Abstract

Abstract

Abstract
Label noise is omnipresent in the annotations process and has an impact on supervised learning algorithms. This work focuses on the impact of label noise on the performance of learning models by examining the effect of random and class-dependent label noise on a binary classification task: quality assessment for photoplethysmography (PPG). PPG signal is used to detect physiological changes and its quality can have a significant impact on the subsequent tasks, which makes PPG quality assessment a particularly good target for examining the impact of label noise in the field of biomedicine. Random and class-dependent label noise was introduced separately into the training set to emulate the errors associated with fatigue and bias in labeling data samples. We also tested different representations of the PPG, including features defined by domain experts, 1D raw signal and 2D image. Three different classifiers are tested on the noisy training data, including support vector machine (SVM), XGBoost, 1D Resnet and 2D Resnet, which handle three representations, respectively. The results showed that the two deep learning models were more robust than the two traditional machine learning models for both the random and class-dependent label noise. From the representation perspective, the 2D image shows better robustness compared to the 1D raw signal. The logits from three classifiers are also analyzed, the predicted probabilities intend to be more dispersed when more label noise is introduced. From this work, we investigated various factors related to label noise, including representations, label noise type, and data imbalance, which can be a good guidebook for designing more robust methods for label noise in future work.

Abstract

Abstract
Energy efficiency and indoor air quality are frequently-two conflicting objectives when establishing the air change rate (ACH) of a dwelling. In Europe, the northern countries have a clear focus on energy conservation, leading to an obvious awareness of the importance of airtightness, which translates into a high level of regulation and implementation. Meanwhile, the southern counterparts experience a more com-plex challenge by having predominantly passive ventilation strategies and milder climates, which often results in a more permissive approach. This work proposes an innovative labelling methodology to classify the performance of naturally ventilated dwellings. A representative sample of a southern European national built stock is used in a stochastic process to create a pool of 43,200 unique dwellings. The simulation period refers to a month of the typical heating season in the southern European mild conditions. The results test the labelling methodology. With feature selection, ACH limits, and a labelling strategy, dwellings classify according to their ability to provide adequate ACHs. The terrain was the best splitter of the dataset from the applied categorical variables. Regarding continuous variables, the airtightness was the one explaining most of the variability of the outputted ACHs, followed by the floor area. From the best performing dwellings labelled as compliant (Com), the average airtightness level was 5.3 h(-1), with 4.9 h(-1) and 5.8 h(-1) in rural and urban locations.

Abstract
In the context of bridge management, three main types of maintenance actions can be considered. Maintenance actions can be taken preventively before the predefined limit condition is reached, or as a corrective measure in case those limits have been reached. The third possibility corresponds to the so-called doing nothing scenario, in which no action is taken on the bridge. To be able to implement preventive maintenance, it is necessary to know the current condition of the bridge, as well as to be able to predict its performance. On the other hand, it is also important to be able to identify potentially threatening events that might occur in the analysis life period. This paper describes an integrated methodology to help bridge managers in defining an efficient maintenance program, considering the specific case of a railway bridge. The novelty of the methodology is focused on updating an existing methodology proposed by COST TU1406, by extending it to railway bridges and also by including the resilience analysis in case of a sudden event occurrence. The analysis considers a multi-hazard future scenario, in which a flood event occurs while corrosion phenomena were already in place. The results show the feasibility of the proposed methodology as a support for the establishment of an efficient maintenance schedule to prevent bridge severe degradation, as well as to establish recovery plans in case of a sudden event.