Search

Since the first studies on bowhead whale singing behaviour, song variations have been consistently reported. However, there has been little discussion regarding variability in bowhead whale singing display and its ecological significance. Unlike the better studied humpback whales, bowhead whales do not appear to share songs at population level, but several studies have reported song sharing within clusters of animals. Over the winter season 2013–2014, in an unstudied wintering ground off Northeast Greenland, 13 song groups sharing similar hierarchical structure and units were identified. Unit types were assessed through multidimensional maps, showing well separated clusters corresponding to manually labelled units, and revealing the presence of unit subtypes. Units presented contrasting levels of variability over their acoustic parameters, suggesting that bowhead whales keep consistency in some units while using a continuum in values of frequency, duration and modulation parameters for other unit types. Those findings emphasise the need to account for variability in song analysis to better understand the behavioural ecology of this endangered species. Additionally, shifting from song toward units or phrase-based analysis, as it has been suggested for humpback whales, offers the opportunity to identify and track similarities in songs over temporal and geographical scales relevant to population monitoring. [ABSTRACT FROM AUTHOR]

This paper studies the value of learning for cognitive transceivers in dynamic wireless networks. We quantify the utility improvement that can be obtained by a wideband user that learns the stationary usage pattern of the spectrum occupied by narrowband users and, based on this learned information, adapts its transmission. Specifically, we investigate the basic tradeoff between the learning duration and the achievable performance in stationary environments. We apply optimization and large- deviation theory to analytically derive an upper bound of the minimum required learning duration, given the user's tolerable performance loss and outage probability. Furthermore, since learning techniques require the information feedback of the spectrum usage pattern between the transceivers, we investigate how a cognitive user can further improve its performance by taking into account its feedback delay. The impact of inaccurate delay estimation on the achievable performance is also quantified. [ABSTRACT FROM AUTHOR]

Abstract: We introduce an efficient and flexible framework for temporal filtering in wavelet-based scalable video codecs called unconstrained motion compensated temporal filtering (UMCTF). UMCTF allows for the use of different filters and temporal decomposition structures through a set of controlling parameters that may be easily modified during the coding process, at different granularities and levels. The proposed framework enables the adaptation of the coding process to the video content, network and end-device characteristics, allows for enhanced scalability, content-adaptivity and reduced delay, while improving the coding efficiency as compared to state-of-the-art motion-compensated wavelet video coders. Additionally, a mechanism for the control of the distortion variation in video coding based on UMCTF employing only the predict step is proposed. The control mechanism is formulated by expressing the distortion in an arbitrary decoded frame, at any temporal level in the pyramid, as a function of the distortions in the reference frames at the same temporal level. All the different scenarios proposed in the paper are experimentally validated through a coding scheme that incorporates advanced features (such as rate-distortion optimized variable block-size multihypothesis prediction and overlapped block motion compensation). Experiments are carried out to determine the relative efficiency of different UMCTF instantiations, as well as to compare against the current state-of-the-art in video coding. [Copyright &y& Elsevier]

Abstract: In the frame of the European Sea-floor Observatory Network (ESONET) and in preparation of the Demonstration Mission Listening to the Deep-Ocean Environment (LIDO) algorithms for the automated real-time detection, classification and localisation of cetaceans have been developed. Such Passive Acoustic Monitoring techniques have the potential to play a key role in cetaceans’ conservation for they allow a non-invasive study of their behaviour, a better knowledge of their population dynamics, and a better understanding of their dynamic relationship with their environment. This paper investigates the development of efficient and accurate techniques to be used as the basis of a localisation module for an automated real-time Passive Acoustic Monitoring system. An opportunity to assess the capabilities of the developed localisation modules was given by recordings collected with a bottom-mounted (around 2080m depth) tetrahedral compact hydrophone array located offshore the port of Catania (Sicily) during the NEMO-ONDE campaign in the years 2005–2006 by INFN and CIBRA . A well-known class of methods for acoustic source localisation is based on time differences of arrival (TDOA). Its capabilities have shown to be useful even in adverse situations (i.e., few sensors, high noise levels and/or poor calibration). A second class of methods, the space–time methods, originated in underwater applications such as sonar but reached its most significant achievements over the last 20years in digital communications with recent progress in the treatment of broadband signals. These developments are here revisited under the scope of the localisation and tracking of cetacean vocalisations. Various broadband space–time methods were implemented and allowed to map the sound radiated during the detected clicks and to consequently localise both sperm whales and vessels. Hybrid methods were also developed which improved the robustness of space–time methods to noise and reverberation and reduced processing time. In most cases, the small variance obtained for these estimates lessened the necessity of additional statistical clustering. Even though not independently confirmed by sightings, the tracks derived in the proposed frame can be considered to be consistent with the known movements of sperm whales and vessels. [Copyright &y& Elsevier]

Despite dedicated research has been carried out to adequately map the distribution of the sperm whale in the Mediterranean Sea, unlike other regions of the world, the species population status is still presently uncertain. The analysis of two years of continuous acoustic data provided by the ANTARES neutrino telescope revealed the year-round presence of sperm whales in the Ligurian Sea, probably associated with the availability of cephalopods in the region. The presence of the Ligurian Sea sperm whales was demonstrated through the real-time analysis of audio data streamed from a cabled-to-shore deep-sea observatory that allowed the hourly tracking of their long-range echolocation behaviour on the Internet. Interestingly, the same acoustic analysis indicated that the occurrence of surface shipping noise would apparently not condition the foraging behaviour of the sperm whale in the area, since shipping noise was almost always present when sperm whales were acoustically detected. The continuous presence of the sperm whale in the region confirms the ecological value of the Ligurian sea and the importance of ANTARES to help monitoring its ecosystems. [ABSTRACT FROM AUTHOR]

Background: Building Machine Learning (ML) models in healthcare may suffer from time-consuming and potentially biased pre-selection of predictors by hand that can result in limited or trivial selection of suitable models. We aimed to assess the predictive performance of automating the process of building ML models (AutoML) in-hospital mortality prediction modelling of triage COVID-19 patients at ICU admission versus expert-based predictor pre-selection followed by logistic regression.Methods: We conducted an observational study of all COVID-19 patients admitted to Dutch ICUs between February and July 2020. We included 2,690 COVID-19 patients from 70 ICUs participating in the Dutch National Intensive Care Evaluation (NICE) registry. The main outcome measure was in-hospital mortality. We asessed model performance (at admission and after 24h, respectively) of AutoML compared to the more traditional approach of predictor pre-selection and logistic regression.Findings: Predictive performance of the autoML models with variables available at admission shows fair discrimination (average AUROC = 0·75-0·76 (sdev = 0·03), PPV = 0·70-0·76 (sdev = 0·1) at cut-off = 0·3 (the observed mortality rate), and good calibration. This performance is on par with a logistic regression model with selection of patient variables by three experts (average AUROC = 0·78 (sdev = 0·03) and PPV = 0·79 (sdev = 0·2)). Extending the models with variables that are available at 24h after admission resulted in models with higher predictive performance (average AUROC = 0·77-0·79 (sdev = 0·03) and PPV = 0·79-0·80 (sdev = 0·10-0·17)).Conclusions: AutoML delivers prediction models with fair discriminatory performance, and good calibration and accuracy, which is as good as regression models with expert-based predictor pre-selection. In the context of the restricted availability of data in an ICU quality registry, extending the models with variables that are available at 24h after admission showed small (but significantly) performance increase. [ABSTRACT FROM AUTHOR]