EvoSpike · Evolving Probabilistic Spiking Neural Networks for Spatio-Temporal Pattern Recognition
7РП — „Хора“ (Действия „Мария Кюри“)
- Период
- 2011-06-01 → 2012-05-31
- Финансиране от ЕС
- 121 353 €
- Участници
- 1
- Схема
- MC-IIF
Линиите свързват координатора с партньорите.
Накратко на български
Еволюциониращите импулсни невронни мрежи анализират данни, които се променят във времето и пространството, като например разпознаването на движещи се обекти или ЕЕГ сигнали. Тези модели помагат за по-доброто разбиране на работата на мозъка и подобряват разпознаването на аудио-визуални образи.
Кратко обяснение, генерирано от езиков модел по текста на CORDIS. Оригиналът е по-долу.
Резултати накратко
Evolving Probabilistic Spiking Neural Networks for Spatio-Temporal Pattern Recognition
The EVOSPIKE project developed novel methods and systems of spiking neural networks (SNN) for spatio- and spectro-temporal data (SSTD). Specifically, it developed models of evolving probabilistic spiking neural networks (epSNN) and evolving probabilistic computational neuro-genetic models (epCNGM). The collabouration between the applicant and host institution led to important results, including the development of software and hardware implementations and pilot applications for audio-visual pattern recognition, EEG data analysis, and neurogenetic cognitive systems. These results further led to 5 journal publications and 13 conference papers. Specifically, there were 12 major achievements: 1. a framework for spatio- and spectro-temporal data (SSTD) modelling and pattern recognition with evolving SNN was developed and published; 2. a new method for SSTD was developed and published - dynamic evolving SNN (deSNN); 3. a new method for SSTD was developed and published - spike pattern association neuron (SPAN); 4. a reservoir type of eSNN for SSTD was developed and published; 5. a pilot application on moving object recognition was developed and published; 6. a pilot application on handwritten digit recognition was developed and published; 7. a pilot application on EEG SSTD was developed and published; 8. a novel computational neuro-genetic model called NeuCube was principally developed and published in LNCS and also submitted as a paper to Nature. NeuCube is superior than other techniques for brain neurogenetic SSTD mining and understanding; 9. a new type of spatio-temporal associative memory and spatio-temporal finite automata were proposed as part of the NeuCube framework; 10. algorithms and programs in Python were developed to run the deSNN and SPAN on the custom analog / digital VLSI hardware (multi-neuron SNN chips) developed in the group of Prof. Indiveri. Preliminary results were included in three joint publications; 11. software was developed and made available, along with all publications, at the project web site: http://ncs.ethz.ch/projects/evospike/; 12. recommendations for further research and long term collabouration in neuromorphic computation were suggested to be based on the NeuCube novel architecture. Two proposals for collabourative work are in a lst stage of preparation - one to IRSES and one to FET-Open to be submitted in January 2013.
Текст от CORDIS, на английски · Данни: CORDIS, © Европейски съюз
Цел на проекта
Spiking neural networks (SNN), considered the third generation of neural networks, are a promising paradigm for the creation of new intelligent ICT and for the study of the brain. This new generation computational models and systems are potentially capable of modelling complex information processes due to their ability to represent and integrate different information dimensions, such as time, space, frequency, phase, and to deal with large volumes of data in an adaptive, self-organising, self-learning way. The progress in this direction has been slow in the past, but now there are more opportunities for a progress to be made and this is the aim of the proposed project. The host organisation, the Institute of Neuro-Informatics (INI), Zurich, has been developing VLSI technologies for implementing SNNs for many years. As it has mainly focused on the hardware development aspects, it is still lacking a theoretical framework for configuring and applying VLSI SNNs to wider computational problems. The contribution of this project and of the incoming researcher Prof. Kasabov will be crucial for making a breakthrough in this domain. The project proposes to devise a theoretical framework and a methodology for the design of novel SNN, namely evolving probabilistic spiking neural networks (epSNN) and evolving probabilistic computational neuro-genetic models (epCNGM) along with their implementation on existing software and hardware platforms at the host organisation INI. The resulting technologies will offer a new way to efficiently solve a wide range of complex spatio-temporal pattern recognition problems, including: audio-visual pattern recognition; EEG brain data analysis; associative memories; neurogenetic cognitive systems. Further applications of the epCNGM are expected to be developed for modelling brain data related to neurodegenerative diseases, such as Alzheimer’s disease. Knowledge will be transferred from the visiting researcher Prof. Kasabov to INI and Europe.
Оригинален текст от CORDIS (на английски).
Участници
- University of Zurich · ZURICHКоординаторШвейцария
Връзки
Данни: CORDIS, © Европейски съюз