15 June 2017

Research Picks Extra – June 2017

Signing on the dotted line, online
A new efficient and dynamic approach to signature recognition has been developed by an international research team. Their approach extracts the stroke-associated features of the signature for the global recognition phase as well as for signal pre-processing prior to local recognition. They explain in detail how the algorithm works and how it can distinguish between genuine and forged signatures with a good degree of accuracy against 100 signatures randomly selected from a database of 5000, with half fake and half genuine in their samples. The algorithm defeated the forgers 96% of the time.
El_Rahman, S.A. (2017) ‘An efficient approach for dynamic signature recognition‘, Int. J. Intelligent Engineering Informatics, Vol. 5, No. 2, pp.167-190

Using weeds to optimize
Optimization algorithms are an important component of countless complicated problems, especially those involve multiple inputs and outputs, random factors, and networks. Scientists in Iran have now turned to the way in which weeds colonize a garden to develop and algorithm that finds solutions to a complex problem by colonizing the network, for example. The green shoots of the weeds represent the solutions to the problem that best fit. The team has demonstrated proof of principle for their discrete invasive weed optimization (DIWO) algorithm to successfully solve a maximum-weighted tree matching problem (MWTMP) a problem of efficiency in hierarchical systems such as sales networks.
Zandieh, M., Shokrollahpour, E. and Bagher, M. (2017) ‘Maximum-weighted tree matching problem: a novel discrete invasive weed optimisation algorithm‘, Int. J. Intelligent Systems Technologies and Applications, Vol. 16, No. 2, pp.95-105

Fixing autocorrect
There’s a joke that one might say is an old joke, at least in terms of modern technology: “The computer scientist who invented autocorrect has died, the funfair will be held next monkey”. It’s old, I didn’t say it was funny! However, it does cut to the quick of a problem many of us face in modern society, the lack of context recognition in autocorrect that leads to spelling errors and malapropisms galore in our personal and business messages, even our presidential tweets. Now, researchers from India are working on developing a new mechanism that can identify context and so boost the accuracy of autocorrect. No more monkey funfairs when the funeral is to be held on Monday from now on.
Nejja, M. and Yousfi, A. (2017) ‘Context’s impact on the automatic spelling correction‘, Int. J. Artificial Intelligence and Soft Computing, Vol. 6, No. 1, pp.56-74

Do you like good music?
Cataloguing music has been an issue for music publishers, artists and shops and suppliers since the days of printed sheet music back in the 19th century, all the way through vinyl era of the 20th century to vinyl’s digital successor the CD and then into the digital download era of mp3, Ogg, and FLAC files. It is possible to categorize a piece of music based on artist and perceived genre, it is also possible to search lyrics for songs. However, a universal classification system that ignores genre (much music crosses genre or cannot be pigeonholed) and does not rely on a listener’s opinion would be most valuable to the music industry, musicians and artists and music consumers alike. A team from Algeria is developing a system that might solve this perennial problem by providing a faithful representation of the semantic content of songs, facilitating music information retrieval in heterogeneous collections by indexing all types of songs from the same concepts, and enabling intelligent search by exploiting the semantic relations between indexing concepts.
Lachtar, N., Bahi, H. and Bouras, Z.E. (2017) ‘Conceptual search of songs using domain ontology and semantic links‘, Int. J. Intelligent Systems Technologies and Applications, Vol. 16, No. 2, pp.153-168

No comments: