Data Distribution Research Articles

Robust Audio-Visual Contrastive Learning for Proposal-Based Self-Supervised Sound Source Localization in Videos.

By observing a scene and listening to corresponding audio cues, humans can easily recognize where the sound is. To achieve such cross-modal perception on machines, existing methods take advantage of the maps obtained by interpolation operations to localize the sound source. As semantic object-level localization is more attractive for prospective practical applications, we argue that these map-based methods only offer a coarse-grained and indirect description of the sound source. Additionally, these methods utilize a single audio-visual tuple at a time during self-supervised learning, causing the model to lose the crucial chance to reason about the data distribution of large-scale audio-visual samples. Although the introduction of Audio-Visual Contrastive Learning (AVCL) can effectively alleviate this issue, the contrastive set constructed by randomly sampling is based on the assumption that the audio and visual segments from all other videos are not semantically related. Since the resulting contrastive set contains a large number of faulty negatives, we believe that this assumption is rough. In this paper, we advocate a novel proposal-based solution that directly localizes the semantic object-level sound source, without any manual annotations. The Global Response Map (GRM) is incorporated as an unsupervised spatial constraint to filter those instances corresponding to a large number of sound-unrelated regions. As a result, our proposal-based Sound Source Localization (SSL) can be cast into a simpler Multiple Instance Learning (MIL) problem. To overcome the limitation of random sampling in AVCL, we propose a novel Active Contrastive Set Mining (ACSM) to mine the contrastive sets with informative and diverse negatives for robust AVCL. Our approaches achieve state-of-the-art (SOTA) performance when compared to several baselines on multiple SSL datasets with diverse scenarios.

Advances and Challenges in Meta-Learning: A Technical Review.

Meta-learning empowers learning systems with the ability to acquire knowledge from multiple tasks, enabling faster adaptation and generalization to new tasks. This review provides a comprehensive technical overview of meta-learning, emphasizing its importance in real-world applications where data may be scarce or expensive to obtain. The article covers the state-of-the-art meta-learning approaches and explores the relationship between meta-learning and multi-task learning, transfer learning, domain adaptation and generalization, self-supervised learning, personalized federated learning, and continual learning. By highlighting the synergies between these topics and the field of meta-learning, the article demonstrates how advancements in one area can benefit the field as a whole, while avoiding unnecessary duplication of efforts. Additionally, the article delves into advanced meta-learning topics such as learning from complex multi-modal task distributions, unsupervised meta-learning, learning to efficiently adapt to data distribution shifts, and continual meta-learning. Lastly, the article highlights open problems and challenges for future research in the field. By synthesizing the latest research developments, this article provides a thorough understanding of meta-learning and its potential impact on various machine learning applications. We believe that this technical overview will contribute to the advancement of meta-learning and its practical implications in addressing real-world problems.

Addressing Data Distribution Shifts in Online Machine Learning Powered Smart City Applications Using Augmented Test-Time Adaptation

Addressing Data Distribution Shifts in Online Machine Learning Powered Smart City Applications Using Augmented Test-Time Adaptation

On Data Distribution Leakage in Cross-Silo Federated Learning

On Data Distribution Leakage in Cross-Silo Federated Learning

Prevalence of the crayfish plague pathogen in red swamp crayfish populations in western France: How serious is the risk for the native white-clawed crayfish?

The crayfish plague pathogen Aphanomyces astaci has been implicated in a number of mass mortalities and irreversible population declines of native crayfish across Europe. At present, the reservoirs of the pathogen in Europe are mainly populations of invasive North American crayfish species. In southwestern Europe, including France, a particularly widespread invader is the red swamp crayfish Procambarus clarkii. Recent distribution data confirm that P. clarkii is present in at least 75 French departments, i.e. more than 78% of those in metropolitan France. We analysed the prevalence and pathogen load of A. astaci in 42 populations of this species in western France (Nouvelle Aquitaine region), where the species is most densely distributed, particularly in a wide range of environments around the Gironde estuary. The pathogen was detected by two different quantitative PCR assays in more than three quarters of the populations studied (34 out of 42); 163 out of 480 analysed crayfish individuals tested positive for the presence of A. astaci. In most cases, individual infection levels were very low, detectable with quantitative PCR but not sufficient for pathogen genotyping. In seven P. clarkii individuals from four populations, however, we were able to assess A. astaci variation by microsatellite markers and sequencing of mitochondrial markers. All these host specimens carried A. astaci genotype group D, haplotype d1, which has caused the majority of crayfish plague outbreaks in neighbouring Spain. In contrast, the French outbreaks genotyped to date (including eight newly analysed in this study) were mostly caused by strains of genotype group B, specific to the signal crayfish Pacifastacus leniusculus. Haplotype d1 found in P. clarkii was involved in one of the newly characterised outbreaks. Our study confirms that P. clarkii is a potentially important reservoir of the crayfish plague pathogen in France, but not the main source of the pathogen in mass mortalities of A. pallipes, probably due to different ecological requirements of the different invasive host crayfish. However, as P. clarkii continues to spread, the threat posed by this species to native crayfish is likely to increase.

Federated Bayesian network approach for cross-regional air pollution classification: a case study of the Beijing-Tianjin-Hebei region.

Although machine learning methods have enabled considerable progress in air quality assessment, challenges persist regarding data privacy, cross-regional data processing, and model generalization. To address these issues, we introduce an advanced federated Bayesian network (FBN) approach. By integrating federated learning, adaptive optimization algorithms, and homomorphic encryption technologies, we substantially enhanced the efficiency and security of cross-regional air quality data processing. The novelty of this research lies in the improvements implemented in federated learning for air quality data analysis, particularly in distributed model training optimization and data consistency. Through the integration of adaptive structural modification strategies and simulated annealing immune optimization algorithms, we markedly enhanced the structural learning accuracy of the Bayesian network, resulting in a 20% improvement in prediction accuracy. Moreover, employing homomorphic encryption ensured data transmission security and confidentiality. In our Beijing-Tianjin-Hebei case study, our method demonstrated a 15% improvement in air quality classification accuracy compared to conventional methods and exhibited superior interpretability in analyzing environmental factor interactions. We quantified complex air pollution patterns across regions and found that a 30% fluctuation in the air quality index correlated with NO2 concentrations. We also observed a moderate positive correlation between specific pollutant indicators in Hebei Province and Tianjin and changes in air quality. Additionally, the FBN exhibited better operational efficiency and data confidentiality than other machine learning models in handling large-scale and multisource environmental data. Our FBN approach presents a novel perspective for environmental monitoring and assessment, vital for understanding complex air pollution patterns and formulating future ecological protection policies.

Uncertainty assessment in unsupervised machine-learning methods for deepwater channel seismic facies using outcrop-derived 3D models and synthetic seismic data

Unsupervised machine-learning (ML) techniques have been widely applied to analyze seismic reflection data, including the identification of seismic facies and structural features. However, interpreting the resulting clusters often relies on geoscientists’ expertise, necessitating a robustness assessment of these methods. To evaluate their reliability, synthetic data generated from an actual outcrop model are used to demonstrate how two unsupervised methods, self-organizing maps (SOMs) and generative topographic maps (GTMs), cluster deepwater channel-related seismic facies and then measure the associated error. Six seismic attributes, comprising root-mean-square amplitude, instantaneous envelope, peak magnitude, and spectral decomposition frequencies at 20, 40, and 55 Hz, served as input variables. Geobodies are assigned to each cluster formed, and the error in facies clustering is quantified by comparing the actual 3D model with the facies grouped by ML methods on a voxel-by-voxel basis. This allowed for error quantification and the computation of metrics such as F1 score and accuracy through correlation matrices. Key findings reveal that (1) GTM and SOM exhibit similar performance, with a clustering configuration of 81 for GTM slightly outperforming others. (2) Error rates are approximately 2% for the predominant facies (background shale) but significantly higher for individual channel-related facies, suggesting that channel clusters might represent multiple facies. (3) Resolution and imbalanced data distribution impact seismic facies predictability, resulting in nonuniqueness in cluster generation. (4) Using synthetic seismic data proved valuable for experimenting with different unsupervised MLs, highlighting the need for assessing uncertainty in these methods, given their implications for crucial economic decisions reliant on reservoir interpretation, modeling, and volumetric estimations.

Statistical estimation of the early to middle Ediacaran ocean redox architecture in the Yangtze block of South China

Early to middle Ediacaran organic-rich black shales host several famous fossil biotas and provide key evidence for understanding the coevolution of multicellular organisms and palaeoceanic environment. The water column redox states play critical roles in organic-rich shale deposition. Amongst multiple geochemical redox indexes, iron-speciation chemistry (FeHR/FeT, Fepy/FeHR) in shales constitutes a reliable proxy to reconstruct the first-order redox framework of the ocean basin. However, iron-speciation is a local marine redox proxy, and only compiled data collected from multiple different sedimentary facies record statistically significant changes of oxidation states within a depositional basin. Here we compiled the spatiotemporal distribution of iron-speciation data in early to middle Ediacaran black shales of the Yangtze block in South China. New high-resolution analysis on two outcrops and one drill-core reflects overall ferruginous condition and dominant euxinic condition locally interrupted by oxic states, respectively, generally indicating pervasive anoxic depositional environment for the Member II shales of the Ediacaran Doushantuo Formation in the Lower Yangtze block. Compilation of fourteen Ediacaran sections from shallow-, slope- and deep-water facies demonstrates frequent spatiotemporal oscillation of the redox conditions throughout the Yangtze block. Widespread ferruginous states, accompanied with euxinic zones mainly focusing in lower slope facies, are detected during the early Ediacaran period, in contrast with more widespread euxinic settings in the middle Ediacaran period. Statistical data including new high-resolution results from the Lower Yangtze block show generally low enrichment for redox sensitive elements (RSEs, e.g., Mo, V and U) in the Member II shales probably due to their contemporary low seawater concentrations. An increase of RSEs concentrations occurs in the Member IV shales, which probably reflects more oxidized Earth surface environment during middle Ediacaran period. The increasing Earth oxygenation documents the occurrence of widespread euxinic states during deposition of the Member IV shales. Compilation of spatial distribution of TOC contents in black shales demonstrates their apparent relevance to spatial distribution of the euxinic rather than ferruginous states, which is likely due to the favorable role of sulfurization process on burial and preservation of organic matters. The euxinic states may also facilitate the exceptional preservation of the soft-bodied Ediacaran fossils.

Boosting biodiversity monitoring using smartphone-driven, rapidly accumulating community-sourced data.

Comprehensive biodiversity data is crucial for ecosystem protection. The Biome mobile app, launched in Japan, efficiently gathers species observations from the public using species identification algorithms and gamification elements. The app has amassed >6 million observations since 2019. Nonetheless, community-sourced data may exhibit spatial and taxonomic biases. Species distribution models (SDMs) estimate species distribution while accommodating such bias. Here, we investigated the quality of Biome data and its impact on SDM performance. Species identification accuracy exceeds 95% for birds, reptiles, mammals, and amphibians, but seed plants, molluscs, and fishes scored below 90%. Our SDMs for 132 terrestrial plants and animals across Japan revealed that incorporating Biome data into traditional survey data improved accuracy. For endangered species, traditional survey data required >2000 records for accurate models (Boyce index ≥ 0.9), while blending the two data sources reduced this to around 300. The uniform coverage of urban-natural gradients by Biome data, compared to traditional data biased towards natural areas, may explain this improvement. Combining multiple data sources better estimates species distributions, aiding in protected area designation and ecosystem service assessment. Establishing a platform for accumulating community-sourced distribution data will contribute to conserving and monitoring natural ecosystems.

Source-free domain adaptation for semantic image segmentation using internal representations.

Semantic segmentation models trained on annotated data fail to generalize well when the input data distribution changes over extended time period, leading to requiring re-training to maintain performance. Classic unsupervised domain adaptation (UDA) attempts to address a similar problem when there is target domain with no annotated data points through transferring knowledge from a source domain with annotated data. We develop an online UDA algorithm for semantic segmentation of images that improves model generalization on unannotated domains in scenarios where source data access is restricted during adaptation. We perform model adaptation by minimizing the distributional distance between the source latent features and the target features in a shared embedding space. Our solution promotes a shared domain-agnostic latent feature space between the two domains, which allows for classifier generalization on the target dataset. To alleviate the need of access to source samples during adaptation, we approximate the source latent feature distribution via an appropriate surrogate distribution, in this case a Gaussian mixture model (GMM).

Evaluation Method for Underwater Ultrasonic Energy Radiation Performance Based on the Spatial Distribution Characteristics of Acoustic Power.

In recent years, underwater wireless ultrasonic energy transmission technology (UWUET) has attracted much attention because it utilizes the propagation characteristics of ultrasound in water. Effectively evaluating the performance of underwater ultrasonic wireless energy transmission is a key issue in engineering design. The current approach to performance evaluation is usually based on the system energy transfer efficiency as the main criterion, but this criterion mainly considers the overall energy conversion efficiency between the transmitting end and the receiving end, without an in-depth analysis of the characteristics of the distribution of the underwater acoustic field and the energy loss that occurs during the propagation of acoustic waves. In addition, existing methods focusing on acoustic field analysis tend to concentrate on a single parameter, ignoring the dynamic distribution of acoustic energy in complex aquatic environments, as well as the effects of changes in the underwater environment on acoustic propagation, such as spatial variability in temperature and salinity. These limitations reduce the usefulness and accuracy of models in complex marine environments, which in turn reduces the efficiency of acoustic energy management and optimization. To solve these problems, this study proposes a method to evaluate the performance of underwater ultrasonic energy radiation based on the spatial distribution characteristics of acoustic power. By establishing an acoustic power distribution model in a complex impedance-density aqueous medium and combining numerical simulation and experimental validation, this paper explores the spatial variation of acoustic power and its impact on the energy transfer efficiency in depth. Using high-resolution spatial distribution data and actual environmental parameters, the method significantly improves the accuracy of the assessment and the adaptability of the model in complex underwater environments. The results show that, compared with the traditional method, this method performs better in terms of the accuracy of the acoustic energy radiation calculation results, and is able to reflect the energy distribution and spatial heterogeneity of the acoustic source more comprehensively, which provides an important theoretical basis and practical guidance for the optimal design and performance enhancement of the underwater ultrasonic wireless energy transmission system.

Unsupervised Imputation of Non-ignorably Missing Data Using Importance-Weighted Autoencoders

Deep Learning (DL) methods have dramatically increased in popularity in recent years. While its initial success was demonstrated in the classification and manipulation of image data, there has been significant growth in the application of DL methods to problems in the biomedical sciences. However, the greater prevalence and complexity of missing data in biomedical datasets present significant challenges for DL methods. Here, we provide a formal treatment of missing data in the context of Variational Autoencoders (VAEs), a popular unsupervised DL architecture commonly utilized for dimension reduction, imputation, and learning latent representations of complex data. We propose a new VAE architecture, NIMIWAE, that is one of the first to flexibly account for both ignorable and non-ignorable patterns of missingness in input features at training time. Following training, samples can be drawn from the approximate posterior distribution of the missing data can be used for multiple imputation, facilitating downstream analyses on high dimensional incomplete datasets. We demonstrate through statistical simulation that our method outperforms existing approaches for unsupervised learning tasks and imputation accuracy. We conclude with a case study of an EHR dataset pertaining to 12,000 ICU patients containing a large number of diagnostic measurements and clinical outcomes, where many features are only partially observed.

A novel causal feature learning-based domain generalization framework for bearing fault diagnosis with a mixture of data from multiple working conditions and machines

Data distribution shift usually deteriorates the performance of traditional supervised learning-based fault diagnosis models. Such issue can be well addressed by domain generalization fault diagnosis methods even if the test data are unavailable in training stage. However, most of these methods are designed and validated only for a single device, and also require the domain labels being provided, which hampers their further application. Therefore, this paper focuses a more realistic scenario, where the source training set is a mixture of data from multiple operation conditions and devices with domain information unavailable. First, a causal analysis is made for the task of domain generalization fault diagnosis, and a structural causal model (SCM) that describes the data generation process is constructed based on the relationships among different variables. A new learning rule for domain generalization, named object-conditional domain invariant (Obj-DI) rule, is also elicited. Then, to achieve domain generalization fault diagnosis, a three-stage based method named LOODG is proposed. Potential domains are mined and samples are given domain labels in latent domain discovery stage. Samples belonging to same object but different domains are explored in object discovery stage. In Obj-DI based causal feature learning stage, generalizable causal features are learned to finally obtain a generalized model. Experiments are conducted on tasks designed from three datasets, and the results demonstrate the superiority of the proposed method.

Assessment of the levels of formation and leading components of the patient safety system of the anesthesiology and resuscitation service (message 2)

The objective was to assess the levels of formation and leading components of the patient safety system of the anesthesiology and resuscitation (A&R) service (department) of multidisciplinary hospitals.Materials and methods. Expert assessment of the patient safety system of the anesthesiology and resuscitation service (department) of 235 multidisciplinary hospitals, including the assessment of its level, leading components, and their structure in the overall sample. Expert assessment method. Statistical analysis of data distribution, Student’s t-test for independent samples of different sizes, correlation analysis.Results. It has been established that in the main part (38.3%) of multidisciplinary hospitals, the A&R service is at an average (sufficient) level of ensuring patient safety. 17% are at a high (organized) and highest (system-organized) level in terms of ensuring patient safety. A fairly large part (44.7%) is at a low (insufficient) and minimal (critical, extremely low) level of ensuring patient safety. At the same time, the level of ensuring patient safety in the A&R service of multidisciplinary hospitals is important and has a close but inverse relationship with the level of digitalization. Comparison of the severity of the five leading (main) components (criteria) for ensuring patient safety in the A&R service of multidisciplinary hospitals made it possible to assess the levels of their completeness and their structure and showed that in the general sample, the most pronounced criteria for ensuring patient safety were «Teamwork, staff expertise» and «Availability of medicines».Conclusion. The current state of the patient safety system of the A&R service of multidisciplinary hospitals is characterized by 5 levels and 5 leading (main) components, the expert assessment of which made it possible to distribute the A&R services of hospitals according to the indicated levels, identify the structure of their leading components an determine target areas for improving the patient safety system.

A physics-based dimension reduction and modeling method for monitoring data and its application to tunnel engineering

The promotion of intelligent control of engineering equipment requires extracting vital information from numerous original features in the monitoring data. Data-driven methods solely rely on data distribution to extract dimensional information, facing difficulties in comprehensively understanding and scrutinizing system behavior. To satisfy the control and optimization requirements of intelligent systems, this paper introduces a physics-based dimension reduction and modeling method. According to scientific principles, original features are transformed into causal variables with physical meaning. Then important causal variables are selected through regularization in machine learning. Finally, a concise prediction model with physical representation is established. An application to tunnel engineering demonstrates that the generated low-dimensional space reveals the essential laws between the control parameter and its main influencing factors (i.e., the geological conditions, operating status, and structural characteristics of the tunneling equipment), which can further provide a better prediction model balancing accuracy and dimensionality.

Online Cross-modal Hashing With Dynamic Prototype

Online cross-modal hashing has received increasing attention due to its efficiency and effectiveness in handling cross-modal streaming data retrieval. Despite the promising performance, these methods mainly focus on the supervised learning paradigm, demanding expensive and laborious work to obtain clean annotated data. Existing unsupervised online hashing methods mostly struggle to construct instructive semantic correlations among data chunks, resulting in the forgetting of accumulated data distribution. To this end, we propose a Dynamic Prototype-based Online Cross-modal Hashing method, called DPOCH. Based on the pre-learned reliable common representations, DPOCH generates prototypes incrementally as sketches of accumulated data and updates them dynamically for adapting streaming data. Thereafter, the prototype-based semantic embedding and similarity graphs are designed to promote stability and generalization of the hashing process, thereby obtaining globally adaptive hash codes and hash functions. Experimental results on benchmarked datasets demonstrate that the proposed DPOCH outperforms state-of-the-art unsupervised online cross-modal hashing methods.

Parallel processing model for low-dose computed tomography image denoising

Low-dose computed tomography (LDCT) has gained increasing attention owing to its crucial role in reducing radiation exposure in patients. However, LDCT-reconstructed images often suffer from significant noise and artifacts, negatively impacting the radiologists’ ability to accurately diagnose. To address this issue, many studies have focused on denoising LDCT images using deep learning (DL) methods. However, these DL-based denoising methods have been hindered by the highly variable feature distribution of LDCT data from different imaging sources, which adversely affects the performance of current denoising models. In this study, we propose a parallel processing model, the multi-encoder deep feature transformation network (MDFTN), which is designed to enhance the performance of LDCT imaging for multisource data. Unlike traditional network structures, which rely on continual learning to process multitask data, the approach can simultaneously handle LDCT images within a unified framework from various imaging sources. The proposed MDFTN consists of multiple encoders and decoders along with a deep feature transformation module (DFTM). During forward propagation in network training, each encoder extracts diverse features from its respective data source in parallel and the DFTM compresses these features into a shared feature space. Subsequently, each decoder performs an inverse operation for multisource loss estimation. Through collaborative training, the proposed MDFTN leverages the complementary advantages of multisource data distribution to enhance its adaptability and generalization. Numerous experiments were conducted on two public datasets and one local dataset, which demonstrated that the proposed network model can simultaneously process multisource data while effectively suppressing noise and preserving fine structures. The source code is available at https://github.com/123456789ey/MDFTN.

Conditional room layout generation based on graph neural networks

In this paper, we present a novel end-to-end variational generative model that utilizes graph-based latent representations for indoor scene synthesis at one time. In contrast to prior research, our method deviates from the practice of gradually introducing and arranging furniture in an empty room using autoregression. Instead, it focuses on acquiring a comprehensive implicit representation of the room’s original architectural structure and the placement of furniture. We initially transform the 3D room scene into a dense scene graph, where nodes correspond to the objects present in the room while edges reflect the spatial location links and functional correlations between objects. Then, a neural network is trained to acquire the graph-based latent representation of the room scene through iterative message passing, ultimately resulting in the acquisition of the data distribution on the latent space of the room layout. Given the architectural structure of an empty room as a prerequisite for scene synthesis, the generative model has the ability to sample from the prior distribution of the room’s latent representation. This allows the model to then decode and generate a variety of room layouts. We evaluate our method with the state-of-the-art 3D indoor scene dataset and generation methods. The experimental results demonstrate that our method achieves more rational and diverse outcomes in the context of generating scenes under specific conditions.

Harnessing Machine Learning and Data Fusion for Accurate Undocumented Well Identification in Satellite Images

This study utilizes satellite data to detect undocumented oil and gas wells, which pose significant environmental concerns, including greenhouse gas emissions. Three key findings emerge from the study. Firstly, the problem of imbalanced data is addressed by recommending oversampling techniques like Rotation–GaussianBlur–Solarization data augmentation (RGS), the Synthetic Minority Over-Sampling Technique (SMOTE), or ADASYN (an extension of SMOTE) over undersampling techniques. The performance of borderline SMOTE is less effective than that of the rest of the oversampling techniques, as its performance relies heavily on the quality and distribution of data near the decision boundary. Secondly, incorporating pre-trained models trained on large-scale datasets enhances the models’ generalization ability, with models trained on one county’s dataset demonstrating high overall accuracy, recall, and F1 scores that can be extended to other areas. This transferability of models allows for wider application. Lastly, including persistent homology (PH) as an additional input improves performance for in-distribution testing but may affect the model’s generalization for out-of-distribution testing. A careful consideration of PH’s impact on overall performance and generalizability is recommended. Overall, this study provides a robust approach to identifying undocumented oil and gas wells, contributing to the acceleration of a net-zero economy and supporting environmental sustainability efforts.

Continuous glucose monitoring and recreational scuba diving in type 1 diabetes: head-to-head comparison between Free Style Libre 3 and Dexcom G7 performance.

Background Scuba diving was previously excluded because of hypoglycemic risks for type 1 diabetic patients (T1DM). Specific eligibility criteria and a safety protocol have been defined, while continuous glucose monitoring (CGM) systems have enhanced diabetes management. This study aims to assess the feasibility and accuracy of CGM Dexcom G7 (DG7) and Free Style Libre 3 (FSL3) in a setting of repetitive scuba diving in T1DM, exploring the possibility of non-adjunctive use. Material and methods The study was conducted during an event of Diabete Sommerso® association in November 2023. Participants followed a safety protocol, with capillary glucose as reference standard (Beurer GL50Evo). Sensors accuracy was evaluated through Median and Mean Absolute Relative Difference (MeARD and MARD) and Surveillance Error Grid (SEG) Analysis. Sub-analyses were performed by glycemic range, timing, gender. Data distribution and correlation was estimated by Spearman test and Bland-Altman plots. The ability of sensors to identify hypoglycaemia was assessed by contingency tables. Participants completed a satisfation questionnaire. Results Data from 202 dives of 13 patients were collected. The overall MARD was 31% for DG7 and 14.2% for FSL3, MeARD 19.7% and 11.6% respectively. FSL3 exhibited better accuracy in normoglycemic and hyperglycemic ranges. SEG analysis showed 82.1% (DG7) and 97.4% (FSL3) data on no-risk zone. FSL3 better performed on hypoglycemia identification (diagnostic odds ratio of 254.10 vs 58.95). Neither of sensors reached the MARD for non-adjunctive use. Patient satisfaction favored FSL3 for size and minimal bulk, while DG7 received positive feedback for versatility. Conclusions The study reveals FSL3 superior accuracy compared to DG7 in a setting of repetitive scuba diving in T1DM, except for hypoglycemic range. Both sensors fails to achieve accuracy for non-adjunctive use. Capillary tests remain crucial for safe dive planning, and sensor data should be interpreted cautiously. We suggest exploring additional factors potentially influencing sensor performance.