The egocentric action anticipation task consists in predicting future (unobserved) actions based on input from a wearable first-person view camera. In this work, we are focusing on applying egocentric action anticipation methods to the personal health domain, i.e., utilizing them for the analysis of dietary and hygienic activities routine and for the prevention of undesirable behavior (such as tasting food before washing hands or adding sugar). We collect a dataset of egocentric videos, capturing specific activities related to food preparation; fine-tune existing action anticipation models on this dataset and analyse effects caused by domain shift.

Fault detection models play a fundamental role in monitoring the health state of engineering systems subject to degradation processes. Data-driven fault detection models, albeit very effective when trained on large databases of failures, fail to perform well under a lack of failure examples. Because reliable engineering systems seldom fail, data shortage is often inevitable. To overcomes failure data scarcity, this work proposes a new model selection framework for the robust selection of fault detection and system health monitoring models. We combine heterogeneous sources of information (time-to-failure and sensors data), a model of the system structure, and mathematical bounds on the sensitivity and specificity of components fault classifiers. We use Support Vector Machines (SVMs) to detect components faults and Scenario theory to derive formal sensitivity and specificity bounds. The component predictions are combined within a system structure-function for system health states estimation. A novel model selection strategy optimizes the hyper-parameters of the SVM ensemble by minimizing system-level prediction errors and generalization error bounds of the individual fault classifiers. One of the main advantages of the proposed framework is a set of formal epistemic bounds on fault detection and false alarm probabilities quantifying the lack of data uncertainty affecting the fault detection rate. We test the method on three representative case studies: (1) On randomized fault detection experiments with synthetic data, (2) on ten SVM models for predictive maintenance of industrial health care imaging systems, and (3) on a real-world PHM challenge problem. The results prove the efficacy of the proposed approach and its usefulness for solving fault detection.

The scalability of learning algorithms has always been a central concern for data mining researchers, and nowadays, with the rapid increase in data storage capacities and availability, its importance has increased. To this end, sampling has been studied by several researchers in an effort to derive sufficiently accurate models using only small data fractions. In this article we focus on spectral k -means, that is, the k -means approximation as derived by the spectral relaxation, and propose a sequential sampling framework that iteratively enlarges the sample size until the k -means results (objective function and cluster structure) become indistinguishable from the asymptotic (infinite-data) output. In the proposed framework we adopt a commonly applied principle in data mining research that considers the use of minimal assumptions concerning the data generating distribution. This restriction imposes several challenges, mainly related to the efficiency of the sequential sampling procedure. These challenges are addressed using elements of matrix perturbation theory and statistics. Moreover, although the main focus is on spectral k -means, we also demonstrate that the proposed framework can be generalized to handle spectral clustering. The proposed sequential sampling framework is consecutively employed for addressing the distributed clustering problem, where the task is to construct a global model for data that resides in distributed network nodes. The main challenge in this context is related to the bandwidth constraints that are commonly imposed, thus requiring that the distributed clustering algorithm consumes a minimal amount of network load. This illustrates the applicability of the proposed approach, as it enables the determination of a minimal sample size that can be used for constructing an accurate clustering model that entails the distributional characteristics of the data. As opposed to the relevant distributed k -means approaches, our framework takes into account the fact that the choice of the number of clusters has a crucial effect on the required amount of communication. More precisely, the proposed algorithm is able to derive a statistical estimation of the required relative sizes for all possible values of k . This unique feature of our distributed clustering framework enables a network administrator to choose an economic solution that identifies the crude cluster structure of a dataset and not devote excessive network resources for identifying all the “correct” detailed clusters.

Semi-supervised learning algorithms commonly incorporate the available background knowledge such that an expression of the derived model's quality is improved.Depending on the specific context quality can take several forms and can be related to the generalization performance or to a simple clustering coherence measure.Recently, a novel perspective of semi-supervised learning has been put forward, that associates semi-supervised clustering with the efficiency of spectral methods.More precisely, it has been demonstrated that the appropriate use of partial supervision can bias the data Laplacian matrix such that the necessary eigenvector computations are provably accelerated.This result allows data mining practitioners to use background knowledge not only for improving the quality of clustering results, but also for accelerating the required computations.In this paper we initially provide a high level overview of the relevant efficiency maximizing semi-supervised methods such that their theoretical intuitions are comprehensively outlined.Consecutively, we demonstrate how these methods can be extended to handle multiple clusters and also discuss possible issues that may arise in the continuous semi-supervised solution.Finally, we illustrate the proposed extensions empirically in the context of text clustering.

Recent studies have demonstrated the prospects of data mining algorithms for addressing the task of seriation in paleontological data (i.e. the age-based ordering of the sites of excavation). A prominent approach is spectral ordering that computes a similarity measure between the sites and orders them such that similar sites become adjacent and dissimilar sites are placed far apart. In the paleontological domain, the similarity measure is based on the mammal genera whose remains are retrieved at each site of excavation. Although spectral ordering achieves good performance in the seriation task, it ignores the background knowledge that is naturally present in the domain, as paleontologists can derive the ages of the sites of excavation within some accuracy. On the other hand, the age information is uncertain, so the best approach would be to combine the background knowledge with the information on mammal co-occurrences. Motivated by this kind of partial supervision we propose a novel semi-supervised spectral ordering algorithm. Our algorithm modifies the Laplacian matrix used in spectral ordering, such that domain knowledge of the ordering is taken into account. Also, it performs feature selection (sparsification) by discarding features that contribute most to the unwanted variability of the data in bootstrap sampling. The theoretical properties of the proposed algorithm are thoroughly analyzed and it is demonstrated that the proposed framework enhances the stability of the spectral ordering output and induces computational gains.