scholarly journals An adaptive approach for anomaly detector selection and fine-tuning in time series

2019 ◽  
Author(s):  
Hui Ye ◽  
Xiaopeng Ma ◽  
Qingfeng Pan ◽  
Huaqiang Fang ◽  
Hang Xiang ◽  
...  
Keyword(s):  
Time Series ◽  
Fine Tuning ◽  
Adaptive Approach ◽  
Anomaly Detector

scholarly journals A Time History of Pre- and Post-Bomb Radiocarbon in the Barents Sea Derived from Arcto-Norwegian Cod Otoliths

Radiocarbon ◽  
2001 ◽  
Vol 43 (2B) ◽  
pp. 843-855 ◽  
Author(s):  
John M Kalish ◽  
Reidar Nydal ◽  
Kjell H Nedreaas ◽  
George S Burr ◽  
Gro L Eine
Keyword(s):  
Time Series ◽  
Ocean Circulation ◽  
General Circulation ◽  
Barents Sea ◽  
Dissolved Inorganic Carbon ◽  
Time History ◽  
General Circulation Models ◽  
Fine Tuning ◽  
Birth Date ◽  
The Barents Sea

Radiocarbon measured in seawater dissolved inorganic carbon (DIC) can be used to investigate ocean circulation, atmosphere/ocean carbon flux, and provide powerful constraints for the fine-tuning of general circulation models (GCMs). Time series of 14C in seawater are derived most frequently from annual bands of hermatypic corals. However, this proxy is unavailable in temperate and polar oceans. Fish otoliths, calcium carbonate auditory, and gravity receptors in the membranous labyrinths of teleost fishes, can act as proxies for 14C in most oceans and at most depths. Arcto-Norwegian cod otoliths are suited to this application due to the well-defined distribution of this species in the Barents Sea, the ability to determine ages of individual Arcto-Norwegian cod with a high level of accuracy, and the availability of archived otoliths collected for fisheries research over the past 60 years. Using measurements of 14C derived from Arcto-Norwegian cod otoliths, we present the first pre- and post-bomb time series (1919–1992) of 14C from polar seas and consider the significance of these data in relation to ocean circulation and atmosphere/ocean flux of 14C. The data provide evidence for a minor Suess effect of only 0.2‰ per year between 1919 and 1950. Bomb 14C was evident in the Barents Sea as early as 1957 and the highest 14C value was measured in an otolith core from a cod with a birth date of 1967. The otolith 14C data display key features common to records of 14C obtained from a Georges Bank mollusc and corals from the tropical and subtropical North Atlantic.

scholarly journals Self-Supervised Pre-Training of Transformers for Satellite Image Time Series Classification

2020 ◽  
Author(s):  
Yuan Yuan ◽  
Lei Lin
Keyword(s):  
Time Series ◽  
Deep Learning ◽  
Large Scale ◽  
Temporal Structure ◽  
Satellite Image ◽  
Fine Tuning ◽  
Small Scale ◽  
Model Parameters ◽  
Learning Approaches ◽  
Wide Range

Satellite image time series (SITS) classification is a major research topic in remote sensing and is relevant for a wide range of applications. Deep learning approaches have been commonly employed for SITS classification and have provided state-of-the-art performance. However, deep learning methods suffer from overfitting when labeled data is scarce. To address this problem, we propose a novel self-supervised pre-training scheme to initialize a Transformer-based network by utilizing large-scale unlabeled data. In detail, the model is asked to predict randomly contaminated observations given an entire time series of a pixel. The main idea of our proposal is to leverage the inherent temporal structure of satellite time series to learn general-purpose spectral-temporal representations related to land cover semantics. Once pre-training is completed, the pre-trained network can be further adapted to various SITS classification tasks by fine-tuning all the model parameters on small-scale task-related labeled data. In this way, the general knowledge and representations about SITS can be transferred to a label-scarce task, thereby improving the generalization performance of the model as well as reducing the risk of overfitting. Comprehensive experiments have been carried out on three benchmark datasets over large study areas. Experimental results demonstrate the effectiveness of the proposed method, leading to a classification accuracy increment up to 1.91% to 6.69%. <div><b>This work has been submitted to the IEEE for possible publication. Copyright may be transferred without notice, after which this version may no longer be accessible.</b></div>

Using hierarchical dynamic time warping to synchronize age-uncertain (proxy) time series

2020 ◽  
Author(s):  
Yuval Burstyn ◽  
Asaf Gazit
Keyword(s):  
Time Series ◽  
Laser Ablation ◽  
Trace Element ◽  
Dynamic Time Warping ◽  
Regional Scale ◽  
Trace Element Analysis ◽  
Fine Tuning ◽  
Time Warping ◽  
Non Local ◽  
Dynamic Time

&lt;p&gt;Climate- and environmental-proxy time series obtained from different archives, such as speleothems, allowed for major leaps in the understanding of past climate and environmental dynamics. However, age uncertainties that arise from the applied dating techniques and from the proxy sampling methodologies, respectively, are often neglected. These age uncertainties are important when leads and lags between different proxy time series are examined or if the relationship to climate-forcing is investigated. This is most pronounced when examining data that detail events of sub-centennial down sub-annual resolution, where noise is not smoothed by a low resolution sampling (e.g. conventional dental drill), or in records karst systems where the noise is inherently high (e.g. water-limited environments).&lt;/p&gt;&lt;p&gt;We explore the use of dynamic time warping with a hierarchical aggregation layer (or HDTW) on multiple trajectories to generate an indexing table for the input samples. We hypothesize that this aggregation process results a temporally aligned references table (of the original trajectories) and allows for an analytical space to investigate and distinguish between local and non-local phenomena. We aim to compare sample derived features, such as peaks in trace element, organic fluorescence analyses and potentially &amp;#948;&lt;sup&gt;18&lt;/sup&gt;O (not tested here), on the derived analytical space, for the purpose of enabling a robust and simplified approach to multi-sample age modelling.&lt;/p&gt;&lt;p&gt;We show HDTW compatibility to existing peak-counting methodologies applied on laser-ablation trace element analysis and confocal fluoresce laser microscopy. As a case study, we use HDTW on three published micron-scale elemental measurements of samples from Mediterranean climates with strong dry summer &amp;#8211; wet winter seasonality - two from south-western Australia (Nagra et al., 2017) and one from the Soreq Cave in the Eastern Mediterranean (Orland et al., 2014). The HDTW continuous space for these samples yields results that are within the published age constraints, without the need to stack multiple traverses and manually account for double or missing peaks.&lt;/p&gt;&lt;p&gt;HDTW is an important new tool for locating and identifying local and non-local phenomena in micron scale measurements (e.g. parallel laser ablation trace element traverses) by automatically aligning several coeval time axes of similar proxies. In the future HDTW could be applied for regional scale investigation (e.g. a coeval speleothems from a single cave or the same region, multiple cores from a single lake) allowing the unbiased fine-tuning between different environmental archives registering similar forcing mechanisms.&lt;/p&gt;&lt;p&gt;Nagra, G., Treble, P.C., Andersen, M.S., Bajo, P., Hellstrom, J.C., Baker, A., 2017. Dating stalagmites in Mediterranean climates using annual trace element cycles. Sci. Rep. 7, 621.&lt;/p&gt;&lt;p&gt;Orland, I.J., Burstyn, Y., Bar-Matthews, M., Kozdon, R., Ayalon, A., Matthews, A., Valley, J.W., 2014. Seasonal climate signals (1990&amp;#8211;2008) in a modern Soreq Cave stalagmite as revealed by high-resolution geochemical analysis. Chem. Geol. 363, 322&amp;#8211;333.&lt;/p&gt;

scholarly journals A simple test to infer mode of selection in metagenomics time series of evolving asexual populations

2020 ◽  
Author(s):  
Rohan Maddamsetti ◽  
Nkrumah A. Grant
Keyword(s):  
Time Series ◽  
Positive Selection ◽  
Gene Regulatory Network ◽  
Regulatory Network ◽  
Fine Tuning ◽  
Simple Test ◽  
Regulatory Regions ◽  
E Coli ◽  
Gene Regulatory ◽  
Asexual Populations

ABSTRACTWe introduce a simple test to infer mode of selection (STIMS) in metagenomic time series of evolving asexual populations. STIMS compares the tempo of molecular evolution for a gene set of interest against a null distribution that is bootstrapped on random gene sets. We test STIMS on metagenomic data spanning 62,750 generations of Lenski’s long-term evolution experiment with E. coli (LTEE). Our method successfully recovers signals of purifying selection and positive selection on gold standard sets of genes. We then use STIMS to study the evolution of genetic modules in the LTEE. We find strong evidence of ongoing positive selection on key regulators of the E. coli gene regulatory network. Key regulatory genes show evidence of positive selection over the entire time series, even in some hypermutator populations. By contrast, we found no signal of selection on the genetic modules that show the strongest transcriptional responses to changes in growth conditions. In addition, the cis-regulatory regions of key regulators are evolving faster than the cis-regulatory regions of their downstream regulatory targets. These results indicate that one mechanistic cause for ongoing fitness gains in the LTEE is ongoing fine-tuning of the gene regulatory network.

scholarly journals Self-Supervised Pre-Training of Transformers for Satellite Image Time Series Classification

2020 ◽  
Author(s):  
Yuan Yuan ◽  
Lei Lin
Keyword(s):  
Time Series ◽  
Deep Learning ◽  
Large Scale ◽  
Temporal Structure ◽  
Satellite Image ◽  
Fine Tuning ◽  
Small Scale ◽  
Model Parameters ◽  
Learning Approaches ◽  
Wide Range

<div>Satellite image time series (SITS) classification is a major research topic in remote sensing and is relevant for a wide range of applications. Deep learning approaches have been commonly employed for SITS classification and have provided state-of-the-art performance. However, deep learning methods suffer from overfitting when labeled data is scarce. To address this problem, we propose a novel self-supervised pre-training scheme to initialize a Transformer-based network by utilizing large-scale unlabeled data. In detail, the model is asked to predict randomly contaminated observations given an entire time series of a pixel. The main idea of our proposal is to leverage the inherent temporal structure of satellite time series to learn general-purpose spectral-temporal representations related to land cover semantics. Once pre-training is completed, the pre-trained network can be further adapted to various SITS classification tasks by fine-tuning all the model parameters on small-scale task-related labeled data. In this way, the general knowledge and representations about SITS can be transferred to a label-scarce task, thereby improving the generalization performance of the model as well as reducing the risk of overfitting. Comprehensive experiments have been carried out on three benchmark datasets over large study areas. Experimental results demonstrate the effectiveness of the proposed method, leading to a classification accuracy increment up to 2.38% to 5.27%. The code and the pre-trained model will be available at https://github.com/linlei1214/SITS-BERT upon publication.</div><div><b>This work has been submitted to the IEEE for possible publication. Copyright may be transferred without notice, after which this version may no longer be accessible.</b></div>

AN ADAPTIVE APPROACH TO TIME-SERIES FORECASTING

1979 ◽  
Vol 10 (2) ◽  
pp. 232-244 ◽  
Author(s):  
Stuart Bretschneider ◽  
Robert Carbone ◽  
Richard L. Longini
Keyword(s):  
Time Series ◽  
Time Series Forecasting ◽  
Adaptive Approach

scholarly journals KPI-TSAD: A Time-Series Anomaly Detector for KPI Monitoring in Cloud Applications

Symmetry ◽  
2019 ◽  
Vol 11 (11) ◽  
pp. 1350 ◽  
Author(s):  
Juan Qiu ◽  
Qingfeng Du ◽  
Chongshu Qian
Keyword(s):  
Time Series ◽  
Anomaly Detection ◽  
Short Term Memory ◽  
Performance Indicator ◽  
Classification Problem ◽  
Stationary Time Series ◽  
Imbalanced Classification ◽  
Usage Patterns ◽  
Benchmark Datasets ◽  
Anomaly Detector

Accurately detecting anomalies and timely interventions are critical for cloud application maintenance. Traditional methods for performance anomaly detection based on thresholds and rules work well for simple key performance indicator (KPI) monitoring. Unfortunately, it is difficult to find the appropriate threshold levels when there are significant differences between KPI values at different times during the day or when there are significant fluctuations stemming from different usage patterns. Therefore, anomaly detection presents a challenge for all types of temporal data, particularly when non-stationary time series have special adaptability requirements or when the nature of potential anomalies is vaguely defined or unknown. To address this limitation, we propose a novel anomaly detector (called KPI-TSAD) for time-series KPIs based on supervised deep-learning models with convolution and long short-term memory (LSTM) neural networks, and a variational auto-encoder (VAE) oversampling model was used to address the imbalanced classification problem. Compared with other related research on Yahoo’s anomaly detection benchmark datasets, KPI-TSAD exhibited better performance, with both its accuracy and F-score exceeding 0.90 on the A1benchmark and A2Benchmark datasets. Finally, KPI-TSAD continued to perform well on several KPI monitoring datasets from real production environments, with the average F-score exceeding 0.72.

scholarly journals Self-Supervised Pre-Training of Transformers for Satellite Image Time Series Classification

2020 ◽  
Author(s):  
Yuan Yuan ◽  
Lei Lin
Keyword(s):  
Time Series ◽  
Deep Learning ◽  
Large Scale ◽  
Temporal Structure ◽  
Satellite Image ◽  
Fine Tuning ◽  
Small Scale ◽  
Model Parameters ◽  
Learning Approaches ◽  
Wide Range

<div>Satellite image time series (SITS) classification is a major research topic in remote sensing and is relevant for a wide range of applications. Deep learning approaches have been commonly employed for SITS classification and have provided state-of-the-art performance. However, deep learning methods suffer from overfitting when labeled data is scarce. To address this problem, we propose a novel self-supervised pre-training scheme to initialize a Transformer-based network by utilizing large-scale unlabeled data. In detail, the model is asked to predict randomly contaminated observations given an entire time series of a pixel. The main idea of our proposal is to leverage the inherent temporal structure of satellite time series to learn general-purpose spectral-temporal representations related to land cover semantics. Once pre-training is completed, the pre-trained network can be further adapted to various SITS classification tasks by fine-tuning all the model parameters on small-scale task-related labeled data. In this way, the general knowledge and representations about SITS can be transferred to a label-scarce task, thereby improving the generalization performance of the model as well as reducing the risk of overfitting. Comprehensive experiments have been carried out on three benchmark datasets over large study areas. Experimental results demonstrate the effectiveness of the proposed method, leading to a classification accuracy increment up to 2.38% to 5.27%. The code and the pre-trained model will be available at https://github.com/linlei1214/SITS-BERT upon publication.</div><div><b>This work has been submitted to the IEEE for possible publication. Copyright may be transferred without notice, after which this version may no longer be accessible.</b></div>

Sign in / Sign up

Export Citation Format

Share Document