MIT: The ability for machines to reach the performance of the human brain in areas such as image interpretation or video ενδεχομένως να μπορεί να δώσει ένα νέο είδος μνήμης computer.
As reported in his publication MIT Technology Review, ερευνητές της ΙΒΜ χρησιμοποίησαν την αποκαλούμενη phase-change memory για να δημιουργήσουν μια συσκευή η οποία επεξεργάζεται δεδομένα με τρόπο ο οποίος παραπέμπει στη λειτουργία ενός βιολογικού εγκεφάλου. Μέσω της χρήσης ενός πρωτότυπου phase-change memory chip, οι ερευνητές καθόρισαν ρύθμισαν το σύστημα έτσι ώστε να λειτουργήσει ως ένα δίκτυο 913 νευρώνων με 165.000 connections, or synapses, between them.
The power of these synapses changes as the chip processes incoming data, changing how virtual neurons affect each other. Taking advantage of this feature, scientists have made the system capable of recognizing handwritten numbers.
The phase-change memory αναμένεται να εξέλθει στην αγορά μέσα στα επόμενα χρόνια. Μπορεί να καταγράψει information με μεγάλη ταχύτητα και να τις «πακετάρει» σε πολύ μεγαλύτερη πυκνότητα από ό,τι τα σημερινά είδη μνήμης. Ένα τσιπ τέτοιας μνήμης αποτελείται από ένα δίκτυο «κυττάρων» που μπορούν να λάβουν δύο καταστάσεις για να παρουσιάσουν ένα ψηφιακό bit πληροφορίας- ένα 1 ή ένα 0. Στο πειραματικό σύστημα της ΙΒΜ, η κάθε σύναψη αντιπροσωπεύεται από ένα ζεύγος κυττάρων που δουλεύουν μαζί.
"Stroke" computers have been the subject of research by computer scientists for some time. Such designs are radically different from today's chips, promising to make computers more efficient at tasks that are currently considered difficult by conventional systems - such as experience learning or video comprehension.
Earlier in the year, IBM announced the most advanced chip of its kind so far, created through techniques and components used to build smartphone processors. The new system is not as powerful, however, as Jeff Beer, an IBM researcher, pointed out that it is important that phase-change memory is used to create 165.000 synapses. According to him, this type of memory is considered to be suitable for "neuromorphic" type systems because it stores data at very high density - it is also easier to reprogram it. In practice, this facilitates the construction of a system that is "capable of" learning, by properly adjusting its behavior while receiving new data.
Previous attempts in this field were of limited scale, with 100 synapses or fewer. The new system, developed in collaboration with researchers at the Pohan Science and Technology University in Korea, is 1.000 of a larger size.
The paper was presented in December at the International Electron Devices Meeting in San Francisco.
Source: naftemporiki.gr