Slashdot Mirror
Memory Molecule Identified
Reader Ostracus informs us of research led by Michael Ehlers of Duke University that has identified a molecule, myosin Vb (five-b), that seems to be a critical component in the formation of memory. "A major puzzle for neurobiologists is how the brain can modify one... synapse at a time in a brain cell and not affect the thousands of other connections nearby. Plasticity, the ability of the brain to precisely rearrange the connections between its nerve cells, is the framework for learning and forming memories ... The discovery of a molecule that moves new receptors to the synapse so that the neuron... can respond more strongly helps to explain several observations about [brain] plasticity ... [The researchers] found that the myosin Vb molecule in hippocampal neurons responded to a flow of calcium ions from the synaptic space by popping up and into action. One end of the myosin is attached to meshlike actin filaments so it can 'walk' to the end of the nerve cells where receptors are. On its other end, it tows an endosome, a packet that contains new receptors. 'These endosomes are like little memories waiting to happen,' Ehlers said."
I see here a possible method of improving AI. If we can indeed model synthetic neurons to perform in a similar way, we might have the key to designing more efficient captcha breaking systems.
Views expressed do not necessarily reflect those of the author.
Google on "artificial neural network" and read a few of the 600,000 hits that you will find. ANN theory is as old as digital computers. Commercial ANN applications have been growing in number and sophistication for over 10 years, e.g,, Dragon NaturallySpeaking and other speech recognition software, Caere OmniPage and other OCR packages.
What TFA is about is reporting the discovery of a key part of the mechanism that changes the weighting factors in a neuron in a biological neural net. Of itself, I doubt that this will trigger any insights on how to improve ANNs: the frankenmeisters already know how to do that with the neurones they work with. But this does open the door for further research by biologists into wetware neural net mechanisms, and that could lead to some interesting things.