domenica 8 novembre 2009

Astounding results. Again.

Another 100% accuracy and with striking sharpness. The third day of testing (Friday 6th november) extended to 21 actions: we had outperforming results, absolutely astounding: 100% precision, no false at all.
We cannot find similar results in literature any far. Resarchers usually reach 90% at their best with only 3-5 actions [see the survey of Bao MIT 2003, still valid in terms of results survey]; and the majority do not even try with 10-12 actions, and who try surely do not reach the 100% . Using 21 action having 100% accuracy , is a really great and impressive performance, expecially because the method do not use any kind of biomechanical model or any a priori knowledge for signals analysis . It is a general method: we can easily extend it at will, and use it in many different contexts, without introducing any domain specific knowledge.

Note: We have already exausted all our dictionary for tests (a total of 252 actions!) in three try because we did not expect to reach a so sharp and net performance at the first shot.

Stefano, dott. Pinardi

dott. S.Pinardi
Nomadislab dept. of Computer Science
University of Milano Bicocca
Milan, Italy


[Italian translation]
Risultati eclatanti, ancora.

Ancora 100% accuracy e di una nitidezza eclatante.
Il terzo gg di test (venerdì) esteso a 21 azioni, dà risultati outperforming, risultati assolutamente sbalorditivi. Non esistono in letteratura risultati neanche lontanamente analoghi: la ricerca arriva nel migliore dei casi al 90% con sole 3-5 azioni da riconoscere e appena sopra queste cardinalità (10-12 azioni) già annaspa...

venerdì 6 novembre 2009

Flawless Victory: One Hundred Percent of Accuracy.

Milan - November 5th, 2009. Today me and Marco Brescianini - a thesis student who is continuing the thesis of Lorenzo Airoldi, which is still present in the laboratory as a coordinator - tested a new algorithm for Movement Science, for the recognition of movements with wearable sensors, based on a method grounded on my doctoral work and on a principle of my intuition. The idea has been realized technically in two stages, first with the work of Lorenzo, then by Marco, with occasional but important technical and architectural support of other people, like Enrico Sada. So, this is a doctoral research carried out with the technical programmatic support of Marco and Lorenzo and the cooperation of others. Friends first, then collaborators: last not least, great professionists.

The algorithm based on my simple idea - very rational, easily extensible, and brand new for the field - gave us a striking result: not entirely unexpected, but astounding for its performance and its unwavering clarity. We obtained a OHP (One Hundred Percent) recognition accuracy. The accuracy of 100 % is the number everyone dreams about for a new method of recognition (artificial intelligence).
An impressive achievement. First of all, because there are not examples of such accuracy in scientific literature of the area, even in better test conditions than ours.
Secondly, because the used method is not dependent on the quality nor the type of the used technology, nor the use of body mechanics model (forces, weights, joint, masses), nor on the ability to understand signal features; but only on a method of cognitive selection, of a context wheghting of the emergent features. A new idea of artificial intelligence in the movement science area.

Besides, the proposed method is easily extensible in terms of vocabulary of actions, and reusable in other technology contexts, because it does not use any a priori knowledge, for example about the body, but it evaluates the information value in its context.
So it is the pure effort of the algorithm, or better of the idea that is at the base, which allows to obtain this quality of results, and this striking performances.

We really did a great job: me, Marco, Lollo (Lorenzo), and incidentally Enrico (Sada) that asked me to explain him why my idea seemed so incomplete, the last thursday, and prompted me to prove him my misconception; instead I found why it could be correct, and that allowed me to return to the core of the original idea, that led us to these excellent results. Everyone counts in a research environment, as I always say, every idea could be relevant or determinant, like in the everyday life.

So let compliments and congratulate ourselfs! And let's keep the good work going...!!

We were very clever. Thanks to Marco, who "sweats" code every day in these days.
Thanks to the work of Lorenzo, his thesis served to comfort us that the direction of research was effective, and to Enrico who spends its free time, unpaid, to be present to resolve our architectural design, and who generates great code on subject as a great jazz musician plays ist instrument at "first sight".

Tomorrow, Friday November the sixth 2009, at 14:00 we will be at Nomadis Lab for other tests, to extended it and to reach higher result. So good luck to us!
We can rest for now, we all deserve it tonight ;-)

Bravo! ;-)
A greeting
Stefano(Dr. Pinardi).

(first and second picture: Dr. S. Pinardi, Marco Brescianini)
(third picture: on the left Dr. Alessandro Frontini, on the right Enrico Sada)
(fourth picture: on the left Davide Landi, on the right Dr. Airoldi Lorenzo