NEUR 2P37 Lecture Notes - Lecture 5: Artificial Neural Network, Confabulation, Thiamine
Document Summary
Research supports that ltp and ltd are necessary for memory formation. Research does not support that these processes provide the full explanation. The roles of timing of action potentials has not been addressed by current models of memory. The order of activity (pre-synaptic activity the post-synaptic activity) matters greatly in memory formation. Not in every case does timing of spikes exist. Information must be relevant to be learned, but artificial neural networks cannot distinguish relevant from irrelevant. Relevance may be encoded by using neuromodulators to turn plasticity at the synapse on or off. Neural networks can extract the identify of the object without seeing the whole thing. Brains encode information differently, encoding the relation between different stimuli. Human memory shows invariance meaning the details of size, lighting, position, etc. do not matter. This supports the idea that we learn the relations between things, not just learn them on a pixel-by-pixel basis.