Skip to content
Free planNo credit card requiredStart writing
AI for Research home
AI for Research home
Dashboard
Sign inGet started

Search 250 million research papers

Read the abstract, see who wrote it and who cites it, and get the citation in APA, MLA, Chicago or Harvard.

About 5,389 papers for “training sample generation”

Label-Aware Pseudo-Training Sample Generation for Text Classification

Arash Yousefi Jordehi, Seyed Abolghasem Mirroshandel, Owen Rambow · Journal of Artificial Intelligence Research · 2026

Deep learning models excel in various Natural Language Processing (NLP) tasks, but their performance (excluding approaches like zero-shot learning or few-shot learning) relies on ample data, posing challenges in fields with limited datasets. To address the poverty in the size of training data, a number of approaches could be taken, such as multi-task learning and data augmentation. Aiming to leverage Large Language Models (LLMs), we propose a data augmentation algorithm. It subtly alters sentences by inserting random words and utilizes LLMs to find the most fitting replacements within their embedding space. Taking inspiration from Prompt Tuning, the focus shifts from optimizing the input prompt to updating the inserted tokens’ embedding vectors by maximizing the conditional generation probability. This allows for vast sample generation while implicitly benefiting from the knowledge within LLMs. The results from our extensive set of experiments on various benchmark text classification tasks show a substantial improvement over the non-augmented outcomes.

Cited by 0Open access

VR-based Image Training Sample Generation System for Substation Inspection

Hongxi Yu, Lu Yang, Dongqi Li et al. · 2022

In the power industry, the defect detection of power equipment is gradually changing from the traditional manual inspection to intelligent fault identification. The number of substation defect image samples is small, the type is small, and the data is unbalanced. The lack of defect sample data sets in deep learning limits the generalization and accuracy of defect recognition algorithms. In view of the above problems, a simulation software design scheme for substation inspection image sample generation is proposed. The three-dimensional modeling is used to restore the scene and power equipment of the substation site, simulate various faults and defects that may occur in the equipment, and automatically mark the fault point and fault type. The simulation software can quickly generate a large number of substation defect image samples, which provides a new idea for the expansion and improvement of substation inspection image data set.

Cited by 0

Data-Free Post-Training Quantization with Block-wise Enhanced Sample Generation

Ruiyao Zhang, Zhiwei Dong, Huang Long et al. · 2025

Data-free quantization is known for quantizing a pre-trained deep neural network without access to any training data, which applies to many real-world scenarios in that the training data is unavailable due to security, user privacy, or proprietary concerns. Most of the existing data-free quantization methods adopt a generator-quantization framework, which generator network to synthesize fake samples and Quantization-Aware Training (QAT) to quantize model. While the combination of the generator network and QAT can result in good accuracy for quantized models, the diversity of generated samples is lacking and quantizing a single model may take over 10 hours, which contrasts with Post-Training Quantization (PTQ)’s time-saving potential but poor accuracy. In order to address these issues, we have made improvements to the data generation and quantization process. In detail, 1) We propose Generator Exploration Enhancement (GEE) for utilizing the batch normalization statistics and adversarial sample exploration to enhance the quality and diversity of synthetic samples; 2) We introduce Block-wise Sample Generation (BSG) to collectively optimize individual blocks and the generator, leveraging PTQ as a foundation to boost workflow efficiency. Experiment results show that our proposed method improves both the performance and the efficiency of the data-free quantization compared to that of existing methods. Significantly, BSG achieves an 18% accuracy improvement and reduces quantization time by over 50% for 3-bit ResNet-18 in ImageNet tasks, surpassing the current state-of-the-art QAT method.

Cited by 0

Virtual Sample Generation and Ensemble Learning Based Image Source Identification With Small Training Samples

Shiqi Wu, Bo Wang, Jianxiang Zhao et al. · International Journal of Digital Crime and Forensics · 2021

Nowadays, source camera identification, which aims to identify the source camera of images, is quite important in the field of forensics. There is a problem that cannot be ignored that the existing methods are unreliable and even out of work in the case of the small training sample. To solve this problem, a virtual sample generation-based method is proposed in this paper, combined with the ensemble learning. In this paper, after constructing sub-sets of LBP features, the authors generate a virtual sample-based on the mega-trend-diffusion (MTD) method, which calculates the diffusion range of samples according to the trend diffusion theory, and then randomly generates virtual sample according to uniform distribution within this range. In the aspect of the classifier, an ensemble learning scheme is proposed to train multiple SVM-based classifiers to improve the accuracy of image source identification. The experimental results demonstrate that the proposed method achieves higher average accuracy than the state-of-the-art, which uses a small number of samples as the training sample set.

Cited by 9Open access

A Frequency Matching Method for Generation of a Priori Sample Models from Training Images

Katrine Lange, Knud Skou Cordua, Jan Frydendall et al. · 2011

This paper presents a Frequency Matching Method (FMM) for generation of a priori sample models based on training images and illustrates its use by an example. In geostatistics, training images are used to represent a priori knowledge or expectations of models, and the FMM can be used to generate new images that share the same multi-point statistics as a given training image. The FMM proceeds by iteratively updating voxel values of an image until the frequency of patterns in the image matches the frequency of patterns in the training image; making the resulting image statistically indistinguishable from the training image.

Cited by 1

Rule-based training sample generation using Sentinel-2 GCVI time series for winter wheat mapping

Fangjie Li, Inbal Becker‐Reshef, Josef Wagner et al. · Science of Remote Sensing · 2026

Timely and accurate winter wheat mapping is essential for agricultural monitoring and food security. However, efficiently acquiring high-quality training data for supervised classification remains a challenge. In this study, we developed a rule-based method to automatically generate training samples using Sentinel-2 green chlorophyll vegetation index (GCVI) time series. Then, the key phenological periods were identified through feature importance analysis, and spectral features from these periods were used with a Random Forest (RF) classifier to produce 10 m resolution winter wheat distribution maps for Hengshui, Kaifeng, and Xiangyang in 2022 and 2023. To evaluate temporal transferability, the automatically generated training samples from 2022 to 2023 were transferred to subsequent years, enabling winter wheat mapping for 2023 and 2024 based on cross-year training data. Accuracy assessments showed that the proposed method achieved high performance, with average overall accuracy (OA) of 96.04 ± 1.97 % and 94.81 ± 2.14 % in 2022 and 2023, respectively, and average F1 scores of 91.21 % and 90.83 %. The winter wheat maps generated using transferred samples also demonstrated good temporal transferability, and maintained high accuracy, with average OA of 94.06 ± 2.19 % in 2023 and 94.58 ± 2.01 % in 2024. Area estimates from stratified random sampling showed that winter wheat planting areas in Hengshui, Kaifeng, and Xiangyang were 290.8 ± 16.82, 199.16 ± 10.65, and 318.05 ± 44.16 thousand hectares (kha) in 2022, increasing to 379.34 ± 19.75, 209.27 ± 12.68, and 342.02 ± 42.81 kha in 2023, respectively. Compared with existing winter wheat products, the map generated in this study achieved higher classification accuracy and finer spatial detail. Overall, this study provides a practical and effective approach for automatic training sample generation in winter wheat mapping, and offers valuable guidance for large-scale, long-term agricultural monitoring.

Cited by 0Open access

A Good Sample is Hard to Find: Noise Injection Sampling and\n Self-Training for Neural Language Generation Models

Chris Kedzie, Kathleen McKeown · arXiv (Cornell University) · 2019

Deep neural networks (DNN) are quickly becoming the de facto standard\nmodeling method for many natural language generation (NLG) tasks. In order for\nsuch models to truly be useful, they must be capable of correctly generating\nutterances for novel meaning representations (MRs) at test time. In practice,\neven sophisticated DNNs with various forms of semantic control frequently fail\nto generate utterances faithful to the input MR. In this paper, we propose an\narchitecture agnostic self-training method to sample novel MR/text utterance\npairs to augment the original training data. Remarkably, after training on the\naugmented data, even simple encoder-decoder models with greedy decoding are\ncapable of generating semantically correct utterances that are as good as\nstate-of-the-art outputs in both automatic and human evaluations of quality.\n

Cited by 0Open access

CBS-GAN: A Band Selection Based Generative Adversarial Net for Hyperspectral Sample Generation

Yulin Qiao, Mostofa Zaman Mohammad, Yi Li et al. · IOP Conference Series Earth and Environmental Science · 2021

Abstract Sample generation is an effective method to improve the performance of hyperspectral image classification by generating virtual samples for training sample expansion in the training process of classification. However, there are some defects existing in the previous sample generation methods including the lack of spatial information, the redundant generation and the damage of the original spectral components. In this paper, we propose conditional band selection generative adversarial net, named CBS-GAN, to handle this problem. Firstly, the band selection net of CBS-GAN is utilized to avoid redundant bands and keep original spectral information, then the generation net of CBS-GAN generates the spatial-spectral data blocks by selected bands for sample generation. The experiments of classification are also used to demonstrate the availability of virtual samples generated by our method.

Cited by 2Open access

Optical-to-SAR Translation Based on CDA-GAN for High-Quality Training Sample Generation for Ship Detection in SAR Amplitude Images

Baolong Wu, Haonan Wang, Cunle Zhang et al. · Remote Sensing · 2024

Abundant datasets are critical to train models based on deep learning technologies for ship detection applications. Compared with optical images, ship detection based on synthetic aperture radar (SAR) (especially the high-Earth-orbit spaceborne SAR launched recently) lacks enough training samples. A novel cross-domain attention GAN (CDA-GAN) model is proposed for optical-to-SAR translation, which can generate high-quality SAR amplitude training samples of a target by optical image conversion. This high quality includes high geometry structure similarity of the target compared with the corresponding optical image and low background noise around the target. In the proposed model, the cross-domain attention mechanism and cross-domain multi-scale feature fusion are designed to improve the quality of samples for detection based on the generative adversarial network (GAN). Specifically, a cross-domain attention mechanism is designed to simultaneously emphasize discriminative features from optical images and SAR images at the same time. Moreover, a designed cross-domain multi-scale feature fusion module further emphasizes the geometric information and semantic information of the target in a feature graph from the perspective of global features. Finally, a reference loss is introduced in CDA-GAN to completely retain the extra features generated by the cross-domain attention mechanism and cross-domain multi-scale feature fusion module. Experimental results demonstrate that the training samples generated by the proposed CDA-GAN can obtain higher ship detection accuracy using real SAR data than the other state-of-the-art methods. The proposed method is generally available for different orbit SARs and can be extended to the high-Earth-orbit spaceborne SAR case.

Cited by 13Open access

Virtual sample generation method using modified Gaussian model and salient region

Song Liu · 2015

Object detection usually needs large sample set for training. We proposed a virtual sample feature generating method for small sample set. Firstly, a generating model for the subcomponent of sample feature is built by using Gaussian distribution simulation. Secondly, the relationship between subcomponents is taken into consideration and modification of generating model is introduced, which can improve the accuracy of generation. Finally, the salient region information is fused into generation model to expand the universality for different object. According to experiments on multi-database, our method effectively improved the detecting rate with small training set.

Cited by 0

An Approach to Generating the Training Sample for Bottom Objects Recognition Neural Networks

Valentina Bykova, Andrey I. Mashoshin, Alexey A. Pavlov · 2023

Detection of bottom objects is one of the key tasks of an autonomous underwater vehicle. Neural networks are being increasingly used for solving these problems. Application of neural networks to object recognition is always associated with the problem of training sample generation. The lack of training material is particularly relevant in processing the hydroacoustic and optical sea bottom survey data. The paper describes an approach consisting in generating the simulated data. The generation of a training sample is based on three-dimensional modeling of the bottom surface and objects located on it, and the use of hydroacoustic calculations for each of the bottom monitoring tools. The paper describes the transfer of training for a neural network and presents the results of checking its operation on a test sample of real data.

Cited by 0

Generative Replay Inspired by Hippocampal Memory Indexing for Continual Language Learning

Aru Maekawa, Hidetaka Kamigaito, Kotaro Funakoshi et al. · 2023

Continual learning aims to accumulate knowledge to solve new tasks without catastrophic forgetting for previously learned tasks. Research on continual learning has led to the development of generative replay, which prevents catastrophic forgetting by generating pseudo-samples for previous tasks and learning them together with new tasks. Inspired by the biological brain, we propose the hippocampal memory indexing to enhance the generative replay by controlling sample generation using compressed features of previous training samples. It enables the generation of a specific training sample from previous tasks, thus improving the balance and quality of generated replay samples. Experimental results indicate that our method effectively controls the sample generation and consistently outperforms the performance of current generative replay methods.

Cited by 4Open access

SeqPATE: Differentially Private Text Generation via Knowledge Distillation

Tian, Zhiliang, Zhao, Yingxiu, Huang, Ziyue et al. · Rare & Special e-Zone (The Hong Kong University of Science and Technology) · 2022

Protecting the privacy of user data is crucial for text generation models, which can leak sensitive information during generation. Differentially private (DP) learning methods provide guarantees against identifying the existence of a training sample from model outputs. PATE is a recent DP learning algorithm that achieves high utility with strong privacy protection on training samples. However, text generation models output tokens sequentially in a large output space; the classic PATE algorithm is not customized for this setting. Furthermore, PATE works well to protect sample-level privacy, but is not designed to protect phrases in samples. In this paper, we propose SeqPATE, an extension of PATE to text generation that protects the privacy of individual training samples and sensitive phrases in training data. To adapt PATE to text generation, we generate pseudo-contexts and reduce the sequence generation problem to a next-word prediction problem. To handle the large output space, we propose a candidate filtering strategy to dynamically reduce the output space, and refine the teacher aggregation of PATE to avoid low agreement due to voting for a large number of candidates. To further reduce privacy losses, we use knowledge distillation to reduce the number of teacher queries. The experiments verify the effectiveness of SeqPATE in protecting both training samples and sensitive phrases.

Cited by 4

Regional Forest Wildfire Mapping Through Integration of Sentinel-2 and Landsat 8 Data in Google Earth Engine with Semi-Automatic Training Sample Generation

Yue Chen, Weili Kou, Xiong Yin et al. · Remote Sensing · 2025

Accurate mapping of burned forest areas in mountainous regions is essential for wildfire assessment and post-fire ecological management. This study develops an FS-SNIC-ML workflow that integrates multi-source optical fusion, semi-automatic sample generation, feature selection, and object-based machine-learning classification to support reliable burned-area mapping under complex terrain conditions. A pseudo-invariant feature (PIFS)-based fusion of Sentinel-2 and Landsat 8 imagery was employed to generate cloud-free, gap-free, and spectrally consistent pre- and post-fire reflectance datasets. Burned and unburned samples were constructed using a semi-automatic SAM–GLCM–PCA–Otsu procedure and county-level stratified sampling to ensure spatial representa-tiveness. Feature selection using LR, RF, and Boruta identified dNBR, dNDVI, and dEVI as the most discriminative variables. Within the SNIC-supported GEOBIA framework, four classifiers were evaluated; RF performed best, achieving overall accuracies of 92.02% for burned areas and 94.04% for unburned areas, outperforming SVM, CART, and KNN. K-means clustering of dNBR revealed spatial variation in fire conditions, while geographical detector analysis showed that NDVI, temperature, soil moisture, and their pairwise interactions were the dominant drivers of wildfire hotspot density. The proposed workflow provides an effective and transferable approach for high-precision burned-area extraction and quantification of wildfire-driving factors in mountainous forest regions.

Cited by 0Open access

A Method for Generating Small Sample Data and Evaluating the Effectiveness of Combat Weapons Based on Conditional Adversarial Nets

Jiayong Fang, Wenmin Zhang, Hongbo Mang et al. · 2023

A neural network has received attention in weapon performance evaluation, one-dimensional combat data in the standard sample generation method, poor sample feature extraction generation, and the network training process could be more convenient. This paper proposes a small sample data generation based on CGAN and weapon combat effectiveness evaluation method - CMGAN algorithm. Based on the advantages of conditional confrontation networks in image recognition, the sample data of one-dimensional combat weapons are projected into two-dimensional space. A two-dimensional convolution neural network is used to extract sample features. Then further optimize and screen the generated samples, and finally, use the generated virtual samples for GA-BP (genetic algorithm back-propagation) neural network effectiveness evaluation of combat weapons. The results show that the improved CMGAN algorithm can generate higher-quality data. It effectively improves the accuracy of combat weapon effectiveness assessment and network training speed and has good engineering application feasibility.

Cited by 0