Prevent Data Leakage Research Articles

Enhanced Dual-Layer Video Steganographic Approach: Enhancing Security and Imperceptibility

In a digital age where the role of digital information is continually expanding, the imperative to ensure secure communication and data protection has intensified. Steganography, a field that conceals sensitive information within apparently innocuous carriers like multimedia files, holds promise for achieving this goal. Recent times have seen a growing interest in video-based steganography, driven by the proliferation of videos online. However, a review has unveiled a noteworthy gap in efforts to increase the safety of video steganography employing the popular Least Significant Bit (LSB) process. LSB steganography is at risk to exploits that could jeopardize the confidentiality of the concealed message. In this research, we present a video steganographic solution achieved through the integration of advanced encryption algorithms and a dual approach involving randomized frame selection using the Fisher-Yates method and pixel selection utilizing the 8-directional technique. Additionally, we have devised an upgraded imperceptible video steganographic approach in the spatial domain, capitalizing on an XOR-based LSB substitution embedding approach. The outcomes underscore the effectiveness of our proposed methodology, revealing significant imperceptibility when compared to recent solutions, thereby reinforcing data security against unauthorized access. This comprehensive solution holds wide-ranging applications across various domains, providing heightened security and confidentiality for sensitive information. Notable applications include secure communications within intelligence agencies, safeguarding patient data in medical sciences, ensuring privacy in video surveillance, fortifying video transmission against errors and data breaches, securing business communications, preventing data leakage, and enhancing rights management for digital content.

Design of energy management strategies for shared energy storage microgrid based on smart contracts under privacy protection

Park microgrids, valued for their efficiency and flexibility, require privacy-conscious energy management to ensure a trusted scheduling and trading environment. This paper, focusing on park microgrids with shared energy storage, designs an energy management strategy that comprehensively considers shared energy storage, scheduling transparency, and privacy security. First, a blockchain-based energy management platform is established, forming an energy dispatch consensus committee to execute decentralized scheduling management and decision-making. Next, an optimized energy scheduling smart contract for park microgrids is designed, considering Time-of-Use (ToU) pricing and storage arbitrage to formulate the day-ahead electricity purchase and sales plans as well as the shared energy storage operation plans. Then, a privacy protection strategy based on the Shamir secret sharing scheme is proposed, effectively preventing data leakage during blockchain interactions. Finally, through case analysis, the superiority of the proposed method in microgrid optimized scheduling, data tamper-resistance, and privacy protection is demonstrated.

Guiding questions to avoid data leakage in biological machine learning applications.

Machine learning methods for extracting patterns from high-dimensional data are very important in the biological sciences. However, in certain cases, real-world applications cannot confirm the reported prediction performance. One of the main reasons for this is data leakage, which can be seen as the illicit sharing of information between the training data and the test data, resulting in performance estimates that are far better than the performance observed in the intended application scenario. Data leakage can be difficult to detect in biological datasets due to their complex dependencies. With this in mind, we present seven questions that should be asked to prevent data leakage when constructing machine learning models in biological domains. We illustrate the usefulness of our questions by applying them to nontrivial examples. Our goal is to raise awareness of potential data leakage problems and to promote robust and reproducible machine learning-based research in biology.

SLO-Net: Enhancing Multiple Sclerosis Diagnosis Beyond Optical Coherence Tomography Using Infrared Reflectance Scanning Laser Ophthalmoscopy Images.

Several machine learning studies have used optical coherence tomography (OCT) for multiple sclerosis (MS) classification with promising outcomes. Infrared reflectance scanning laser ophthalmoscopy (IR-SLO) captures high-resolution fundus images, commonly combined with OCT for fixed B-scan positions. However, no machine learning research has utilized IR-SLO images for automated MS diagnosis. This study utilized a dataset comprised of IR-SLO images and OCT data from Isfahan, Iran, encompassing 32 MS and 70 healthy individuals. A number of convolutional neural networks (CNNs)-namely, VGG-16, VGG-19, ResNet-50, ResNet-101, and a custom architecture-were trained with both IR-SLO images and OCT thickness maps as two separate input datasets. The highest performing models for each modality were then integrated to create a bimodal model that receives the combination of OCT thickness maps and IR-SLO images. Subject-wise data splitting was employed to prevent data leakage among training, validation, and testing sets. Overall, images of the 102 patients from the internal dataset were divided into test, validation, and training subsets. Subsequently, we employed a bootstrapping approach on the training data through iterative sampling with replacement. The performance of the proposed bimodal model was evaluated on the internal test dataset, demonstrating an accuracy of 92.40% ± 4.1% (95% confidence interval [CI], 83.61-98.08), sensitivity of 95.43% ± 5.75% (95% CI, 83.71-100.0), specificity of 92.82% ± 3.72% (95% CI, 81.15-96.77), area under the receiver operating characteristic (AUROC) curve of 96.99% ± 2.99% (95% CI, 86.11-99.78), and area under the precision-recall curve (AUPRC) of 97.27% ± 2.94% (95% CI, 86.83-99.83). Furthermore, to assess the model generalization ability, we examined its performance on an external test dataset following the same bootstrap methodology, achieving promising results, with accuracy of 85.43% ± 0.08% (95% CI, 71.43-100.0), sensitivity of 97.33% ± 0.06% (95% CI, 83.33-100.0), specificity of 84.6% ± 0.10% (95% CI, 71.43-100.0), AUROC curve of 99.67% ± 0.02% (95% CI, 95.63-100.0), and AUPRC of 99.65% ± 0.02% (95% CI, 94.90-100.0). Incorporating both modalities improves the performance of automated diagnosis of MS, showcasing the potential of utilizing IR-SLO as a complementary tool alongside OCT. Should the results of our proposed bimodal model be validated in future work with larger and more diverse datasets, diagnosis of MS based on both OCT and IR-SLO can be reliably integrated into routine clinical practice.

Empowering Data Owners: An Efficient and Verifiable Scheme for Secure Data Deletion

Cloud services have attracted numerous enterprises, organizations, and individual users due to their exceptional computing power and almost limitless storage capacity. A vast amount of business data and private data are continuously uploaded to the cloud platform, driven by a series of attractive services offered by the cloud. Unfortunately, once data is uploaded to the cloud, its owner has no way of ensuring that it is actually deleted as intended. This obviously increases the concerns of data owners about the security of their data, because it is related to the privacy of users. Therefore, there must be a reliable solution to prove that data is deleted as requested by users, to prevent data leakage or abuse. In existing data deletion schemes, most are designed based on cryptographic knowledge rather than erasure or overwrite techniques, in order not to cause incalculable damage to the storage medium. However, most cryptographic-based data deletion schemes, particularly attribute-based encryption, involve numerous complex bilinear mapping operations, which are expensive for most devices. To address this issue, the paper proposes an Efficient and Verifiable Scheme for Secure Data Deletion (EVSD). Firstly, Elliptic Curve Cryptography (ECC) is introduced to achieve efficient encryption of data. Then, leveraging Linear Secret Sharing Scheme (LSSS), fine-grained data deletion policies supporting logical operations are implemented. Finally, the deletion of the data is efficiently verified using the root of the Merkle Hash Tree (MHT) generated by the defined illegal and legal attributes, while the deletion proof is also generated. Satisfactorily, security analysis shows that the EVSD scheme is much more advantageous compared to existing schemes, and a trait likewise is also observed in the performance evaluation.

Advancing deep learning-based acoustic leak detection methods towards application for water distribution systems from a data-centric perspective

Against the backdrop of severe leakage issue in water distribution systems (WDSs), numerous researchers have focused on the development of deep learning-based acoustic leak detection technologies. However, these studies often prioritize model development while neglecting the importance of data. This research explores the impact of data augmentation techniques on enhancing deep learning-based acoustic leak detection methods. Five random transformation-based methods—jittering, scaling, warping, iterated amplitude adjusted Fourier transform (IAAFT), and masking—are proposed. Jittering, scaling, warping, and IAAFT directly process original signals, while masking operating on time-frequency spectrograms. Acoustic signals from a real-world WDS are augmented, and the efficacy is validated using convolutional neural network classifiers to identify the spectrograms of acoustic signals. Results indicate the importance of implementing data augmentation before data splitting to prevent data leakage and overly optimistic outcomes. Among the techniques, IAAFT stands out, significantly increasing data volume and diversity, improving recognition accuracy by over 7%. Masking enhances performance mainly by compelling the classifier to learn global features of the spectrograms. Sequential application of IAAFT and masking further strengthens leak detection performance. Furthermore, when applying a complex model to acoustic leakage detection through transfer learning, data augmentation can also enhance the effectiveness of transfer learning. These findings advance artificial intelligence-driven acoustic leak detection technology from a data-centric perspective towards more mature applications.

A Deep Learning-Based Method for Preventing Data Leakage in Electric Power Industrial Internet of Things Business Data Interactions.

In the development of the Power Industry Internet of Things, the security of data interaction has always been an important challenge. In the power-based blockchain Industrial Internet of Things, node data interaction involves a large amount of sensitive data. In the current anti-leakage strategy for power business data interaction, regular expressions are used to identify sensitive data for matching. This approach is only suitable for simple structured data. For the processing of unstructured data, there is a lack of practical matching strategies. Therefore, this paper proposes a deep learning-based anti-leakage method for power business data interaction, aiming to ensure the security of power business data interaction between the State Grid business platform and third-party platforms. This method combines named entity recognition technologies and comprehensively uses regular expressions and the DeBERTa (Decoding-enhanced BERT with disentangled attention)-BiLSTM (Bidirectional Long Short-Term Memory)-CRF (Conditional Random Field) model. This method is based on the DeBERTa (Decoding-enhanced BERT with disentangled attention) model for pre-training feature extraction. It extracts sequence context semantic features through the BiLSTM, and finally obtains the global optimal through the CRF layer tag sequence. Sensitive data matching is performed on interactive structured and unstructured data to identify privacy-sensitive information in the power business. The experimental results show that the F1 score of the proposed method in this paper for identifying sensitive data entities using the CLUENER 2020 dataset reaches 81.26%, which can effectively prevent the risk of power business data leakage and provide innovative solutions for the power industry to ensure data security.

A multifaceted survey on privacy preservation of federated learning: progress, challenges, and opportunities

Federated learning (FL) refers to a system of training and stabilizing local machine learning models at the global level by aggregating the learning gradients of the models. It reduces the concern of sharing the private data of participating entities for statistical analysis to be carried out at the server. It allows participating entities called clients or users to infer useful information from their raw data. As a consequence, the need to share their confidential information with any other entity or the central entity called server is eliminated. FL can be clearly interpreted as a privacy-preserving version of traditional machine learning and deep learning algorithms. However, despite this being an efficient distributed training scheme, the client’s sensitive information can still be exposed to various security threats from the shared parameters. Since data has always been a major priority for any user or organization, this article is primarily concerned with discussing the significant problems and issues relevant to the preservation of data privacy and the viability and feasibility of several proposed solutions in the FL context. In this work, we conduct a detailed study on FL, the categorization of FL, the challenges of FL, and various attacks that can be executed to disclose the users’ sensitive data used during learning. In this survey, we review and compare different privacy solutions for FL to prevent data leakage and discuss secret sharing (SS)-based security solutions for FL proposed by various researchers in concise form. We also briefly discuss quantum federated learning (QFL) and privacy-preservation techniques in QFL. In addition to these, a comparison and contrast of several survey works on FL is included in this work. We highlight the major applications based on FL. We discuss certain future directions pertaining to the open issues in the field of FL and finally conclude our work.

Are Next-Generation Pathogenicity Predictors Applicable to Cancer?

Next-generation pathogenicity predictors are designed to identify pathogenic mutations in genetic disorders but are increasingly used to detect driver mutations in cancer. Despite this, their suitability for cancer is not fully established. Here we have assessed the effectiveness of next-generation pathogenicity predictors when applied to cancer by using a comprehensive experimental benchmark of cancer driver and neutral mutations. Our findings indicate that state-of-the-art methods AlphaMissense and VARITY demonstrate commendable performance despite generally underperforming compared to cancer-specific methods. This is notable considering that these methods do not explicitly incorporate cancer-related data in their training and have made concerted efforts to prevent data leakage from the human-curated training and test sets. Nevertheless, it should be mentioned that a significant limitation of using pathogenicity predictors for cancer arises from their inability to detect cancer potential driver mutations specific for a particular cancer type.

A privacy-preserving distributed energy management framework based on vertical federated learning-based smart data cleaning for smart home electricity data

Nonintrusive load monitoring (NILM) is a part of home energy management systems ((H)EMSs), which can advance the systems to leverage and use gathered electricity data (load data) to achieve cost-effective load monitoring for efficient (residential) demand-side management (DSM). Load data monitored by existing (H)EMSs, whose load monitoring is based on intrusive load monitoring, can be used as a preliminary stage for NILM to be accelerated in energy management for the improvement of its practicality. However, they can be polluted. For example, power-consumption data gathered by appliance-level smart plugs may have an incorrect appliance label (a mislabeled appliance ID). Polluted data must be preprocessed through a data cleaning process before they are leveraged and used. Data-cleaning approaches that improve data quality should be developed considering a decentralized paradigm, because a centralized data-cleaning paradigm cannot be applied to future edge-based IoT applications. Additionally, preventing data leakage is paramount in data management for numerous field applications. This study develops a privacy-preserving distributed energy management framework based on vertical federated learning for smart data cleaning as a demonstrative application of smart home electricity data toward distributed NILM. The framework developed in this study with the advent of AI methodology can achieve smart data cleaning for further distributed NILM to be accelerated for its practical applications; its feasibility and effectiveness have been verified experimentally.

The integration algorithm of digital resources in business administration based on cluster analysis

The field of business management involves a large amount of data and information sources, including market data, customer data, supply chain data, etc. In order to quantify and analyze different resources, help enterprises better plan and allocate resources, and improve resource utilization efficiency, a clustering analysis based digital resource integration algorithm for business management is studied. Build a business management digital resource integration framework, including data layer, integration layer, and storage layer, to integrate and store data from different sources of business management databases, thereby facilitating unified management and utilization of digital resources by enterprises. The data layer collects data from different business management databases and stores it in the database according to different sources; The integration layer preprocesses the collected data, simply fixes errors and missing information in the data, and improves data quality. Adopting a feature extraction method based on the projection direction uncorrelation strategy of the labeled power set conversion method, the useful feature information of digital resources in enterprise management can be effectively extracted; Based on the two-step clustering analysis method, business management digital resources are clustered according to similar characteristics to complete the classification and integration of business management digital resources, and improve the efficiency of resource utilization; The storage layer adopts the Security Information Diffusion Algorithm (IDA) storage model to store integrated and classified digital resources managed by enterprises, ensuring data security and effectively preventing data leakage and illegal access. The experimental results show that the digital resource structure of business management integrated by this algorithm is clear, with a data redundancy of less than 8% and a difference of less than 11%. The time consumption for data integration is less than 2.11 minutes, indicating good resource integration ability.

Adversarial Machine-Learning-Enabled Anonymization of OpenWiFi Data

Data privacy and protection through anonymization is a critical issue for network operators or data owners before it is forwarded for other possible use of data. With the adoption of Artificial Intelligence (AI), data anonymization augments the likelihood of covering up necessary sensitive information; preventing data leakage and information loss. OpenWiFi networks are vulnerable to any adversary who is trying to gain access or knowledge on traffic regardless of the knowledge possessed by data owners. The odds for discovery of actual traffic information is addressed by applied conditional tabular generative adversarial network (CTGAN). CTGAN yields synthetic data; which disguises as actual data but fostering hidden acute information of actual data. In this paper, the similarity assessment of synthetic with actual data is showcased in terms of clustering algorithms followed by a comparison of performance for unsupervised cluster validation metrics. A well-known algorithm, K-means outperforms other algorithms in terms of similarity assessment of synthetic data over real data while achieving nearest scores 0.634, 23714.57, and 0.598 as Silhouette, Calinski and Harabasz and Davies Bouldin metric respectively. On exploiting a comparative analysis in validation scores among several algorithms, K-means forms the epitome of unsupervised clustering algorithms ensuring explicit usage of synthetic data at the same time a replacement for real data. Hence, the experimental results aim to show the viability of using CTGAN-generated synthetic data in lieu of publishing anonymized data to be utilized in various applications.

A Blockchain-Based Fairness Guarantee Approach for Privacy-Preserving Collaborative Training in Computing Force Network

The advent of the big data era has brought unprecedented data demands. The integration of computing resources with network resources in the computing force network enables the possibility of distributed collaborative training. However, unencrypted collaborative training is vulnerable to threats such as gradient inversion attacks and model theft. To address this issue, the data in collaborative training are usually protected by cryptographic methods. However, the semantic meaninglessness of encrypted data makes it difficult to prevent potential data poisoning attacks and free-riding attacks. In this paper, we propose a fairness guarantee approach for privacy-preserving collaborative training, employing blockchain technology to enable participants to share data and exclude potential violators from normal users. We utilize a cryptography-based secure aggregation method to prevent data leakage during blockchain transactions, and employ a contribution evaluation method for encrypted data to prevent data poisoning and free-riding attacks. Additionally, utilizing Shamir’s secret sharing for secret key negotiation within the group, the negotiated key is directly introduced as noise into the model, ensuring the encryption process is computationally lightweight. Decryption is efficiently achieved through the aggregation of encrypted models within the group, without incurring additional computational costs, thereby enhancing the computational efficiency of the encryption and decryption processes. Finally, the experimental results demonstrate the effectiveness and efficiency of our proposed approach.

Securing Medical Data: The Integration of Advanced Encryption Standard and Blockchain

Objective: The research is to ensure the security of medical data and prevent data breaches. To achieve this objective, An efficient solution was proposed - the integration of blockchain technology with Advanced Encryption Standard (AES) encryption techniques to provide medical data security. Methods: The cryptography which served as the cornerstone of information security provides robust encryption and decryption methods to ensure data confidentiality and integrity. At the same time, the attributes of tamper-proof and decentralization in blockchain ensure the true validity and security of data. We utilized the AES encryption algorithm to secure data and integrate the encrypted data into the blockchain technology, providing a solid and reliable guarantee for medical data privacy. Results: The AES encryption algorithm was apply to blockchain technology. With this integrated solution, the security of medical data and effectively prevent data leakage was ensured. Conclusion: This solution not only addressed the issue of medical data leakage, but also provided a solid foundation for the future development of information security. Integrating blockchain technology with AES encryption provided a reliable solution for medical data security, giving both patients and medical professionals greater confidence in data security. At the same time, the combination of the AES encryption algorithm and blockchain technology is being widely applied in various other fields.

Application of privacy protection technology to healthcare big data.

With the advent of the big data era, data security issues are becoming more common. Healthcare organizations have more data to use for analysis, but they lose money every year due to their inability to prevent data leakage. To overcome these challenges, research on the use of data protection technologies in healthcare is actively underway, particularly research on state-of-the-art technologies, such as federated learning announced by Google and blockchain technology, which has recently attracted attention. To learn about these research efforts, we explored the research, methods, and limitations of the most widely used privacy technologies. After investigating related papers published between 2017 and 2023 and identifying the latest technology trends, we selected related papers and reviewed related technologies. In the process, four technologies were the focus of this study: blockchain, federated learning, isomorphic encryption, and differential privacy. Overall, our analysis provides researchers with insight into privacy technology research by suggesting the limitations of current privacy technologies and suggesting future research directions.

Optimizing age-related hearing risk predictions: an advanced machine learning integration with HHIE-S

ObjectivesThe elderly are disproportionately affected by age-related hearing loss (ARHL). Despite being a well-known tool for ARHL evaluation, the Hearing Handicap Inventory for the Elderly Screening version (HHIE-S) has only traditionally been used for direct screening using self-reported outcomes. This work uses a novel integration of machine learning approaches to improve the predicted accuracy of the HHIE-S tool for ARHL in older adults.MethodsWe employed a dataset that was gathered between 2016 and 2018 and included 1,526 senior citizens from several Taipei City Hospital branches. 80% of the data were used for training (n = 1220) and 20% were used for testing (n = 356). XGBoost, Gradient Boosting, and LightGBM were among the machine learning models that were only used and assessed on the training set. In order to prevent data leakage and overfitting, the Light Gradient Boosting Machine (LGBM) model—which had the greatest AUC of 0.83 (95% CI 0.81–0.85)—was then only used on the holdout testing data.ResultsOn the testing set, the LGBM model showed a strong AUC of 0.82 (95% CI 0.79–0.86), far outperforming conventional techniques. Notably, several HHIE-S items and age were found to be significant characteristics. In contrast to traditional HHIE research, which concentrates on the psychological effects of hearing loss, this study combines cutting-edge machine learning techniques—specifically, the LGBM classifier—with the HHIE-S tool. The incorporation of SHAP values enhances the interpretability of the model's predictions and provides a more comprehensive comprehension of the significance of various aspects.ConclusionsOur methodology highlights the great potential that arises from combining machine learning with validated hearing evaluation instruments such as the HHIE-S. Healthcare practitioners can anticipate ARHL more accurately thanks to this integration, which makes it easier to intervene quickly and precisely.

Multimedia image evaluation based on blockchain, visual communication design and color balance optimization

The relationship between image processing and image analysis is inseparable. With the increasing demand for multimedia visual images, the quality of image analysis is also increasing. However, in image processing and computer vision tasks, protecting users' privacy and preventing data leakage and abuse are not handled well. Image enhancement and nonlinear image color balance algorithm are applied to improve the visual quality of multimedia visual images and make them clearer and fuller. The article utilized image enhancement and a non-linear image color balance algorithm to improve the processing effect before visual image analysis. It also utilized the encryption mechanism of blockchain technology to detect the similarity of multimedia visual images. By comparing the feature points of the images, similar images were matched to address the copyright issue of the images. After experimental testing, the effect of image enhancement is significant, and the histogram of image equalization is significantly better than the original image. In the experiment of image analysis, the computer accurately classified visual images with different attributes. Finally, in the similarity detection algorithm of blockchain, the test results showed that when the number of image transactions reaches 500, the difference hash algorithm takes 1.13 s and 0.78 s to calculate the similarity comparison between the original and secondary images. The differential hash algorithm of blockchain is significantly superior to the Message-Digest Algorithm (MD5) in terms of computational speed and resource consumption. It has better image similarity detection performance and can also provide better image copyright protection mechanisms.

Analysis of Data Leakage Detection Algorithms

Abstract: Data refers to information about a particular mention. There are several types of data and they can be stored either physically or virtually. The use of Cloud Computing has gained popularity in many IT organizations in order to store data effectively. With the growth of the population, data management becomes more critical. A management system relies heavily on the protection of data and preventing leakage. A data leak not only damages the company's reputation but also damages its customers' trust. It is high time to address the issue, especially considering previous incidents like Yahoo's data breach that compromised over 2 billion accounts, Friend Finder Networks’ leak that caused 412 million accounts to be compromised, and more. In this paper, we present methodologies for detecting and preventing data leakage in cloud computing. The cloud platform is being used to perform various algorithms, which will lead to an efficient detection of data leakage, as well as an immediate identification of guilt.

Application of Big Data Privacy Protection Based on Edge Computing in the Prediction of Martial Arts Training Movement Trajectory

With the development of science and technology, edge computing and big data privacy protection are more and more widely used in various fields. The application of big data privacy protection based on edge computing in the prediction of sport trajectories for martial arts training shows good performance and privacy protection. Edge computing can process and analyze data in real time to improve the accuracy and efficiency of sport trajectory prediction. Big data privacy protection can ensure the security of athletes' personal information and training data and prevent data leakage and misuse. However, existing related works still have some deficiencies in data processing speed, accuracy, and privacy protection. In this paper, the authors address these issues and propose an edge computing-based big data privacy protection method to improve the accuracy and security of sport trajectory prediction for martial arts training.

A federated pedestrian trajectory prediction model with data privacy protection

Pedestrian trajectory prediction is essential for self-driving vehicles, social robots, and intelligent monitoring applications. Diverse trajectory data is critical for high-accuracy trajectory prediction. However, the trajectory data is captured in scattered scenes, which can cause the problem of data island. Furthermore, artificial aggregation of trajectory data suffers from the risk of data leakage, ignoring the rule of privacy protection. We propose a multi-scene federated trajectory prediction (Fed-TP) method to solve the above problems. As our key contribution, a destination-oriented LSTM (Long Short-Term Memory)-based trajectory prediction (DO-TP) network is proposed in each scene to forecast future trajectories in an encoder-decoder manner. The independent training using trajectory data in each scene can prevent data leakage and achieves high privacy security. As another key contribution, a federated learning framework is introduced to break the scene limitation by conducting distributed collaborative training. The performance of different federated learning methods is compared on public datasets, including ETH, UCY, and Stanford Drone Dataset (SDD). Compared with FedAvg and FedProx, FedAtt is more suitable for pedestrian trajectory prediction. Experimental results demonstrate that the proposed method has better data privacy security than directly training on multiple scenes and superior prediction performance than training on a single scene.