New Stochastic Method for Sequence Recognition
We offer a new stochastic method for sequence recognition, which is suitable for detection of for example speech, text or handwriting recognition. In tests it outperformed Hidden Markov Models (HMM’s) and other approaches significantly, which were state of the art for pattern recognition of sequences up till now.<!–break–></p>
<p>The new sequence recognition method is nearly under and over fitting free. The training procedure doesn’t require an initial set of parameters or incremental optimization, so that the training can run fully automated. The method is able to do “word spotting” in very long sequences. Because of its unique nature it is very insensitive to non-linear distortion. The needed computing cost is low, thus the new method for sequence recognition can work in real time.<br> <br> <strong>IP Rights</strong><br> <strong>IP Rights</strong><br> EP and US patent application with priority on July 2007<br> <br> <strong>Patent Owner</strong><br> Technische Universität Berlin</p>
Further Information: PDF
ipal GmbH
Phone: +49 (0)30/2125-4820
Contact
Dr. Dirk Dantz
Media Contact
All latest news from the category: Technology Offerings
Newest articles
Superradiant atoms could push the boundaries of how precisely time can be measured
Superradiant atoms can help us measure time more precisely than ever. In a new study, researchers from the University of Copenhagen present a new method for measuring the time interval,…
Ion thermoelectric conversion devices for near room temperature
The electrode sheet of the thermoelectric device consists of ionic hydrogel, which is sandwiched between the electrodes to form, and the Prussian blue on the electrode undergoes a redox reaction…
Zap Energy achieves 37-million-degree temperatures in a compact device
New publication reports record electron temperatures for a small-scale, sheared-flow-stabilized Z-pinch fusion device. In the nine decades since humans first produced fusion reactions, only a few fusion technologies have demonstrated…