Jan. 2021), p. 121444. doi: 10.1016/j.talanta.2020.121444.

Título do Evento: Workshop do INEO 2026
Cidade do Evento: Nazaré Paulista
Título dos Anais do Evento: Anais do Workshop do INEO 2026
Nome da Editora: Even3
Meio de Divulgação: Meio Digital

Como citar

BÖHM, Louis Adrian; BRUNO, Odemir. AI FRAMEWORK TO SELECT ROBUST CONVOLUTED NEURAL NETWORKS FOR SENSOR DATA.. In: Anais do Workshop do INEO 2026. Anais...Nazaré Paulista(SP) Hotel Estância Atibainha, 2026. Disponível em: https//www.even3.com.br/anais/workshop-do-ineo-2026/1462694-AI-FRAMEWORK-TO-SELECT-ROBUST-CONVOLUTED-NEURAL-NETWORKS-FOR-SENSOR-DATA. Acesso em: 10/08/2026