On-chip Time Series Classification: Towards Achieving State-of-the-art Results in Real Time (LINE...

Employer
University of East Anglia
Location
Other
Posted
September 12 2017
Position Type
Full Time
Organization Type
Academia

Time series classification (TSC) problems arise across a rich and diverse range of domains.  There is an active community of researchers developing algorithms for solving TSC problems, leading to many published algorithms and abundant sources of data (85 such problems are hosted by the UEA TSC group at www.timeseriesclassification.com).  Recently, one of the largest ever studies in machine learning was carried out at UEA [i]; more than 20 leading TSC algorithms were evaluated using 100 resamples of 85 public datasets, culminating in over 30 million individual experiments.  The results established that a UEA-developed algorithm, COTE [ii], was the state-of-the-art algorithm for TSC.


COTE has recently been updated to HIVE-COTE [iii], which is the new state of the art and significantly more accurate than any other TSC algorithm. However, while highly accurate, HIVE-COTE has certain deficiencies. It is computationally-intensive which means that it is not suitable for deploying in various emerging real-world applications that demand fast or real-time decisions. The evaluation in [i] investigated the accuracy of the best TSC algorithms but this only half of the story.


This PhD project will define and improve the state-of-the-art in TSC under restricted operating conditions, such as limited time constraints and computing resources.  The applicant will join a vibrant TSC group at UEA and the research will have direct application to ongoing projects in the group, including classification of electrical devices from usage patterns, fraudulent alcohol from spectrograms, mosquitos from wingbeat frequencies, and marine mammals from autonomous vehicle sensors.  The project will initially benchmark leading algorithms [e.g. ii-v] under constrained conditions, and using the insight gleaned from these investigations, the candidate will develop new state-of-the-art algorithms under restricted operating conditions for application to real-world TSC problems.


Informal project enquiries are welcomed by the primary supervisor (j.lines@uea.ac.uk).


Interviews will be held w/c 22 January 2018.


Funding notes


This PhD project is in a Faculty of Science competition for funded studentships. These studentships are funded for 3 years and comprise home/EU fees, an annual stipend of £14,553 and £1000 per annum to support research training. Overseas applicants may apply but they are required to fund the difference between home/EU and overseas tuition fees (in 2017/18 the difference is £13,805 for the Schools of CHE, PHA & MTH (Engineering), and £10,605 for CMP & MTH but fees are subject to an annual increase).


 



This job comes from a partnership with Science Magazine and Euraxess