1 paul scherrer institutuoroboros model, bica, aaai 2008 knud thomsen the ouroboros model knud...

1 Paul Scherrer Institut

Uoroboros Model, BICA, AAAI 2008 Knud Thomsen

The Ouroboros Model

Knud Thomsen



• Taking fresh look at the BIG PICTURE• Avoid getting bogged down in details

• Observing different levels of analysis:

• functional (‘computational’)• algorithmic• implementational (D. Marr)

• Suggesting a synthesis Reuters

David Noelle,

Christian Lebiere



The Ouroboros Model, Main Features, I

.…anticipation, action / perception, evaluation, anticipation,...

The Ouroboros Model proposes a dynamic and self-organizing cognitive architecture with one fundamental recursive algorithm in the form of an extended action / perception cycle :

(K. Thomsen, Copenhagen 2005)

(Don Perlis,

Stephen Grossberg)



The Ouroboros Model, Main Features, II

• All concepts are stored in SCHEMATA, i.e. frames.. with features linked

together

• Activation of any one feature biases associated ones, i.e. provokes expectations

• A monitoring loop “CONSUMPTION ANALYSIS”

checks the fulfillment of these expectations

• Expectations can be violated, met, exceeded;

feedback = EMOTIONS (K. Thomsen, Tucson 2006)

• Assignment produces “LEFTOVERS” (K.T. Budapest

2007)

• Directs Flow of Activity (see below)

Antonio Chella

(Wei Chen)



One concept with features There are many of them

The whole concept gets biased,Empty slots are highlighted Input excites some features



Checking for fitting input Comparing result / expectation

Pointing out discrepanciesInclude highlighted featuresin new input

EMOTIONSEMOTIONS

continue



Originally foreseen for the paper:

Action-, Memory structure, Consumption Analysis, selected effects:

• Attention• Emotion• Problem Solving• Cognitive Growth• Teaching• Priming / Masking• Sleep• Self Awareness --> Consciousness• Pleasure in Play and Art• Real-world Applications



Please, let us discuss over the poster



Consumption Analysis highlights slots of the activated schemata and thus directs Attention to the most urgent issues, the general level of match together with herited Importance give a measure for the Relevance of an issue

Attention:

Problem Solving:

The Algorithm underlying the Ouroboros Model can be seen as a specific version of pattern matching and constraint satisfaction It can also be understood as an extension of production systems, any feature can serve as a starting point for activating a schemata

Cognitive Growth:

Resources, Concept Formation and –Refinement are directed to the topics where they are needed most

(Susan Epstein)



Teaching:

Supervision can be most effective when it taps into the same processes as working during self-steered learning; encouraging curiosity and providing goals and supportive conditions for AHA-experiences secures knowledge acquisition and fosters optimum growth



Assignment produces “LEFTOVERS”:

K. Thomsen et al, "The Laser Surveillance System", 7th Symposium on Safeguards and Nuclear Material Management, Liege, Belgium, 21 - 23 May 1989

www.cartoonstock.com/ catalog reference mwi0131

(~Pigeonhole principle, picture: wikipedia)



Hypothesis:

Sleep is a specific and multifaceted housekeeping function primarily aimed at getting rid of leftovers accumulating during all wake activities, maintaining appropriate signal/noise conditions in the brain •extra and especially novel activities necessitate an increased need for sleep as the

involved schemata are overflowing more, in particular when they are still developing • for the same reason babies and children need more sleep than adults • disturbing, threatening and unresolved issues would predominantly surface as dream

content as their processing could not be concluded yet• episodes and situations which stir emotions, because expectations - e.g. norms - are violated,

will preferably provide material for dreaming •erasing erroneous associations and traces of expectations or of perceptions unaccounted for

enhances correct and well established connections; the ensuing increase in signal / noise would make memory entries, checked for consistency by the consumption analysis, stand out and thus easier to retrieve and reactivate

• suppressed content preferably produces remnants• the variations in neuromodulator levels during sleep phases reported in the literature appear

to be well in line with their proposed role in “cleaning up”• the paradoxical activity of brain cells and areas during sleep seem to fit with the functions

which they are assumed to perform during awake behavior



Some Simple Observations concerning Sleep & Dream:

• Sleep follows waking like a (mirrored) shadow• We spend a good portion of our lives sleeping • Sleep is (almost) ubiquitous• Sleep is required, it did not evolve away• Dreaming evolved together with cognitive capabilities• The brain is very active during such “rest”• Dream content is not very different from wake content• The brain usually does not keep a direct memory of it • There is no consensus (yet) on the function of sleep & dream• There are many (not really contradictory) different proposals• … (K. Thomsen, Budapest 2007)

(Logo TSC 2005 Copenhagen)



Hypothesis refined:

• Junks (features @ frames) “approved” by the consumption analysis are marked as such producing “index-entries” in the hippocampus

• removing leftovers, i.e. improving signal/noise by specifically erasing disproved material, i.e. partly ‘empty’ frames and ‘loose’/unbound features, is a main function of sleeping and dreaming

• index-entries in the hippocampus are strengthened during SWS

• leftover frames and features in the association cortex are unprimed and weakened by REM sleep• we do not dream of highly over-learned stuff



Summary for Sleep:

Trivially, more activity produces more remnants; even more so, when categories do not (yet) fit

When suitable concepts, i.e. schemata, are not (yet) available, a lot is left unaccounted for; structures have to be destilled

Threatening stuff is paradigmatic for something not digested

Negative emotions arise when something does not (yet) fit

Keep useful chunks and make them stand out more clearly,Discard useless stuff and make place for new associations

Suppression leaves content apart by definition

Most residue accumulate where top and bottom meet

“Replay” and “Playing off” might be different functions/phases

The most basic requirement during cleaning: make sure that you produce less mess than you try to get rid of



The Ouroboros Model provides an integrative account of a great many features of Cognition as resulting over time from the efficient working of an iterative cyclic (self-referential & self-organizing) process implemented in a recurrent architecture:

ONE PROCESS CAN DO IT ALL

Conclusion:Reuters



Acknowledgements:Special thanks for inspiring discussion amongst others to:

Dorothée Legrand and Christopher Holvenstot



Reuters

Thank you



Your questions ? :

1 paul scherrer institutuoroboros model, bica, aaai 2008 knud thomsen the ouroboros model knud...

Documents