Martine Ceberio
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Research Interests
Martine Ceberio has a total of 77 co-authors. Here are the top co-authors by number of shared publications:
- Vladik Kreinovich
Department of Computer Science, University of Texas at El Paso, 500 W. University, El Paso, TX 79968, USA
41 shared publications - Olga Kosheleva
University of Texas at El Paso
22 shared publications - Marina Tuyako Mizukoshi
Universidade Federal de Goiás, IME
6 shared publications Olga Kosheleva
University of Texas at El Paso
5 shared publicationsVladik Kreinovich
University of Texas at El Paso
5 shared publications- Weldon Alexander Lodwick
University of Colorado Denver
5 shared publications
Latest Publications
- Open Access
- 17 November 2026
- Published by Springer Nature
- p. 197-214
A Causal Argumentation Method for Explainability of Machine Learning Models
- 17 November 2026
- Published by Springer Nature
- p. 197-214
Open AccessA Causal Argumentation Method for Explainability of Machine Learning Models
- 4 November 2026
- Published by Springer Nature
- p. 252-263
FedAWL: Client Selection via Age-Weighted Loss for Non-IID Federated Learning
- 4 November 2026
- Published by Springer Nature
- p. 252-263
FedAWL: Client Selection via Age-Weighted Loss for Non-IID Federated Learning
- 4 November 2026
- Published by Springer Nature
- p. 224-237
Adaptive Optimizer Sequencing for Kolmogorov-Arnold Networks
- 4 November 2026
- Published by Springer Nature
- p. 224-237
Adaptive Optimizer Sequencing for Kolmogorov-Arnold Networks
- 2 July 2026
- Published by Springer Nature
- p. 161-166
Data Fusion Is More Complex Than Data Processing: A Proof
- 2 July 2026
- Published by Springer Nature
- p. 161-166
Data Fusion Is More Complex Than Data Processing: A Proof
- 1 April 2026
- Published by Springer Nature
- p. 47-56
Does the Model Say What the Data Says? A Simple Heuristic for Model–Data Alignment
- 1 April 2026
- Published by Springer Nature
- p. 47-56
Does the Model Say What the Data Says? A Simple Heuristic for Model–Data Alignment
- Open Access
Contribution to Error Analysis of Deep Neural Networks: Case of the Activation Functions
Deep neural networks are widely used in high-stakes applications such as medical imaging and autonomous perception, where numerical reliability is as critical as predictive accuracy. While prior work has focused on gradient stability, the propagation of small floating-point errors through nonlinear network components remains underexplored. This paper provides a quantitative analysis of error amplification in common activation functions, including ReLU, sigmoid, tanh, and softmax. For each function, we evaluate the perturbation response using two complementary approaches: absolute-error analysis, which captures the exact or bounded output change for a given additive input perturbation, and relative-error analysis, which measures the multiplicative amplification of errors relative to the nominal function value. By systematically deriving these bounds and expressions, we compare the numerical behavior across activations. These results enable a detailed description of the error propagation characteristics of each function, offering practical guidance for designing numerically stable and robust neural architectures in contexts where even minor computational errors can significantly affect downstream predictions.
Open AccessContribution to Error Analysis of Deep Neural Networks: Case of the Activation Functions
Deep neural networks are widely used in high-stakes applications such as medical imaging and autonomous perception, where numerical reliability is as critical as predictive accuracy. While prior work has focused on gradient stability, the propagation of small floating-point errors through nonlinear network components remains underexplored. This paper provides a quantitative analysis of error amplification in common activation functions, including ReLU, sigmoid, tanh, and softmax. For each function, we evaluate the perturbation response using two complementary approaches: absolute-error analysis, which captures the exact or bounded output change for a given additive input perturbation, and relative-error analysis, which measures the multiplicative amplification of errors relative to the nominal function value. By systematically deriving these bounds and expressions, we compare the numerical behavior across activations. These results enable a detailed description of the error propagation characteristics of each function, offering practical guidance for designing numerically stable and robust neural architectures in contexts where even minor computational errors can significantly affect downstream predictions.
- 16 October 2025
- Published by Springer Nature
- p. 13-25
What Makes a 7 a 7? Exploration into a Neural Network’s Features of Interest
- 16 October 2025
- Published by Springer Nature
- p. 13-25
What Makes a 7 a 7? Exploration into a Neural Network’s Features of Interest
- 27 October 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1891-1894
Machine Learning-Based Screening for Pediatric Hip Dysplasia: Towards a Validated Approach
Developmental Dysplasia of the Hip (DDH) is a congenital disorder with preventable consequences upon early diagnosis. It is a problem of significance since the incidence is in 1-3% of the newborns. This paper explores Machine Learning (ML) to enhance diagnosing DDH in a easy-to-use, reliable and economic manner, in order to enhance screening. In terms of methodology, we observe that diagnosing DDH requires an ultrasound imaging produced by a trained operator, who needs to be specially trained at this time. Such operators are scarce in the field. However, classical business-type ML is not rigorous and hence not valid for matters of human life and health. We explore approaches to make ML rigorous for this type of applications and to solve the screening and detection challenge in a safe way.
- 27 October 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1891-1894
Machine Learning-Based Screening for Pediatric Hip Dysplasia: Towards a Validated Approach
Developmental Dysplasia of the Hip (DDH) is a congenital disorder with preventable consequences upon early diagnosis. It is a problem of significance since the incidence is in 1-3% of the newborns. This paper explores Machine Learning (ML) to enhance diagnosing DDH in a easy-to-use, reliable and economic manner, in order to enhance screening. In terms of methodology, we observe that diagnosing DDH requires an ultrasound imaging produced by a trained operator, who needs to be specially trained at this time. Such operators are scarce in the field. However, classical business-type ML is not rigorous and hence not valid for matters of human life and health. We explore approaches to make ML rigorous for this type of applications and to solve the screening and detection challenge in a safe way.
- Open Access
- 6 May 2024
- Published by River Publishers in Journal of Mobile Multimedia
- p. 665-678
Just-in-Accuracy: Mobile Approach to Uncertainty
To make a mobile device last longer, we need to limit computations to a bare minimum. One way to do that, in complex control and decision making problems, is to limit precision with which we do computations, i.e., limit the number of bits in the numbers’ representation. A problem is that often, we do not know with what precision should we do computations to get the desired accuracy of the result. What we propose is to first do computations with very low precision, then, based on these computations, estimate what precision is needed to achieve the given accuracy, and then perform computations with this precision.
- 6 May 2024
- Published by River Publishers in Journal of Mobile Multimedia
- p. 665-678
Open AccessJust-in-Accuracy: Mobile Approach to Uncertainty
To make a mobile device last longer, we need to limit computations to a bare minimum. One way to do that, in complex control and decision making problems, is to limit precision with which we do computations, i.e., limit the number of bits in the numbers’ representation. A problem is that often, we do not know with what precision should we do computations to get the desired accuracy of the result. What we propose is to first do computations with very low precision, then, based on these computations, estimate what precision is needed to achieve the given accuracy, and then perform computations with this precision.
- 5 December 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 496-500
Which Activation Function Works Best for Training Artificial Pancreas: Empirical Fact and Its Theoretical Explanation
One of the most effective ways to help patients at the dangerous levels of diabetes is an artificial pancreas, a device that constantly monitors the patient's blood sugar level and injects insulin based on this level. Patient's reaction to insulin is highly individualized, so the artificial pancreas needs to be trained on each patient. It turns out that the best training results are attained when instead of the usual ReLU neurons, we use their minor modification known as Exponential Linear Units (ELU). In this paper, we provide a theoretical explanation for the empirically observed effectiveness of ELUs.
- 5 December 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 496-500
Which Activation Function Works Best for Training Artificial Pancreas: Empirical Fact and Its Theoretical Explanation
One of the most effective ways to help patients at the dangerous levels of diabetes is an artificial pancreas, a device that constantly monitors the patient's blood sugar level and injects insulin based on this level. Patient's reaction to insulin is highly individualized, so the artificial pancreas needs to be trained on each patient. It turns out that the best training results are attained when instead of the usual ReLU neurons, we use their minor modification known as Exponential Linear Units (ELU). In this paper, we provide a theoretical explanation for the empirically observed effectiveness of ELUs.