资源描述
Click to edit Master title style,Click to edit Master text styles,Second level,Third level,Fourth level,Fifth level,*,*,A MODEL OF HUMAN MEMORYJuly 2021,Memory patterns are probably not stored in the brains neurons.Why?,They take too long to train.,New“on-the-fly training disturbs old knowledge.,Old knowledge must be retrained in order to be maintained.Where would the old knowledge be stored?Catch 22.,Once neural networks are trained,the training patterns are discarded.Human memory does not discard training patterns.,Human memory records input patterns in great detail.,Cite:Bla Juleszs work with random dot stereograms and eidetic subject.,SENSORY INPUT PATTERN STORAGE,Sensory input patterns come from eyes,ears,tactile,olfactory,vestibular and other sensors.,Incoming patterns are stored in empty folders,wherever they are located.,Sequences of patterns,like videos are stored in the same folder.,Visual,auditory,and other sensory patterns that were received at the same time are stored in the same folder.,Only“interesting input patterns are stored in the memory.They remain for the rest of ones life.An eidetic stores everything,interesting or not.,Sensory input patterns go to short term memory and,if interesting,are transferred to main memory and recorded for life.,Short term memory(of the order of a few seconds)is used in the determination of what is interesting.The problem solver can also decide what is interesting.,Autoassociative neural networks are used in the pattern retrieval process.,Pattern retrieval occurs in response to prompt patterns.,Prompt patterns may come from sensory inputs.,The memory is organized in the form of independent segments to make possible a very large storage capacity.,SEGMENT N,NN,SENSING STRONG TRAINING,MUX,SEGMENT N+1,PROBLEM,SOLVER,HIT?,HIT LINE,SENSORY,INPUT,NN,SENSING STRONG TRAINING,MUX,HIT?,VC,AC,SC,SHORT,TERM,MEMORY,“INTERESTING INPUT,SWITCH,SENSORY INPUT LINE,MEMORY INPUT LINE,VC=VISUAL CORTEX,AC=AUDITORY CORTEX,SC=SENSORY CORTEX,SENSORY INPUT PATTERN STORAGE,INBORN PATTERN STORAGE,Inborn knowledge in the form of patterns is pre-loaded in the developing brains memory and remains intact throughout ones lifetime.,Examples of inborn knowledge:,(a)A bird building a nest involves complex construction in“safe places such as roof tops,tree tops,telephone poles,etc.,(b)Baby horse walking and finding lunch within half hour of birth.,(c)Human baby sucking,crying,peeing and pooping.,It is conjectured that the memory storage means for inborn knowledge is the,same as for sensory knowledge gained during a lifetime.,It is conjectured that the memory retrieval means for inborn data is the,same as for sensory input data.,Inborn patterns are stored in folders in“memory segment 0.,MEMORY INPUT LINE,SEGMENT 0,NN,SENSING,MUX,MEMORY INPUT LINE,INBORN PATTERNS,SENSORY INPUT LINE,HIT LINE,SHORT,TERM,MEMORY,“INTERESTING INPUT SWITCH,INBORN PATTERN STORAGE,VC,AC,HIT?,SC,VC=VISUAL CORTEX,AC=AUDITORY CORTEX,SC=SENSORY CORTEX,THOUGHT PATTERN STORAGE,Thought patterns are also stored in memory.Storage means and retrieval means for thought patterns are the same as for sensory input patterns.,Thought patterns come from the“problem solver.,The design of the problem solver is not yet part of this study,but could be thought of as a mechanism based on Arthur Samuels checker player.,Thought patterns are always interesting and stored in empty memory locations.,Storage of thought patterns takes precedence over storage of sensory input patterns.,SEGMENT N,NN,STRONG TRAINING,MUX,SEGMENT N+1,STRONG TRAINING,MUX,PROBLEM,SOLVER,MEMORY INPUT LINE,THOUGHT PATTERNS,THOUGHT PATTERN STORAGE,NN,PATTERN RETRIEVAL SYSTEM,Patterns stored in memory can be retrieved without knowledge of their storage location.,Autoassociative neural networks are part of the retrieval mechanism.,Autoassociative neural networks are trained by using their input patterns as both input and desired response patterns.They are trained to produce outputs that are reproductions of their inputs.,Once trained,autoassociative networks produce small input/output differences when presented with patterns that were trained in,but large differences when presented with patterns that were not trained in.Dj Vu?Hit or no hit?,Autoassociative networks are trained with all the patterns stored in the connected memory folders.,The autoassociative networks are prompted with sensory input patterns or thought patterns.Visual input patterns for example are rotated,translated,scaled,brightened,contrasted,etc.by the“visual cortex VC while attempting to make a hit.If there is a hit,the hit pattern is saved and compared with the contents of all the connected memory folders.The patterns of the folder containing the hit pattern are retrieved and sent to the problem solver which is the memory output“customer.,These patterns in turn may be used as prompts to retrieve other folders.This type of feedback could cause a“chain reaction resulting in the retrieval of many interrelated fold
展开阅读全文