Browsing by Author "Yigit,G."

Now showing 1 - 4 of 4

Citation Count: 0
Data Augmentation With In-Context Learning and Comparative Evaluation in Math Word Problem Solving
(Springer, 2024) Yigit,G.; Amasyali,M.F.
Math Word Problem (MWP) solving presents a challenging task in Natural Language Processing (NLP). This study aims to provide MWP solvers with a more diverse training set, ultimately improving their ability to solve various math problems. We propose several methods for data augmentation by modifying the problem texts and equations, such as synonym replacement, rule-based: question replacement, and rule based: reversing question methodologies over two English MWP datasets. This study extends by introducing a new in-context learning augmentation method, employing the Llama-7b language model. This approach involves instruction-based prompting for rephrasing the math problem texts. Performance evaluations are conducted on 9 baseline models, revealing that augmentation methods outperform baseline models. Moreover, concatenating examples generated by various augmentation methods further improves performance. © The Author(s), under exclusive licence to Springer Nature Singapore Pte Ltd. 2024.
Citation Count: 1
Exploring the Benefits of Data Augmentation in Math Word Problem Solving
(Institute of Electrical and Electronics Engineers Inc., 2023) Yigit,G.; Amasyali,M.F.
Math Word Problem (MWP) is a challenging Natural Language Processing (NLP) task. Existing MWP solvers have shown that current models need to generalize better and obtain higher performances. In this study, we aim to enrich existing MWP datasets with high-quality data, which may improve MWP solvers' performances. We propose several data augmentation methods by applying minor modifications to the problem texts and equations of English MWPs datasets which contain equations with one unknown. Extensive experiments on two MWPs datasets have shown that data created by augmented methods have considerably improved performance. Moreover, further increasing the training samples by combining the samples generated by the proposed augmentation methods provides further performance improvements. © 2023 IEEE.
Citation Count: 1
Improving Diabetic Retinopathy Detection Using Patchwise Cnn With Bigru Model
(Institute of Electrical and Electronics Engineers Inc., 2023) Darici,M.B.; Yigit,G.
This study addresses Diabetic Retinopathy (DR), a diabetes complication that can lead to vision loss if not promptly diagnosed and treated. Recent advances in deep learning have shown promising results in detecting DR from retinal images. The study introduces a novel patch-based CNN-biGRU model for DR detection. The proposed model extracts patches from retinal images employing a sliding window strategy and then uses a Convolutional Neural Network (CNN) architecture to extract features from each patch. The features extracted from each patch are then concatenated, and a 4-layer bidirectional Gated Recurrent Unit (biGRU) is applied to predict the whole image. We assessed the proposed model on a publicly available dataset named APTOS 2019 Blindness Detection and achieved an accuracy of 73.5%, outperforming existing state-of-the-art approaches. The given patch-based CNN model can improve the accuracy of DR detection and aims to assist ophthalmologists in making more accurate diagnoses. © 2023 IEEE.
Citation Count: 0
A Siamese Network-Based Approach for Autism Spectrum Disorder Detection With Dual Architecture
(Institute of Electrical and Electronics Engineers Inc., 2023) Yigit,G.; Darici,M.B.
Autism Spectrum Disorder (ASD) is a sophisticated neuro-developmental condition impacting numerous children. Early detection of ASD is crucial to implement suitable treatments to improve the daily activities of people with ASD. This paper introduces a system for ASD detection using facial images. The proposed model presents a unique system inspired by Siamese networks. Unlike traditional Siamese networks focusing on input pairs, our model leverages architectural pairs for feature combinations. During training, we combine features learned from different or the same architectures. This enables information transfer and improves the model's capture of comprehensive patterns. Experimental results on the 2940 facial images dataset demonstrate the effectiveness of our system, which exhibits improved accuracy compared to using individual architectures. When (ResNet50, VGG16) architecture pairs are employed in the proposed approach, the highest performance is obtained with an accuracy of 78.57%. Leveraging the strengths of multiple architectures, our model provides a comprehensive and robust representation of input data, leading to improved performance. © 2023 IEEE.