lecture 3: learning and memory prof.dr. jaap murre university of maastricht university of amsterdam...
Post on 16-Dec-2015
219 Views
Preview:
TRANSCRIPT
Lecture 3:Learning and Memory
Prof.dr. Jaap Murre
University of Maastricht
University of Amsterdam
jaap@murre.com
http://neuromod.uva.nl
Overview• We will study Hebbian learning and the formation of
categories• We will do some basic memory experiments• Examine various forms of memory• We will try to locate memory in the brain and relate
brain lesions to amnesia• We will also briefly explore executive functions in
the frontal lobes• We will look at memory improvement
With Hebbian learning, two learning methods are possible
• With unsupervised learning there is no teacher: the network tries to discern regularities in the input patterns
• With supervised learning an input is associated with an output– If the input and output are the same, we speak of
auto-associative learning– If they are different it is called hetero-associative
learning
Supervised learning with Hopfield (1982) network
• Bipolar activations – -1 or 1
• Symmetric weights (no self weights) – wij= wji
• Asynchronous update rule– Select one neuron randomly and update it
• Simple threshold rule for updating
Energy of a Hopfield network
Energy E = - ½ i,jwjiaiaj
E = - ½ i(wjiai + wijai)aj = - iwjiai aj
Net input to node j is iwjiai = netj
Thus, we can write E = - netj aj
The energy minimization question can also be turned around
• Given ai and aj, how should we set the weight wji
= wji so that the energy is minimized?
E = - ½ wjiaiaj, so that
– when aiaj = 1, wji must be positive
– when aiaj = -1, wji must be negative
• For example, wji= aiaj, where is a learning
constant
Hebb and Hopfield
• When used with Hopfield type activation rules, the Hebb learning rule places patterns at attractors
• If a network has n nodes, 0.15n random patterns can be reliably stored by such a system
• For complete retrieval it is typically necessary to present the network with over 90% of the original pattern
We will look at an example of competitive learning
• Competitive learning is a form of unsupervised learning
Example of competitive learning:Hebbian learning takes place
a t o
1 2
Category node 2 now represents ‘at’
Category 1 is established through Hebbian learning as well
a t o
1 2
Category node 1 now represents ‘to’
Before we continue...
• Everybody on the right of the classroom, please, close their eyes until the following words have been presented
• The others, pay attention to the following 10 words. You will be asked to remember them later
• Don’t write them down!
Now for the other half...
• Everybody on the left of the classroom, please, close their eyes until the following words have been presented
• The others, pay attention to the following 10 words. You will be asked to remember them later
• Don’t write them down!
Memory and attention are strongly intertwined
• Paying attention can be seen as holding in memory
• Attention is required for rehearsal
• The longer an item is attended (held in memory), the higher the chance it will be remembered later
Brown-Peterson task
• Try to remember three letters, e.g., XJC
• When given a number (e.g., 307), start counting backward in threes (307, 304, 301, 298, …)
• When the Write! text appears, write down the letters you remember
• This has to be done at least several times to obtain the effect
Typical results of the Brown-Peterson task
• The results typically show very low memory performance
• The reason is that rehearsal of the letters is prevented by the counting task
Before we continue,
• Write down all the words you remember from the presentation
• Make sure you do not verbalize them at this moment
• We will verify the result in a minute, but first we have the following two puzzles
Fragment completion
• Try to complete the following English word fragments
• You have 30 seconds
• Each dot (.) stands for a letter
• Don’t verbalize! (So, we can obtain a better sample)
The presented words were:
• Left half– table– car– tree– computer– monkey– paper– scissors– tennis– dessert– bread
• Right half– table– car– tree– computer– monkey– paper– assassin– tennis– dessert– bread
This model had some limitations
• Ba, ta, pa, ta, pa, ba is much more difficult to remember than ba, bu, bi, bu, bi, ba
• Hence, there are phonological effects in short-term memory
Executive functions
• What controls the memory retrieval process?
• How does the control process work?
• What determines which areas of brain are ‘allowed’ to active in the first place?
Anatomy of prefrontal cortex
• Strong lateral connectivity via stellates in Layer IV
• No direct connections to motor outputs
• Certain cells fire strongly and selectively during the delay period of a task in relation to certain aspects of the taks (e.g., position), especially in area 46 surrounding the principal sulcus
Goldman-Rakic studies of Piaget’s AB Paradigm
• Infants persist in reaching for a target even if they have observed it being hidden in another place and older infants will do this if the delay is large enough (2-5 s at 7.5-9 months)
• Still older infants will not do this
• Monkeys with dorsolateral prefrontal lesions show similar behavior (delays > 2 s)
Similarities with ‘prefrontal’ patients
• Prefrontal patients show perseveration on the Wisconsin card sorting test
• There is evidence that also in adult humans such behavior is mainly caused by lesions to the dorsolateral prefrontal cortex
There is also a working memory aspect to the task
• The subject must keep in mind where the hiding place was, which may involve a kind of ‘working memory’ lasting several seconds
• In other experiments Patricia Goldman-Rakic has implicated area 46 as performing a type of working memory function
• Alan Baddeley is not convinced that this type of working memory is similar to his own concept
Forgetting
• There is currently no theory that explains why we forget
• Forgetting seems to follow rather strict rules, but even these have not been fully explored
• It is postulated that very well rehearsed knowledge will never be forgotten (Harry Barrick’s ‘permastore’)
1000 PC hard disks
• The neocortex contains about 10 billion
• Every neuron connects to 10,000 others
• That amounts to 100,000 billion connections that each can store about 1 byte
• The neocortex thus has the equivalent capacity of at least a 1000 hard disks of 100 giga bytes
System 1: Trace system
• Function: Substrate for bulk storage of memories, ‘association machine’
• Corresponds roughly to neocortex
System 2: Link system
• Function: Initial ‘scaffold’ for episodes
• Corresponds roughly to hippocampus and certain temporal and perhaps frontal areas
System 3: Modulatory system • Function: Control of plasticity• Involves at least parts of the hippocampus,
amygdala, fornix, and certain nuclei in the basal forebrain and in the brain stem
Semantic dementia
• The term was adopted recently to describe a new form of dementia, notably by Julie Snowden et al. (1989, 1994) and by John Hodges et al. (1992, 1994)
• Semantic dementia is almost a mirror-image of amnesia
Neuropsychology of semantic dementia
• Progressive loss of semantic knowledge
• Word-finding problems
• Comprehension difficulties
• No problems with new learning
• Lesions mainly located in the infero-lateral temporal cortex but (early in the disease) with sparing of the hippocampus
Severe loss of traceconnections
Stage-2 learning proceedsas normal
Stage 3 learning stronglyimpaired
Non-rehearsed memorieswill be lost
No consolidation in semantic dementia
Nadel and Moscovitch (1997): Trace Replication Theory
• They reject the ‘Standard Theory’ of consolidation• Hippocampus always remains involved• Hippocampal representations increase in strength
with time
• For review and assessment, see Meeter, M., & J. M. J. Murre (2004). Consolidation of long-term memory: Evidence and alternatives. Psychological Bulletin, in press.
“We dream in order to forget”
• Or do we?
• Theory by Francis Crick and Graeme Mitchison (1983)
• Main problem: Overloading of memory
• Solution: Reverse learning leads to removal of ‘obsessions’
Dreaming and memory consolidation
• When should this reverse learning take place?
• During REM sleep– Normal input is deactivated– Semi-random activations from the brain stem– REM sleep may have lively hallucinations
Consolidation may also strengthen memory
• This may occur during deep sleep (as opposed to REM sleep)
• Both hypothetical processes may work together to achieve an increase in the clarity of representations in the cortex
Relevant data by Matt Wilson and Bruce McNaughton (1994)
• 120 neurons in rat hippocampus
• PRE: Slow-wave sleep before being in the experimental environment (cage)
• RUN: During experimental environment
• POST: Slow-wave sleep after having been in the experimental environment
Wilson en McNaughton Data
• PRE: Slow-wave sleep before being in the experimental environment (cage)
• RUN: During experimental environment
• POST: Slow-wave sleep after having been in the experimental environment
Experiment by Robert Stickgold
• Difficult visual discrimination problem
• Several hours of practice
• One group goes home• Other group stays in
the lab and skips a night of sleep
Improvement without further training due to sleep
0
5
10
15
20
25
0 2 4 6 8 10
Days after training
Imp
rove
me
nt (
ms)
Normal sleep
Skipped first night sleep
How the simulations work: One simulated ‘day’
• A new memory is learned
• A period of ‘simulated dreaming’ follows– Artificial neurons are activated randomly– This random activity causes ‘recall’ of a
memory– The recalled memory is strengthened in the
neocortex
Strongly and weakly encoded patterns
• Mixture of weak, middle and strong patterns
• Strong patterns had a higher learning parameter (cf. longer learning time)
0
0.5
1
0 5 10 15
Weak patterns
Middle patterns
Strong patterns
0
0.5
1
0 5 10 15
Weak patterns
Middle patterns
Strong patterns
A ‘Darwinian’ competition?
• Over time, the consolidation process squeezes out the weak patterns
Memory Improvement
• Strengthening of existing memory
• Not suitable for anterograde amnesia– Memory book/-electronic agenda– Errorless learning (Baddeley and Wilson)
The two pillars of effective memory
• Elaboration or Making words more memorable
• Rehearsal or Going back to what you are about to forget
Elaboration: Making words more memorable
• Partition (break it up!)
• Link– use anything that comes into your mind
• Imagine– visualize (bizarre)– hear, feel, smell, etc.– verbalize
Elaboration of names
• Pat Galveston• Pat Galve-ston• Pot Gulf Stone• An enormous pot with
a ‘sea’ (gulf) inside and with a massive rock in it
• Al Kane• (no need to
break up)• Eel Cane• An eel
slithering down a cane
Rehearsal: Go back to what you are about to forget
• The more rehearsal, the better the memory retention
• Rehearse at progressively longer intervals: expanding rehearsal
Expanding rehearsal
• Example schedule– immediately after the lecture (lesson, meeting,
experience, …)– the next day– three days later– a week later– a month later– after half a year
A very useful memory trick
• Uses the journey technique• Best used with lists of objects or
names• In your mind, walk along a
familiar route• Mentally, ‘place’ the objects at
locations along the route• Elaborate upon the
locations
Example of the journey technique
• In front of my building
• The revolving entrance door
• The lobby• Waiting for the
elevator• In the elevator
• Buy cat food
• Call Lucy
• Order chair
• E-mail Ted
• Bill John
Further hints on the journey technique
• Combine different journeys to remember long lists
• Always use the same locations
• This allows reference by number (e.g., 7-th on the list)
A trick to remember numbers
• One is a bun• Two is a shoe• Three is a knee• Four is a door• Five is a hive
• Six is sticks• Seven is heaven• Eight is a gate• Nine is wine• Ten is a hen
This is a peg technique
• Combine with the journey technique to remember long numbers, e.g., 597928641– bee hive in front of the building
– lots of bottles of wine in the door
– lobby has turned into heaven
– wine is presented while waiting for the elevator
– elevator is full of shoes, etc.
top related