Limited Data Set Research Articles

Cross-Field Road Markings Detection Based on Inverse Perspective Mapping

With the rapid development of the autonomous vehicles industry, there has been a dramatic proliferation of research concerned with related works, where road markings detection is an important issue. When there is no public open data in a field, we must collect road markings data and label them by ourselves manually, which is huge labor work and takes lots of time. Moreover, object detection often encounters the problem of small object detection. The detection accuracy often decreases when the detection distance increases. This is primarily because distant objects on the road take up few pixels in the image and object scales vary depending on different distances and perspectives. For the sake of solving the issues mentioned above, this paper utilizes a virtual dataset and open dataset to train the object detection model and cross-field testing in the field of Taiwan roads. In order to make the model more robust and stable, the data augmentation method is employed to generate more data. Therefore, the data are increased through the data augmentation method and homography transformation of images in the limited dataset. Additionally, Inverse Perspective Mapping is performed on the input images to transform them into the bird’s eye view, which solves the “small objects at far distance” problem and the “perspective distortion of objects” problem so that the model can clearly recognize the objects on the road. The model testing on the front-view images and bird’s eye view images also shows a remarkable improvement of accuracy by 18.62%.

Data Augmentation for Voiceprint Recognition Using Generative Adversarial Networks

Voiceprint recognition systems often face challenges related to limited and diverse datasets, which hinder their performance and generalization capabilities. This study proposes a novel approach that integrates generative adversarial networks (GANs) for data augmentation and convolutional neural networks (CNNs) with mel-frequency cepstral coefficients (MFCCs) for voiceprint classification. Experimental results demonstrate that the proposed methodology improves recognition accuracy by up to 15% in low-resource scenarios. The optimal ratio of real-to-GAN-generated samples was determined to be 3:2, which balanced dataset diversity and model performance. In specific cases, the model achieved an accuracy of 96.6%, showcasing its effectiveness in capturing unique voice characteristics while mitigating overfitting. These results highlight the potential of combining GAN-augmented data and CNN-based classification to enhance voiceprint recognition in diverse and resource-constrained environments.

An Enhanced Cycle Generative Adversarial Network Approach for Nighttime Pineapple Detection of Automated Harvesting Robots

Nighttime pineapple detection for automated harvesting robots is a significant challenge in intelligent agriculture. As a crucial component of robotic vision systems, accurate fruit detection is essential for round-the-clock operations. The study compared advanced end-to-end style transfer models, including U-GAT-IT, SCTNet, and CycleGAN, finding that CycleGAN produced relatively good-quality images but had issues such as the inadequate restoration of nighttime details, color distortion, and artifacts. Therefore, this study further proposed an enhanced CycleGAN approach to address limited nighttime datasets and poor visibility, combining style transfer with small-sample object detection. The improved model features a novel generator structure with ResNeXtBlocks, an optimized upsampling module, and a hyperparameter optimization strategy. This approach achieves a 29.7% reduction in FID score compared to the original CycleGAN. When applied to YOLOv7-based detection, this method significantly outperforms existing approaches, improving precision, recall, average precision, and F1 score by 13.34%, 45.11%, 56.52%, and 30.52%, respectively. These results demonstrate the effectiveness of our enhanced CycleGAN in expanding limited nighttime datasets and supporting efficient automated harvesting in low-light conditions, contributing to the development of more versatile agricultural robots capable of continuous operation.

Enhancing Software Effort Estimation with Pre-Trained Word Embeddings: A Small-Dataset Solution for Accurate Story Point Prediction

Traditional software effort estimation methods, such as term frequency–inverse document frequency (TF-IDF), are widely used due to their simplicity and interpretability. However, they struggle with limited datasets, fail to capture intricate semantics, and suffer from dimensionality, sparsity, and computational inefficiency. This study used pre-trained word embeddings, including FastText and GPT-2, to improve estimation accuracy in such cases. Seven pre-trained models were evaluated for their ability to effectively represent textual data, addressing the fundamental limitations of TF-IDF through contextualized embeddings. The results show that combining FastText embeddings with support vector machines (SVMs) consistently outperforms traditional approaches, reducing the mean absolute error (MAE) by 5–18% while achieving accuracy comparable to deep learning models like GPT-2. This approach demonstrated the adaptability of pre-trained embeddings for small datasets, balancing semantic richness with computational efficiency. The proposed method optimized project planning and resource allocation while enhancing software development through accurate story point prediction while safeguarding privacy and security through data anonymization. Future research will explore task-specific embeddings tailored to software engineering domains and investigate how dataset characteristics, such as cultural variations, influence model performance, ensuring the development of adaptable, robust, and secure machine learning models for diverse contexts.

Re-Estimating GEDI Ground Elevation Using Deep Learning: Impacts on Canopy Height and Aboveground Biomass

This paper presents a method to improve ground elevation estimates through waveform analysis from the Global Ecosystem Dynamics Investigation (GEDI) and examines its impact on canopy height and aboveground biomass (AGB) estimation. The method uses a deep learning model to estimate ground elevation from the GEDI waveform. Geographic transferability was demonstrated by recalculating canopy height and AGB estimation accuracy using the improved ground elevation without changing established GEDI formulas for relative height (RH) and AGB. The study covers four regions in Japan and South America, from subarctic to tropical zones, integrating GEDI waveform data with airborne laser scan (ALS) data. Transfer learning was explored to enhance accuracy in regions not used for training. Ground elevation estimates using deep learning showed an RMSE improvement of over 3 m compared to the conventional GEDI L2A product, with generalization performance. Applying transfer learning and retraining with additional data further improved the estimation accuracy, even with limited datasets. The findings suggest that improving ground elevation estimates enhances canopy height and AGB accuracy, maximizing GEDI’s global AGB estimation algorithms. Optimizing models for each region could further enhance accuracy. The broader application of this method may improve global carbon cycle understanding and climate models.

Adapting SAM2 Model from Natural Images for Tooth Segmentation in Dental Panoramic X-Ray Images

Dental panoramic X-ray imaging, due to its high cost-effectiveness and low radiation dose, has become a widely used diagnostic tool in dentistry. Accurate tooth segmentation is crucial for lesion analysis and treatment planning, helping dentists to quickly and precisely assess the condition of teeth. However, dental X-ray images often suffer from noise, low contrast, and overlapping anatomical structures, coupled with limited available datasets, leading traditional deep learning models to experience overfitting, which affects generalization ability. In addition, high-precision deep models typically require significant computational resources for inference, making deployment in real-world applications challenging. To address these challenges, this paper proposes a tooth segmentation method based on the pre-trained SAM2 model. We employ adapter modules to fine-tune the SAM2 model and introduce ScConv modules and gated attention mechanisms to enhance the model’s semantic understanding and multi-scale feature extraction capabilities for medical images. In terms of efficiency, we utilize knowledge distillation, using the fine-tuned SAM2 model as the teacher model for distilling knowledge to a smaller model named LightUNet. Experimental results on the UFBA-UESC dataset show that, in terms of performance, our model significantly outperforms the traditional UNet model in multiple metrics such as IoU, effectively improving segmentation accuracy and model robustness, particularly with limited sample datasets. In terms of efficiency, LightUNet achieves comparable performance to UNet, but with only 1.6% of its parameters and 24.0% of the inference time, demonstrating its feasibility for deployment on edge devices.

Magnetic resonance image denoising for Rician noise using a novel hybrid transformer-CNN network (HTC-net) and self-supervised pretraining.

Magnetic resonance imaging (MRI) is a crucial technique for both scientific research and clinical diagnosis. However, noise generated during MR data acquisition degrades image quality, particularly in hyperpolarized (HP) gas MRI. While deep learning (DL) methods have shown promise for MR image denoising, most of them fail to adequately utilize the long-range information which is important to improve denoising performance. Furthermore, the sample size of paired noisy and noise-free MR images also limits denoising performance. To develop an effective DL method that enhances denoising performance and reduces the requirement of paired MR images by utilizing the long-range information and pretraining. In this work, a hybrid Transformer-convolutional neural network (CNN) network (HTC-net) and a self-supervised pretraining strategy are proposed, which effectively enhance the denoising performance. In HTC-net, a CNN branch is exploited to extract the local features. Then a Transformer-CNN branch with two parallel encoders is designed to capture the long-range information. Within this branch, a residual fusion block (RFB) with a residual feature processing module and a feature fusion module is proposed to aggregate features at different resolutions extracted by two parallel encoders. After that, HTC-net exploits the comprehensive features from the CNN branch and the Transformer-CNN branch to accurately predict noise-free MR images through a reconstruction module. To further enhance the performance on limited MRI datasets, a self-supervised pretraining strategy is proposed. This strategy employs self-supervised denoising to equip the HTC-net with denoising capabilities during pretraining, and then the pre-trained parameters are transferred to facilitate subsequent supervised training. Experimental results on the pulmonary HP 129Xe MRI dataset (1059 images) and IXI dataset (5000 images) all demonstrate the proposed method outperforms the state-of-the-art methods, exhibiting superior preservation of edges and structures. Quantitatively, on the pulmonary HP 129Xe MRI dataset, the proposed method outperforms the state-of-the-art methods by 0.254-0.597dB in PSNR and 0.007-0.013 in SSIM. On the IXI dataset, the proposed method outperforms the state-of-the-art methods by 0.3-0.927dB in PSNR and 0.003-0.016 in SSIM. The proposed method can effectively enhance the quality of MR images, which helps improve the diagnosis accuracy in clinical.

Survey of Quantum Generative Adversarial Networks (QGAN) to Generate Images

Quantum Generative Adversarial Networks (QGANs) represent a useful development in quantum machine learning, using the particular properties of quantum mechanics to solve the challenges of data analysis and modeling. This paper brings up a general analysis of five QGAN architectures, focusing on their evolution, strengths, weaknesses, and limitations in noisy intermediate-scale quantum (NISQ) devices. Primary methods like Entangling Quantum GAN (EQ-GAN) and Quantum state fidelity (QuGAN) concentrate on stability, convergence, and robust performance on small-scale datasets such as 2 × 2 grayscale images. Intermediate models such as Image Quantum GAN (IQGAN) and Experimental Quantum GAN (EXQGAN) provide new ideas like trainable encoders and patch-based sub-generators that are scalable to 8 × 8 datasets with increasing noise resilience. The most advanced method is Parameterized Quantum Wasserstein GAN (PQWGAN), which uses a hybrid quantum-classical structure to obtain high-resolution image processing for 28 × 28 grayscale datasets while trying to maintain parameter efficiency. This study explores, analyzes, and summarizes critical problems of QGANs, including accuracy, convergence, parameter efficiency, image quality, performance metrics, and training stability under noisy conditions. In addition, developing QGANs can generate and train parameters in quantum approximation optimization algorithms. One of the useful applications of QGAN is generating medical datasets that can generate medical images from limited datasets to train specific medical models for the recognition of diseases.

Empathetic Deep Learning: Transferring Adult Speech Emotion Models to Children With Gender-Specific Adaptations Using Neural Embeddings

AbstractUnderstanding and recognizing emotional states through speech has vast implications in areas ranging from customer service to mental health. In this paper, we investigate the relationship between adults and children for the task of automatic speech emotion recognition, focusing on the critical issue of limited datasets for children’s emotions. We use two databases: IEMOCAP, which contains emotional speech recordings from adults, and AIBO, which includes recordings from children. To address the dataset limitations, we employ transfer learning by training a neural network to classify adult emotional speech using a Wav2Vec model for feature extraction, followed by a classification head for the downstream task. However, the labels between IEMOCAP and AIBO do not align perfectly, presenting a challenge in emotional mapping. To tackle this, we perform inference on children’s data to examine how emotional labels in IEMOCAP correspond to those in AIBO, highlighting the complexities of cross-age emotional transfer. This approach achieved F-scores of up to 0.47. In addition, we trained male and female IEMOCAP models to determine how variations in gender within adult speech affect emotional mapping in children data. Some of our findings indicate that female samples align more with high arousal emotions, while male samples align more with low arousal emotion, underscoring the importance of gender in emotion recognition. To the best of our knowledge, this is the first study in the field of deep learning applications on emotion recognition that analyses the effects of genders and age groups on emotional mapping.

Prediction of Higher Education Student Dropout based on Regularized Regression Models

This study explores the critical topic of student dropout in higher education institutions. To allow early and precise interventions and to provide a multifaceted view of student performance, this study combined two predictive models for dropout classification and score prediction. At first, a logistic regression model was developed to predict student dropout at an early stage. Then, to enhance dropout prediction, a second-degree polynomial regression model was used to predict student results based on available academic variables (access, tests, exams, projects, and assignments) from a Moodle course. Dealing with a limited dataset is a key challenge due to the high risk of overfitting. To address this issue and achieve a balance between overfitting, data size, and model complexity, the predictive models were evaluated with L1 (Lasso) and L2 (Ridge) regularization terms. The regularization techniques of the predictive models led to an accuracy of up to 89% and an R2 score of up to 86%.

Advancements and Challenges: A Comprehensive Review of GAN-based Models for the Mitigation of Small Dataset and Texture Sticking Issues in Fake License Plate Recognition

This review paper critically examines the recent advancements in refining Generative Adversarial Networks (GANs) to address the challenges posed by small datasets and the persisting issue of texture sticking in the domain of fake license plate recognition. Recognizing the limitations posed by insufficient data, the survey begins with an exploration of various GAN architectures, including pix2pix_GAN, CycleGAN, and SRGAN, that have been employed to synthesize diverse and realistic license plate images. Notable achievements include high accuracy in License Plate Character Recognition (LPCR), advancements in generating new format license plates, and improvements in license plate detection using YOLO. The second focal point of this review centers on mitigating the texture sticking problem, a crucial concern in GAN-generated content. Recent enhancements, such as the integration of StyleGAN2-ADA and StyleGAN3, aim to address challenges related to texture dynamics during video generation. Additionally, adaptive data augmentation mechanisms have been introduced to stabilize GAN training, particularly when confronted with limited datasets. The synthesis of these findings provides a comprehensive overview of the evolving landscape in mitigating challenges associated with small datasets and texture sticking in fake license plate recognition. The review not only underscores the progress made but also identifies emerging trends and areas for future exploration. These insights are vital for researchers, practitioners, and policymakers aiming to bolster the effectiveness and reliability of GAN-based models in the critical domain of license plate recognition.

Automated surgical skill assessment in endoscopic pituitary surgery using real‐time instrument tracking on a high‐fidelity bench‐top phantom

AbstractImproved surgical skill is generally associated with improved patient outcomes, although assessment is subjective, labour intensive, and requires domain‐specific expertise. Automated data‐driven metrics can alleviate these difficulties, as demonstrated by existing machine learning instrument tracking models. However, these models are tested on limited datasets of laparoscopic surgery, with a focus on isolated tasks and robotic surgery. Here, a new public dataset is introduced: the nasal phase of simulated endoscopic pituitary surgery. Simulated surgery allows for a realistic yet repeatable environment, meaning the insights gained from automated assessment can be used by novice surgeons to hone their skills on the simulator before moving to real surgery. Pituitary Real‐time INstrument Tracking Network (PRINTNet) has been created as a baseline model for this automated assessment. Consisting of DeepLabV3 for classification and segmentation, StrongSORT for tracking, and the NVIDIA Holoscan for real‐time performance, PRINTNet achieved 71.9% multiple object tracking precision running at 22 frames per second. Using this tracking output, a multilayer perceptron achieved 87% accuracy in predicting surgical skill level (novice or expert), with the ‘ratio of total procedure time to instrument visible time’ correlated with higher surgical skill. The new publicly available dataset can be found at https://doi.org/10.5522/04/26511049.

Safety, efficacy, and tolerability of alemtuzumab in pediatric patients with active relapsing-remitting multiple sclerosis: The LemKids study.

Limited licensed medications are available for multiple sclerosis (MS) in pediatric patients. To evaluate the efficacy, safety, and tolerability of alemtuzumab in pediatric patients with relapsing-remitting multiple sclerosis (RRMS) and disease activity on prior disease-modifying therapies (DMTs). LemKids was a multicenter, multinational, single-arm, open-label, switch (from ongoing DMT to alemtuzumab treatment) study in pediatric RRMS patients (aged 10-<18 years), with disease activity on DMT. The primary endpoint was a comparison of the number of new/enlarging T2 lesions on the magnetic resonance imaging of the brain between the prior-DMT period and alemtuzumab treatment. This study was prematurely terminated due to low enrollment and an European Medicines Agency Article-20 pharmacovigilance review of alemtuzumab in adult RRMS. Of 46 screened patients, 16 were enrolled; 12 completed prior-DMT treatment period; 11 received alemtuzumab of whom 7 completed treatment. Patients on alemtuzumab developed fewer new/enlarging T2 lesions compared with prior-DMT (7 vs 178, relative risk (95% confidence interval): 0.04 (0.01-0.14)). No significant pharmacodynamic changes or safety concerns were noted in this limited dataset. Alemtuzumab treatment was associated with a low number of new/enlarging T2 lesions in pediatric patients with RRMS and was safe and well tolerated in seven patients during infusion and the initial 4 months.

Effects of viscosity and induced magnetic fields on weakly nonlinear wave transmission in a viscoelastic tube using physics-informed neural networks

This study presents an advanced deep learning methodology that utilizes physics-informed neural networks (PINNs), to analyze the transmission of weakly nonlinear waves in a prestressed viscoelastic arterial tube. Using the long wave approximation, a mathematical model is constructed to replicate the propagation of weakly nonlinear waves in a viscoelastic arterial tube filled with viscous nanofluid, taking into account the influence of an induced magnetic field. The perturbed Burger, perturbed Korteweg–de Vries, and perturbed Korteweg–de Vries-Burgers equations are formulated based on the combined effects of nanofluid viscosity and the applied magnetic field using the reductive perturbation technique. Semi-supervised physics-informed neural network models are utilized to solve these perturbed evolutionary equations, trained on a limited dataset within their rectangular domain of definition. Gaussian process-based Bayesian optimization is used to determine the hyperparameters of the neural network, ensuring optimal model is performance. The effectiveness of the optimal models is evaluated by calculating the residual losses associated with the perturbed partial differential equations (PDEs). Visual representations of weakly non-linear wave propagation, considering nanofluid viscosity and induced magnetic fields, enhance the comprehension of dissipative effects in the cardiovascular system. These insights aid in obtaining precise measurements of pulse wave velocity for cardiovascular health monitoring. Consequently, the application of PINN proves to be a valuable tool for solving real-world PDEs and highlights its importance in advancing medical machine learning fields.

Machine Learning for Automated Oil Palm Fruit Grading: The Role of Fuzzy C-Means Segmentation and Textural Features

Oil palm fruit, a high-demand and economically valuable crop, faces grading challenges in Malaysia due to labour-intensive methods, resulting in unharvested ripe fruits and yield losses. Extensive research has been conducted on outdoor palm fruit classifiers, but most high-accuracy research in this field relies heavily on large image datasets. This creates a trade-off between employing more sophisticated deep-learning approaches that require extensive data and utilising traditional techniques that have lower data requirements. Therefore, the aim is to propose an outdoor image-based fresh fruit bunch (FFB) classification system, emphasising pre-processing technique and feature extraction techniques suitable for limited datasets. To achieve this, the FFB was localised within the original images using the YOLOv4-Tiny algorithm. Subsequently, a salient segmentation method utilising superpixel-based Fast Fuzzy C-means (FFCM) clustering is employed to remove the background from the images. Next, colour moment and opposite colour local binary pattern (OCLBP) features are extracted from the segmented images to capture important information for classification. Finally, the extracted features are fed into a Multilayer Perceptron (MLP) classifier, which enables the system to predict five classes: damaged, empty, unripe, ripe, and overripe. The developed system demonstrated a commendable performance in accurately classifying fruit bunches, achieving an accuracy of 93.68%. In conclusion, the proposed system effectively addresses the issue of unripe fruit harvesting and contributes to the advancement of state-of-the-art methods in classifying outdoor FFB images.

Noise robust classification of carbide tool wear in machining mild steel using texture extraction based transfer learning approach for predictive maintenance

Acoustics based smart condition monitoring is a viable alternative to mechanical vibrations or image-capture based predictive maintenance methods. In this study, a texture analysis based transfer learning methodology was applied to classify tool wear based on the noise generated during mild steel machining. The machining acoustics were converted to spectrogram images and transfer learning was applied for their classification into high/medium/low tool wear using four pre-trained deep learning models (SqueezeNet, ResNet50, InceptionV3, GoogLeNet). Moreover, three optimizers (RMSPROP, ADAM, SGDM) were applied to each of the deep learning models to enhance classification accuracies. Primary results indicate that the InceptionV3-RMSPROP obtained the highest testing accuracy of 87.50%, followed by the SqueezeNet-RMSPROP and ResNet50-SGDM at 75.00% and 62.50% respectively. However, SqueezeNet-RMSPROP was determined to be more desirable from a practical machining quality and safety perspective, owing to its greater recall value for the highest tool wear class. The proposed acoustics-texture extraction-transfer learning approach is especially suitable for cost effective tool wear condition monitoring involving limited datasets.

EA-CNN: Enhanced Attention-CNN with Explainable AI for Fruit and Vegetable Classification

The quality of vegetables and fruits are judged by their visual features. Misclassification of fruits and vegetables lead to a financial loss. To prevent the loss, superstores need to classify fruits and vegetables in terms of size, color and shape. To improve the accuracy of models for classifying fruits and vegetables, researchers have introduced various CNN architectures like VGG16, AlexNet, DenseNet-121 and different attention-based variants by compromising models’ complexity leading to high computational cost and overlooking feature’s contribution for interpretable predictions. Additionally, the major drawback in most of the existing works is the utilization of the limited dataset and number of classes for fruit and vegetable classification task instead of the benchmark dataset like Fruit-360 which comprises of 141 distinct fruit and vegetable classes. In this study, an explainable artificial intelligence (XAI) driven enhanced attention-CNN (EA-CNN) is proposed for accomplishing the fruit and vegetable classification task accurately and efficiently. The proposed EA-CNN model (a) detects the class of fruits and vegetable accurately through visual features by manipulating undiscover customized pooling technique and enhanced attention feature extraction mechanism, (b) classifies the fruits and vegetables efficiently through a simplified architecture, (c) provides interpretable predictions through XAI approach. By utilizing entire fruit-360 database, this research noteworthily enhances the model’s capability to classify wide range of fruits and vegetables, thus providing an effective and reliable solution for practical applications. The proposed study outclasses baseline models with regard to accuracy and computational cost on fruit-360 benchmark dataset. EA-CNN achieves better accuracy of 98.1% in smaller number of iterations as compared to benchmark models. Moreover, to further prove the efficiency, generalization ability, robustness, scalability and adaptability of this research, the architecture of proposed EA-CNN model is validated on another real-world dataset named as ‘Fruit Recognition’. The experimental outcomes display that the proposed EA-CNN model produces substantial performance on Fruit-Recognition dataset as well by attaining a generalized accuracy of 96%.

Knowledge assisted machine learning to clarify pore influence on fatigue life of forging/additive hybrid manufactured Ti-17 alloy

Forging/additive hybrid manufactured Ti alloy parts suffer from relatively low fatigue life due to the existence of metallurgical defects in the transition zone, which also brings difficulty to fatigue life modeling. In this work, the synergistic effect of pore size and location on the rotating-bending fatigue life of hybrid manufactured Ti-5Al-2Sn-2Zr-4Mo-4Cr (Ti-17) samples was systematically investigated with the combination of machine learning approaches and physical knowledge. A machine learning framework with a back propagation neural network and generative adversarial network (GAN) was constructed and employed on sparse and limited datasets. A general and interpretable model was obtained with a high level of 90% confidence. In general, the fatigue life of hybrid manufactured Ti-17 alloys decreases with pore size and increases with its distance to surface. Specifically, critical sizes were obtained for near-surface and in-depth pores that have negligible influence on fatigue life of hybrid manufactured samples with respect to pore-free samples. The present work thus provides a systematic platform for the evaluation of the fatigue performance of hybrid manufactured titanium alloys.

Self-supervised learning for small-scale medical imaging dataset

In recent years, advancements in deep learning have dramatically improved performance in medical image analysis, yet these models typically rely on large-scale labeled datasets, which are often unattainable in medical settings due to privacy concerns, limited data availability, and high annotation costs. This study explores the application of self-supervised learning (SSL) techniques to overcome these limitations and effectively utilize small-scale medical imaging datasets. By leveraging SSL, we enable models to learn useful feature representations without requiring extensive labeled data. We investigate various self-supervised approaches, including contrastive learning and masked image modeling, and evaluate their effectiveness on a limited dataset of medical images. Our experiments demonstrate that SSL-based models can achieve competitive performance, even when trained on a fraction of the labeled data typically required for supervised methods. Additionally, we explore the impact of SSL on model robustness and generalization across diverse medical imaging modalities. The findings suggest that self-supervised techniques could reduce dependency on annotated data, paving the way for broader, more scalable applications in medical imaging. This research contributes to the development of efficient, scalable diagnostic tools that can be deployed in data-constrained environments, potentially improving diagnostic accuracy and accessibility in smaller healthcare facilities.

LLMOverTab: Tabular data augmentation with language model-driven oversampling

In recent years, Large Language Model (LLM) have seen significant advancements, attracting attention for their applications in various fields. These models have shown promising results in handling tabular data, especially in cases with limited datasets, by leveraging pre-trained knowledge. However, their effectiveness in addressing imbalanced data in tabular formats is less explored. To bridge this gap, our study introduces LLMOverTab, a novel approach using LLMs for oversampling in imbalanced tabular data. We conducted comprehensive experiments on diverse tabular datasets to assess the effectiveness of LLMOverTab, demonstrating its potential in improving the handling of imbalanced data. The study also explores application of LLMOverTab in zero-shot and few-shot learning contexts, providing insights into its adaptability. Additionally, we analyze the oversampled data, offering reflections on the quality of generated samples. Our research not only showcases the utility of LLMOverTab in managing imbalanced tabular data, but also opens new avenues for the application of language models in various tasks of tabular data. This study adds to the increasing interest in applying LLMs to various task domains. It provides new perspectives for the innovative use of LLMs in structured tabular data fields, highlighting their potential in a range of applications.