Unsupervised Learning Approach Research Articles

Unsupervised learning approach to quantum wave-packet dynamics from coupled temporal-spatial correlations

Unsupervised learning approach to quantum wave-packet dynamics from coupled temporal-spatial correlations

Enhancing public opinion monitoring for social hot events with a time series neural network-based logic map

A method for constructing a commercial map of social hotspots based on time-series neural networks is proposed. It explores map construction through three stages: atomic event graph building, fusion, and public opinion analysis. The neural network captures event dynamics and public opinion evolution. An unsupervised learning approach trains the model to explore deep event connections. A time map reflects event trends and opinion changes. A sentence-level event extraction model provides rich information. An empirical mode decomposition method analyzes IMF data to understand intrinsic patterns. Time-series neural networks classify and reconstruct event graphs, constructing a knowledge graph. Experiments show this method reduces storage costs, simplifies prediction, and exhibits robustness. It offers new insights for commercial mapping and contributes to public opinion monitoring.

Unveiling digital mirrors: Decoding gendered body poses in instagram imagery

Social media platforms have had a significant impact on people's everyday lives worldwide and imagery on social media has become a vital means of practicing views of the self and communicating them to others. Albeit a significant proportion of people frequently upload self-representations on social media, this source of information on individuals, groups, and societies remains under-examined within the social sciences and neighboring disciplines. In our study, we focus on gender-stereotypical body-posing in self-portraits on social media. While sociology has examined gender stereotypes for decades, research lacks empirical evidence on representations in digital contexts as well as novel forms of stereotyping. We present a scalable and transferable methodology for analyzing body poses in images by combining neural network pose detection with an unsupervised learning approach and applying this methodology to data from the social media platform Instagram. Based on a clustering algorithm applied to gender-annotated imagery, we identify 150 body posing clusters. Our results reveal significant gender differences in 20 percent of clusters, many of which represent gender-stereotypical body poses addressed in sociological literature. Moreover, we can identify new stereotypical poses related to smartphone technology and social media trends. This study represents a novel approach to utilizing large-scale image data for social science research and contributes to a better understanding of the consolidation and reproduction of gender stereotypes in digital realms.

Addressing the non-perturbative regime of the quantum anharmonic oscillator by physics-informed neural networks

Abstract The use of deep learning in physical sciences has recently boosted the ability of researchers to tackle physical systems where little or no analytical insight is available. Recently, the Physics−Informed Neural Networks (PINNs) have been introduced as one of the most promising tools to solve systems of differential equations guided by some physically grounded constraints. In the quantum realm, such an approach paves the way to a novel approach to solve the Schrödinger equation for non-integrable systems. By following an unsupervised learning approach, we apply the PINNs to the anharmonic oscillator in which an interaction term proportional to the fourth power of the position coordinate is present. We compute the eigenenergies and the corresponding eigenfunctions while varying the weight of the quartic interaction. We bridge our solutions to the regime where both the perturbative and the strong coupling theory work, including the pure quartic oscillator. We investigate systems with real and imaginary frequency, laying the foundation for novel numerical methods to tackle problems emerging in quantum field theory.

Reexamining Transplant Outcomes in Acute Kidney Injury Kidneys Through Machine Learning.

Despite many people awaiting kidney transplant, kidney allografts from acute kidney injury (AKI) donors continue to be underutilized. We aimed to cluster kidney transplant recipients of AKI kidney allografts using an unsupervised machine learning (ML) approach. Using Organ Procurement and Transplantation Network-United Network for Organ Sharing (OPTN/UNOS) data, a consensus clustering cohort analysis was performed in 12356 deceased donor kidney transplant recipients between 2015 and 2019 in whom donors had a terminal serum creatinine ≥1.5mg/dL. Significant cluster characteristics were determined, and outcomes were compared. The median donor terminal creatinine was 2.2 (interquartile range [IQR] 1.7-3.3) mg/dL. Cluster analysis was performed on 12356 AKI kidney recipients, and three clinically distinct clusters were identified. Young, sensitized kidney re-transplant patients characterized Cluster 1. Cluster 2 was characterized by first-time kidney transplant patients with hypertensive and diabetic kidney diseases. Older diabetic recipients characterized Cluster 3. Clusters 1 and 2 donors were young and met standard kidney donor profile index (KDPI) criteria; Cluster 3 donors were older, more likely to have hypertension or diabetes, and meet high KDPI criteria. Cluster 1 had a higher risk of acute rejection, 3-year patient death, and graft failure. Cluster 3 had a higher risk of death-censored graft failure, patient death, and graft failure at 1 and 3 years. Cluster 2 had the best patient-, graft-, and death-censored graft survival at 1 and 3 years. Compared to non-AKI kidney recipients, the AKI clusters showed a higher incidence of delayed graft function (DGF, AKI: 43.2%, 41.7%, 45.3% vs. non-AKI: 25.5%); however, there were comparable long-term outcomes specific to death-censored graft survival (AKI: 93.6%, 93.4%, 90.4% vs. non-AKI: 92.3%), patient survival (AKI: 89.1%, 93.2%, 84.2% vs. non-AKI: 91.2%), and overall graft survival (AKI: 84.7%, 88.2%, 79.0% vs. non-AKI: 86.0%). In this unsupervised ML approach study, AKI recipient clusters demonstrated differing, but good clinical outcomes, suggesting opportunities for transplant centers to incrementally increase kidney utilization from AKI donors.

An Interference Cancelation and Signal Decoding Method for Non-orthogonal Multiple Access Based on Deep Learning

In wireless communication, spectral efficiency is one of the key performance parameters. The available limited resources must be effectively shared among different applications. Due to the tremendous demand for wireless applications, an effective technique could be used for sharing the resources (time/frequency). Non-orthogonal Multiple Access (NOMA), is a wireless network technique aimed at boosting spectral efficiency. In NOMA, Users in the downlink have received a super-imposed signal from a Base Station (BS) with varying power allocations. Depending on the power allocation, the receiver employs interference cancelation techniques to decode its own data. The effectiveness of interference cancelation hinges on achieving a perfect channel condition, which is challenging to achieve in practical scenarios. This study introduces an unsupervised Deep Learning (DL) approach known as autoencoder, aimed at interference cancelation to ensure signal decoding from super-imposed signal. In contrast to the traditional approach, simulation results of the proposed structure show that the DL-based receiver achieves superior performance in correctly decoding the signal.

Categorizing E-cigarette-related tweets using BERT topic modeling

BackgroundSocial media platforms are critical channels for promoting e-cigarettes, particularly among youth, making analysis of their vast and diverse content essential for public health interventions. Prevalence rates of e-cigarette use are high and evidence suggests that social media are popular forums that promote e-cigarette use through direct and indirect marketing techniques. The volume and diverse nature of e-cigarette-related information on social media is challenging and may obfuscate public health prevention messaging. Traditional hand-coding methods are labor-intensive and limit scalability. In contrast, unsupervised machine learning approaches, such as topic modeling, allow for efficient analysis of large datasets, uncovering patterns and trends that manual methods cannot achieve at scale. The present study focused on ascertaining the extent to which themes and topics in tweets related to e-cigarettes can be successfully rendered into useful homogenous units using machine learning. A better understanding of current depictions and discussions around e-cigarette products and use on social media can inform public health counter messaging and policy interventions. MethodsWe used topic modeling (BERTopic) to iteratively derive vape-related tweet clusters and calculate the importance of particular words to these groupings. We conducted a qualitative content analysis to study clustered tweets. We also sought to determine the geographic locations of e-cigarette conversations using automated geoparsing methods, which translate toponyms in textual data into geographic identifiers, to attempt to infer the location of tweets. ResultsWe were able to successfully identify >100,000 tweets in broad thematic categories in English and Spanish. Our correlation and inter-topic map analysis of the machine-derived topics, which examines the relationships between topics, indicated that most of the topics were unique (correlation value < 0.5) and did not overlap with each other. We identified six topics: Flavors and Disposable Vapes, Cannabis, Vape Shops and Refillable Vapes, Vape Culture, Anti-vaping and Quitting, and Spanish Tweets and Vaping Nicotine. Further analysis of these topics using qualitative methods identified themes within each topic. For example, Category 6 (Spanish Tweets and Vaping Nicotine) included four topics focused on the health risks of vaping, personal motivations for vaping, and the regulation of vaping products. Using geoparsing, which automatically detects location information, we found that the United States had the highest number of tweets related to vaping. Discussion/conclusionResults underscore the possibility of leveraging BERTopic modeling to reduce large quantities of data to comprehensively describe and categorize myriad e-cigarette related messages to which social media users are exposed. This data reduction approach can be applied to various social media platforms to describe and categorize e-cigarette posts and thereby triangulate and validate findings. Thematic content analysis of the topics identified through this technique requires supervision and human inputs. Our approach provides a comprehensive understanding of the evolving e-cigarette discourse, informing public health counter-messaging and policy interventions. Moreover, findings support the need for regulation, such as reducing appealing flavors and suggest that social media can be used effectively to support public health messaging (i.e., quitting messages).

Classifying the clouds of Venus using unsupervised machine learning

Because Venus is completely shrouded by clouds, they play an important role in the planet’s atmospheric dynamics. Studying the various morphological features observed on satellite imagery of the Venusian clouds is crucial to understanding not only the dynamic atmospheric processes, but also interactions between the planet’s surface structures and atmosphere. While attempts at manually categorizing and classifying these features have been made many times throughout Venus’ observational history, they have been limited in scope and prone to subjective bias. We therefore present and investigate an automated, objective, and scalable approach for their classification using unsupervised machine learning that can leverage full datasets of past, ongoing, and future missions.To achieve this, we introduce a novel framework to generate nadir observation patches of Venus’ clouds at fixed consistent scales from satellite imagery data of the Venus Express and Akatsuki missions. Such patches are then divided into classes using an unsupervised machine learning approach that consists of encoding the patch images into feature vectors via a convolutional neural network trained on the patch datasets and subsequently clustering the obtained embeddings using hierarchical agglomerative clustering.We find that our approach demonstrates considerable accuracy when tested against a curated benchmark dataset of Earth cloud categories, is able to identify meaningful classes for global-scale (3000km) cloud features on Venus and can detect small-scale (25km) wave patterns. However, at medium scales (∼500km) challenges are encountered, as available resolution and distinctive features start to diminish and blended features complicate the separation of well defined clusters.

Illuminating the Hierarchical Segmentation of Faults Through an Unsupervised Learning Approach Applied to Clouds of Earthquake Hypocenters

AbstractWe propose a workflow for the recognition of the hierarchical segmentation of faults through earthquake hypocenter clustering without prior information. Our approach combines density‐based clustering algorithms (DBSCAN and OPTICS), and principal component analysis (PCA). Given a spatial distribution of earthquake hypocenters, DBSCAN identifies first‐order clusters, representing regions with the highest density of connected seismic events. Within each first‐order cluster, OPTICS further identifies nested higher‐order clusters, providing information on their number and size. PCA analysis is applied to first‐ and higher‐order clusters to evaluate eigenvalues, allowing discrimination between seismicity associated with planar features and distributed seismicity that remains uncategorized. The identified planes are then geometrically characterized in terms of their location and orientation in the space, length, and height. This automated procedure operates within two spatial scales: the largest scale corresponds to the longest pattern of approximately equally dense earthquake clouds, while the smallest scale relates to earthquake location errors. By applying PCA analysis, a planar feature outputted from a first‐order cluster can be interpreted as a fault surface while planes outputted after OPTICS can be interpreted as fault segments comprised within the fault surface. The evenness between the orientation of illuminated fault surfaces and fault segments, and that of the nodal planes of earthquake focal mechanisms calculated along the same faults, corroborates this interpretation. Our workflow has been successfully applied to earthquake hypocenter distributions from various seismically active areas (Italy, Taiwan, and California) associated with faults exhibiting diverse kinematics.

Reproducible, data-driven characterization of sleep based on brain dynamics and transitions from whole-night fMRI.

Understanding the function of sleep requires studying the dynamics of brain activity across whole-night sleep and their transitions. However, current gold standard polysomnography (PSG) has limited spatial resolution to track brain activity. Additionally, previous fMRI studies were too short to capture full sleep stages and their cycling. To study whole-brain dynamics and transitions across whole-night sleep, we used an unsupervised learning approach, the Hidden Markov model (HMM), on two-night, 16 hr fMRI recordings of 12 non-sleep-deprived participants who reached all PSG-based sleep stages. This method identified 21 recurring brain states and their transition probabilities, beyond PSG-defined sleep stages. The HMM trained on one night accurately predicted the other, demonstrating unprecedented reproducibility. We also found functionally relevant subdivisions within rapid eye movement (REM) and within non-REM 2 stages. This study provides new insights into brain dynamics and transitions during sleep, aiding our understanding of sleep disorders that impact sleep transitions.

Advanced data-driven FBG sensor-based pavement monitoring system using multi-sensor data fusion and an unsupervised learning approach

Pavement health monitoring and assessment using sensor technology provides valuable insights into the condition of roads and helps in effective maintenance and management. This paper presents an advanced fiber Bragg grating (FBG) sensor-based pavement monitoring system that leverages multi-sensor data fusion techniques and an unsupervised learning approach for enhanced structural health assessment. The study commences with the fusion of data from various FBG sensors based on specific considerations and employing a novel function to improve the accuracy of pavement condition assessment. A dedicated data management system is introduced to address the challenges associated with continuous monitoring. This system automates data organization, pre-processing, and storage, facilitating historical trend analysis and post-processing. Then, the research introduces a novel unsupervised learning approach utilizing the density-based spatial clustering of applications with noise (DBSCAN) algorithm. This algorithm identifies abnormal patterns in the FBG sensor data, which may indicate structural anomalies or unexpected events, aiding the operator by automatically flagging abnormal data. By incorporating principle component analysis (PCA)-based feature fusion, the system reduces the dimensionality of the data, which enhances the efficiency of the clustering process. Various anomalies were detected in the data collected for four months from the test track. The cause and relevance of these anomalies were assessed to facilitate informed decision-making. This approach enables the operator to achieve early detection of potential structural issues and distinguish abnormal damage data from other anomalous data, thereby contributing to the overall success of the pavement monitoring system. In conclusion, the integration of advanced FBG sensor technology with multi-sensor data fusion and unsupervised learning techniques results in a sophisticated and robust pavement monitoring system. The presented approach enhances accuracy by detecting anomalies at any strain level, computational time using multi-sensor data fusion, and reliability of structural health assessment through anomaly evaluation. It also provides a framework for proactive decision-making, ultimately improving the longevity and structural integrity of critical infrastructure.

Dynamics of phytoplankton communities in the Baltic Sea: insights from a multi-dimensional analysis of pigment and spectral data—part I, pigment dataset

This study aims to investigate the seasonal and spatial distribution of surface phytoplankton communities in the Baltic Sea, using pigment analysis and hydrological parameters. Data were collected during six oceanographic campaigns between 2005 and 2008, including high-performance liquid chromatography (HPLC) pigment characterization and hydrological measurements. The first part of this comprehensive study was focused on the HPLC phytoplankton pigment dataset in relation to hydrological conditions. The research highlighted the importance of high-quality input data for accurate taxonomic analysis. Several unsupervised machine learning approaches, such as hierarchical cluster analysis (HCA), principal component analysis (PCA), and network-based community detection analysis (NCA), were used to analyze the data and identify phytoplankton communities based on biomarker pigments. Five main phytoplankton communities were identified: diatoms, dinoflagellates, cryptophytes, green algae, and cyanobacteria. The results evidenced distinct seasonal patterns, with diatom blooms dominating in spring, cyanobacterial blooms in mid-summer, and haptophyte and dinoflagellate peaks occurring in late summer and autumn. While PCA and NCA provided consistent insights into community structure, HCA offered less clarity in distinguishing between groups. The results of the statistical analysis were then compared with those of traditional approaches such as CHEMTAX and region-specific bio-optical algorithms, providing new perspectives on the taxonomic composition of phytoplankton groups. This study provides valuable insights into phytoplankton dynamics in the Baltic Sea and the effectiveness of different analytical approaches in understanding community structure, providing metrics that can enhance current and future advancements in remote sensing, including support for hyperspectral ocean color remote sensors.

Structural mechanism of glass transition uncovered by unsupervised machine learning

Uncovering the structural origins of the ubiquitous dynamic arrest phenomenon at the glass transition has long been a challenge due to the difficulty in identifying a rational structural representation from a disordered medium. To address this challenge, we propose a novel approach based on unsupervised learning to define a set of structural fingerprints. In this approach, complex local atomic environments, ranging from short to medium range, are captured by the discretized radial distribution function and projected onto a simple two-dimensional space using a neural network-based autoencoder. This two-dimensional space is characterized by two static structural indicators, P1 and P2, providing a comprehensive and user-friendly representation of the mysterious “glassy structure”. By employing Gaussian mixture modeling, the structural space is autonomously divided into three sections, each representing a unique cluster with similar environments. These indicators not only elucidate the glass transition but also allow for the quantitative prediction of activation barriers for local structural excitations. Furthermore, the unsupervised clustering technique can distinguish between the structural features of “hard zones” and “soft zones”, as well as recently proposed superfast “liquid-like” atoms in glass. This unsupervised machine learning approach demonstrates the utility of seemingly agnostic local structure in amorphous materials, offering insights into the long-sought structural origins of the glass transition.

A federated learning approach to automated and secure supplier selection in cyber manufacturing as-a-service

The emergence of cyber or platform-based manufacturing as-a-service is rapidly disrupting the way discrete parts are sourced and manufactured. However, the centralized business model of cyber manufacturing as-a-service platforms raises concerns about data ownership and access control of independent manufacturing suppliers. Contrary to centralized platforms, cyber manufacturing as-a-service aims to connect designers with geographically distributed manufacturers by serving as a broker who matches the query part design requirements with the manufacturing capabilities of candidate suppliers in its network. One of the key challenges in realizing the vision of cyber manufacturing as-a-service is the lack of a computationally efficient method for manufacturing capability search while maintaining data security of the proprietary datasets of the suppliers in the network. In this paper, we propose a federated learning approach that utilizes a deep unsupervised part retrieval model (FL-DUPR) to learn a federated embedding of suppliers’ manufacturing capabilities without directly accessing their proprietary datasets. We demonstrate through two case studies that this approach yields a supplier selection accuracy of 89 % when the manufacturing capabilities of the suppliers do not overlap, and a multi-label supplier selection accuracy of 87 % when there are significant overlaps in the suppliers’ manufacturing capabilities. We also show that our unsupervised learning approach outperforms the baseline supervised learning classification model trained under the same federated learning framework. The results demonstrate the promise of the proposed federated embedding approach for automated identification of the required manufacturing capabilities offered by various suppliers without directly accessing their proprietary data, thus paving the way for a more secure cyber manufacturing as-a-service business model.

Unsupervised machine learning and cepstral analysis with 4D-STEM for characterizing complex microstructures of metallic alloys

Four-dimensional scanning transmission electron microscopy, coupled with a wide array of data analytics, has unveiled new insights into complex materials. Here, we introduce a straightforward unsupervised machine learning approach that entails dimensionality reduction and clustering with minimal hyperparameter tuning to semi-automatically identify unique coexisting structures in metallic alloys. Applying cepstral transformation to the original diffraction dataset improves this process by effectively isolating phase information from potential signal ambiguity caused by sample tilt and thickness variations, commonly observed in electron diffraction patterns. In a case study of a NiTiHfAl shape memory alloy, conventional scanning transmission electron microscopy imaging struggles to accurately identify a low-contrast precipitate at lower magnifications, posing challenges for microscale analyses. We find that our method efficiently separates multiple coherent structures while using objective means of determining hyperparameters. Furthermore, we demonstrate how the clustering result facilitates more robust strain mapping to provide immediate and quantitative structural insights.

Patients carrying pathogenic SCN8A variants with loss‐ and gain‐of‐function effects can be classified into five subgroups exhibiting varying developmental and epileptic components of encephalopathy

AbstractObjectivePhenotypic heterogeneity presents challenges in providing clinical care to patients with pathogenic SCN8A variants, which underly a wide disease spectrum ranging from neurodevelopmental delays without seizures to a continuum of mild to severe developmental and epileptic encephalopathies (DEEs). An important unanswered question is whether there are clinically important subgroups within this wide spectrum. Using both supervised and unsupervised machine learning (ML) approaches, we previously found statistical support for two and three subgroups associated with loss‐ and gain‐ of‐ function vari‐ants, respectively. Here, we test the hypothesis that the unsupervised subgroups (U1–U3) are distinguished by differential contributions of developmental and epileptic components.MethodsWe predicted that patients in the U1 and U2 subgroups would differ in timing of developmental delay and seizure onset, with earlier and concurrent onset of both features for the U3 subgroup. Standard statistical procedures were used to test these predictions, as well as to investigate clinically relevant associations among all five subgroups.ResultsTwo‐population proportion and Kruskal–Wallis tests supported the hypothesis of a reversed order of developmental delay and seizure onset for patients in U1 and U2, and nearly synchronous developmental delay/seizure onset for the U3 (termed DEE) subgroup. Association testing identified subgroup variation in treatment response, frequency of initial seizure type, and comorbidities, as well as different median ages of developmental delay onset for all five subgroups.SignificanceUnsupervised ML approaches discern differential developmental and epileptic components among patients with SCN8A‐related epilepsy. Patients in U1 (termed developmental encephalopathy) typically gain seizure control yet rarely experience improvements in development, whereas those in U2 (termed epileptic encephalopathy) have fewer if any developmental impairments despite difficulty in achieving seizure control. This understanding improves prognosis and clinical management and provides a framework to discover mechanisms underlying variability in clinical outcome of patients with SCN8A‐related disorders.

Deep learning for surgical workflow analysis: a survey of progresses, limitations, and trends

Automatic surgical workflow analysis, which aims to recognize the ongoing surgical events in videos, is fundamental for developing context-aware computer-assisted systems. This paper reviews representative surgical workflow recognition algorithms based on deep learning, outlining their merits, limitations, and future research directions. The literature survey was performed on three large bibliographic databases, covering 67 lary sources, which were comparatively analyzed in terms of spatial feature modeling, spatio-temporal feature modeling, input pre-processing, regularization and post-processing algorithms, as well as learning strategies. Then, common public datasets and evaluation metrics for surgical workflow recognition are also described in detail. Finally, we discuss all literature from different perspectives, and point out the challenges, possible solutions and future trends. The need for more diverse and larger datasets, the potential of unsupervised and semi-supervised learning approaches, comprehensive and equitable metrics, establishing complete regulatory and data standards, and interoperability will be key challenges in translating models to clinical operating rooms. And we propose that surgical activity anticipation and employing large language model as training assistant are interesting research directions in surgical workflow analysis.

Identifying patterns of co-occurring chronic conditions preceding dementia: An unsupervised machine learning approach using health administrative data

ObjectivesIndividual risk factors for dementia are well known, but the influence of co-occurring chronic conditions has not been considered. We identified clusters of chronic conditions using an unsupervised machine learning approach and examined associations with incident dementia. ApproachUsing linked population-based administrative databases, we followed all community-dwelling adults aged 40-54 years in Ontario, Canada from April 2002 until March 2019 for incident dementia. We estimated the prevalence of 29 chronic conditions using validated algorithms and/or diagnosis codes. We reduced dataset dimensionality using multiple correspondence analysis and a fuzzy c-means clustering algorithm identified the optimal number of clusters (between 3-6 tested). Associations between clusters and incident dementia were examined using a cause-specific hazard model adjusted for sociodemographic characteristics and accounting for the competing risk of death. ResultsWe identified 82,359 eligible individuals (random 3% sample of total eligible individuals; mean age 46.5 years; 50.4% female). Regression analyses were based on 5 comorbidity clusters (fuzzy silhouette index:0.69). Compared to the low comorbidity cluster, persons in the cerebrovascular disease/metabolic (HRadj=3.06, 95%CI[2.42,3.86]) and neuro-related/mental health clusters (HRadj=2.51, 95%CI[2.05,3.07]) had the highest rates of incident dementia, followed by the cardiovascular risk factor cluster (HRadj=1.66,95%CI[1.32,2.09]). Persons in the cancer cluster did not have an increased incidence of dementia (HRadj=0.96,95%CI[0.77,1.20]). ConclusionsWe found significant associations between machine learning-derived clusters of chronic conditions and dementia. ImplicationsUnsupervised machine learning approaches to identify clusters of chronic conditions may be a useful tool for considering the impact of multimorbidity on dementia risk.

Typologies of South African small-scale farmers and their risk perceptions: an unsupervised machine learning approach

PurposeSmall-scale farmers are highly heterogeneous with regard to their types of farming, levels of technology adoption, degree of commercialization and many other factors. Such heterogeneous types, respectively groups of small-scale farming systems require different forms of government interventions. This paper applies a machine learning approach to analyze the typologies of small-scale farmers in South Africa based on a wide range of objective variables regarding their personal, farm and context characteristics, which support an effective, target-group-specific design and communication of policies.Design/methodology/approachA cluster analysis is performed based on a comprehensive quantitative and qualitative survey among 212 small-scale farmers, which was conducted in 2019 in the Limpopo Province of South Africa. An unsupervised machine learning approach, namely Partitioning Around Medoids (PAM), is applied to the survey data. Subsequently, the farmers' risk perceptions between the different clusters are analyzed and compared.FindingsAccording to the results of the cluster analysis, the small-scale farmers of the investigated sample can be grouped into four types: subsistence-oriented farmers, semi-subsistence livestock-oriented farmers, semi-subsistence crop-oriented farmers and market-oriented farmers. The subsequently analyzed risk perceptions and attitudes differ considerably between these types.Originality/valueThis is the first typologisation of small-scale farmers based on a comprehensive collection of quantitative and qualitative variables, which can all be considered in the analysis through the application of an unsupervised machine learning approach, namely PAM. Such typologisation is a pre-requisite for the design of more target-group-specific and suitable policy interventions.

An Unsupervised Approach for Treatment Effectiveness Monitoring Using Curvature Learning

Contrast-enhanced ultrasound could assess whether cancer chemotherapeutic agents work in days, rather than waiting 2–3 months, as is typical using the Response Evaluation Criteria in Solid Tumors (RECIST), therefore avoiding toxic side effects and expensive, ineffective therapy. A total of 40 mice were implanted with human colon cancer cells: treatment-sensitive mice in control (n = 10, receiving saline) and treated (n = 10, receiving bevacizumab) groups and treatment-resistant mice in control (n = 10) and treated (n = 10) groups. Each mouse was imaged using 3D dynamic contrast-enhanced ultrasound with Definity microbubbles. Curvature learning, an unsupervised learning approach, quantized pixels into three classes—blue, yellow, and red—representing normal, intermediate, and high cancer probability, both at baseline and after treatment. Next, a curvature learning score was calculated for each mouse using statistical measures representing variations in these three color classes across each frame from cine ultrasound images obtained during contrast administration on a given day (intra-day variability) and between pre- and post-treatment days (inter-day variability). A Wilcoxon rank-sum test compared score distributions between treated, treatment-sensitive mice and all others. There was a statistically significant difference in tumor score between the treated, treatment-sensitive group (n = 10) and all others (n = 30) (p = 0.0051). Curvature learning successfully identified treatment response, detecting changes in tumor perfusion before changes in tumor size. A similar technique could be developed for humans.