131,346 research outputs found
State of the Art in Computational Bioacoustics and Machine Learning: How far have we come?
No full textTerrestrial bioacoustics, like many other domains, has recently witnessed some transformative results from the application of deep learning and big data (Stowell 2017, Mac Aodha et al. 2018, Fairbrass et al. 2018, Mercado III and Sturdy 2017). Generalising over specific projects, which bioacoustic tasks can we consider "solved"? What can we expect in the near future, and what remains hard to do? What does a bioacoustician need to understand about deep learning? This contribution will address these questions, giving the audience a concise summary of recent developments and ways forward. It builds on recent projects and evaluation campaigns led by the author (Stowell et al. 2015, Stowell et al. 2018), as well as broader developments in signal processing, machine learning and bioacoustic applications of these. We will discuss which type of deep learning networks are appropriate for audio data, how to address zoological/ecological applications which often have few available data, and issues in integrating deep learning predictions with existing workflows in statistical ecology
1863-01-18 Major D. Porter Stowell recommends Lieutenant John D. Myrick for promotion
No full texthttps://digitalmaine.com/cw_me_1st_cav/1452/thumbnail.jp
1863-01-18 Major D. Porter Stowell recommends Lieutenant John D. Myrick for promotion
No full texthttps://digitalmaine.com/cw_me_1st_cav/1452/thumbnail.jp
1863-01-22 Major D. Porter Stowell writes regarding Henry McIntire
No full texthttps://digitalmaine.com/cw_me_1st_cav/1455/thumbnail.jp
1863-01-22 Major D. Porter Stowell writes regarding Henry McIntire
No full texthttps://digitalmaine.com/cw_me_1st_cav/1455/thumbnail.jp
Automatic large-scale classification of bird sounds is strongly improved by unsupervised feature learning
No full textAutomatic species classification of birds from their sound is a computational tool of increasing importance in ecology, conservation monitoring and vocal communication studies. To make classification useful in practice, it is crucial to improve its accuracy while ensuring that it can run at big data scales. Many approaches use acoustic measures based on spectrogram-type data, such as the Mel-frequency cepstral coefficient (MFCC) features which represent a manually-designed summary of spectral information. However, recent work in machine learning has demonstrated that features learnt automatically from data can often outperform manually-designed feature transforms. Feature learning can be performed at large scale and “unsupervised”, meaning it requires no manual data labelling, yet it can improve performance on “supervised” tasks such as classification. In this work we introduce a technique for feature learning from large volumes of bird sound recordings, inspired by techniques that have proven useful in other domains. We experimentally compare twelve different feature representations derived from the Mel spectrum (of which six use this technique), using four large and diverse databases of bird vocalisations, classified using a random forest classifier. We demonstrate that in our classification tasks, MFCCs can often lead to worse performance than the raw Mel spectral data from which they are derived. Conversely, we demonstrate that unsupervised feature learning provides a substantial boost over MFCCs and Mel spectra without adding computational complexity after the model has been trained. The boost is particularly notable for single-label classification tasks at large scale. The spectro-temporal activations learned through our procedure resemble spectro-temporal receptive fields calculated from avian primary auditory forebrain. However, for one of our datasets, which contains substantial audio data but few annotations, increased performance is not discernible. We study the interaction between dataset characteristics and choice of feature representation through further empirical analysis
MeSH term explosion and author rank improve expert recommendations
No full textInformation overload is an often-cited phenomenon that reduces the productivity, efficiency and efficacy of scientists. One challenge for scientists is to find appropriate collaborators in their research. The literature describes various solutions to the problem of expertise location, but most current approaches do not appear to be very suitable for expert recommendations in biomedical research. In this study, we present the development and initial evaluation of a vector space model-based algorithm to calculate researcher similarity using four inputs: 1) MeSH terms of publications; 2) MeSH terms and author rank; 3) exploded MeSH terms; and 4) exploded MeSH terms and author rank. We developed and evaluated the algorithm using a data set of 17,525 authors and their 22,542 papers. On average, our algorithms correctly predicted 2.5 of the top 5/10 coauthors of individual scientists. Exploded MeSH and author rank outperformed all other algorithms in accuracy, followed closely by MeSH and author rank. Our results show that the accuracy of MeSH term-based matching can be enhanced with other metadata such as author rank
1862-04-14 D.P. Stowell writes regarding the regiment
No full texthttps://digitalmaine.com/cw_me_1st_cav/1249/thumbnail.jp
Going Beyond Counting First Authors in Author Co-citation Analysis
No full textThe present study examines one of the fundamental aspects of author co-citation analysis (ACA) - the way co-citation
counts are defined. Co-citation counting provides the data on which all subsequent statistical analyses and mappings
are based, and we compare ACA results based on two different types of co-citation counting - the traditional type that
only counts the first one among a cited work's authors on the one hand and a non-traditional type that takes into
account the first 5 authors of a cited work on the other hand. Results indicate that the picture produced through this non-traditional author co-citation counting contains more coherent author groups and is therefore considerably clearer. However, this picture represents fewer specialties in the research field being studied than that produced through the traditional first-author co-citation counting when the same number of top-ranked authors is selected and analyzed. Reasons for these effects are discussed
"Closing the R&D Gap, Evaluating the Sources of R&D Spending"
No full textBoth spending and tax policies have been implemented in the United States with the goal of stimulating private sector research and development (R&D). Karier questions whether current R&D policy, especially the research and experimentation tax credit, can contribute to closing the gap between nondefense expenditures on R&D in the United States and such expenditures in other countries, such as Japan and Germany. He also explores possible changes to our current R&D policy to make it more effective.
- …
