Hyperparameter Tuning over an Attention Model for Image Captioning

Castro, Roberto; Pineda, Israel; Morocho-Cayamcela, Manuel Eugenio

doi:10.1007/978-3-030-89941-7_13

Cited by 3 publications

(3 citation statements)

References 12 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…These activation functions, each with distinct characteristics, play a crucial role in introducing non-linearity into the neural network, enabling it to learn complex patterns in the data [33,34]. Possible optimizers are Adam, SGD, RMSprop, and AdamW [35]. The batch size options are set at 64, 128, or 256.…”

Section: A Multi-objective Optimizationmentioning

confidence: 99%

A Data-Driven Model for Power Loss Estimation of Magnetic Materials Based on Multi-Objective Optimization and Transfer Learning

Li,

Wang,

Liu

et al. 2024

IEEE Open J. Power Electron.

View full text Add to dashboard Cite

Traditional methods such as Steinmetz's equation (SE) and its improved variant (iGSE) have demonstrated limited precision in estimating power loss for magnetic materials. The introduction of Neural Network technology for assessing magnetic component power loss has significantly enhanced accuracy. Yet, an efficient method to incorporate detailed flux density information-which critically impacts accuracy-remains elusive. Our study introduces an innovative approach that merges Fast Fourier Transform (FFT) with a Feedforward Neural Network (FNN), aiming to overcome this challenge. To optimize the model further and strike a refined balance between complexity and accuracy, Multi-Objective Optimization (MOO) is employed to identify the ideal combination of hyperparameters, such as layer count, neuron number, activation functions, optimizers, and batch size. This optimized Neural Network outperforms traditionally intuitive models in both accuracy and size. Leveraging the optimized base model for known materials, transfer learning is applied to new materials with limited data, effectively addressing data scarcity. The proposed approach substantially enhances model training efficiency, achieves remarkable accuracy, and sets an example for Artificial Intelligence applications in loss and electrical characteristic predictions with challenges of model size, accuracy goals, and limited data.

show abstract

Section: A Multi-objective Optimizationmentioning

confidence: 99%

A Data-Driven Model for Power Loss Estimation of Magnetic Materials Based on Multi-Objective Optimization and Transfer Learning

Li,

Wang,

Liu

et al. 2024

IEEE Open J. Power Electron.

View full text Add to dashboard Cite

show abstract

“…In response to the uncertainties raised by the previously described experimental scenarios, the combination of crossentropy loss and Adam optimizer was highlighted as the best hyperparameter configuration according to the Top-5 Accuracy, BLEU-4, and loss value metrics. By reusing this configuration for the following experiment, different decisions can be made depending on the final purpose of the researcher [10]. If the architecture with the best metrics concerning response quality is required, the convolutional models ResNet-152 and ResNeXt-101 provided the best results in the metrics used in the previous experimentation.…”

Section: Introductionmentioning

confidence: 99%

Deep Learning Approaches Based on Transformer Architectures for Image Captioning Tasks

et al. 2022

Self Cite

View full text Add to dashboard Cite

This paper focuses on visual attention, a state-of-the-art approach for image captioning tasks within the computer vision research area. We study the impact that different hyperparemeter configurations on an encoder-decoder visual attention architecture in terms of efficiency. Results show that the correct selection of both the cost function and the gradient-based optimizer can significantly impact the captioning results. Our system considers the cross-entropy, Kullback-Leibler divergence, mean squared error, and negative log-likelihood loss functions; the adaptive momentum (Adam), AdamW, RMSprop, stochastic gradient descent, and Adadelta optimizers. Experimentation shows that a combination of cross-entropy with Adam is the best alternative returning a Top-5 accuracy value of 73.092 and a BLEU-4 value of 20.10. Furthermore, a comparative analysis of alternative convolutional architectures demonstrated their performance as an encoder. Our results show that ResNext-101 stands out with a Top-5 accuracy of 73.128 and a BLEU-4 of 19.80; positioning itself as the best option when looking for the optimum captioning quality. However, MobileNetV3 proved to be a much more compact alternative with 2,971,952 parameters and 0.23 Giga fixed-point Multiply-Accumulate operations per Second (GMACS). Consequently, MobileNetV3 offers a competitive output quality at the cost of lower computational performance, supported by values of 19.50 and 72.928 for the BLEU-4 and Top-5 accuracy, respectively. Finally, when testing vision transformer (ViT), and data-efficient image transformer (DeiT) models to replace the convolutional component of the architecture, DeiT achieved an improvement over ViT, obtaining a value of 34.44 in the BLEU-4 metric.

show abstract

“…Currently, computer vision (CV) tasks are useful for solving problems related to object detection, classification, object counting, visual surveillance, etc., taking advantage of video resources from public surveillance cameras located in many public areas (i.e., shopping malls, supermarkets, airports, train stations, stadiums, etc.) [9][10][11][12]. The problem of the correct/incorrect wearing of face masks implies two CV tasks: (1) object detection and (2) object classification.…”

Section: Introductionmentioning

confidence: 99%

A Computer Vision Model to Identify the Incorrect Use of Face Masks for COVID-19 Awareness

Crespo

Martínez

et al. 2022

Applied Sciences

Self Cite

View full text Add to dashboard Cite

Face mask detection has become a great challenge in computer vision, demanding the coalition of technology with COVID-19 awareness. Researchers have proposed deep learning models to detect the use of face masks. However, the incorrect use of a face mask can be as harmful as not wearing any protection at all. In this paper, we propose a compound convolutional neural network (CNN) architecture based on two computer vision tasks: object localization to discover faces in images/videos, followed by an image classification CNN to categorize the faces and show if someone is using a face mask correctly, incorrectly, or not at all. The first CNN is built upon RetinaFace, a model to detect faces in images, whereas the second CNN uses a ResNet-18 architecture as a classification backbone. Our model enables an accurate identification of people who are not correctly following the COVID-19 healthcare recommendations on face mask use. To enable further global use of our technology, we have released both the dataset used to train the classification model and our proposed computer vision pipeline to the public, and optimized it for embedded systems deployment.

show abstract

Hyperparameter Tuning over an Attention Model for Image Captioning

Cited by 3 publications

References 12 publications

A Data-Driven Model for Power Loss Estimation of Magnetic Materials Based on Multi-Objective Optimization and Transfer Learning

A Data-Driven Model for Power Loss Estimation of Magnetic Materials Based on Multi-Objective Optimization and Transfer Learning

Deep Learning Approaches Based on Transformer Architectures for Image Captioning Tasks

A Computer Vision Model to Identify the Incorrect Use of Face Masks for COVID-19 Awareness

Contact Info

Product

Resources

About