EDS / ELT || (Extractor Discovery System / Evolutionary Learning Technology)

A system that makes means of recognition autonomously



A way to determine what is really important from a sea of information.

We live in the information era, deluged by text, music, photos, motion pictures, and more. Luckily, a number of recognition systems enable us to sort an enormous amount of information to obtain what we really need. A facial recognition system, for example, is designed to exclusively choose photos showing human faces — and to select them accurately. Recognition systems resulted from years of scientific trial and error, and experiments. The fact is, as people have become interested in more subjects than ever, we can no longer depend on scientists and experts alone for advanced recognition technologies to fill our needs. For example, plane spotters cannot expect much from scientists if they want a sorting system to classify their photographs of planes by carrier, because the photos almost always show aircraft from different airlines (JAL, ANA, etc.) at the same time. Problems of this kind would be solved with a system that can recognize objects autonomously. Sony CSL’s ELT is exactly that.


Genetic programming for searching complex data-classifying functions


The core of ELT is a software technology called Genetic Programming. This technology treats various formulae as genetic elements. By using a hybrid evaluation approach, it continues to search for complex formulae that can classify data. In the JAL and ANA example, when the system is given sample photos of JAL and ANA planes, it combines various formulae and tries to classify the planes. The system would then create an algorithm that can distinguish between JAL and ANA planes with a high degree of precision. EDS/ELT was jointly developed by Dr. Francois Pachet of Sony CSL-Paris and Yoshiyuki Kobayashi of Sony, both scientists who had long studied EDS. Its effectiveness in classifying music and pictures was already confirmed. EDS/ELT and the recognition abilities based on it have found their way into a range of Sony products, including mobile phones, tablets, the Walkman and, most recently, an autonomous product classification system for Sony’s manufacturing lines. Applications will likely broaden if cloud computing advances and as computers become more powerful.

Related Publications

- Zils, A. and Pachet, F. Automatic Extraction of Music Descriptors from Acoustic Signals using EDS. Proceedings of the 116th AES Convention, Berlin, Germany, May 2004- Yoshiyuki Kobayashi Automatic Generation of Musical Instrument Detector by Using Evolutionary Learning Method Proceedings of the 10th International Society for Music Information Retrieval Conference, Kobe, Japan, October 26-30,2009, POS 93-98