Brain-inspired ICT

Brain-inspired ICT

memory A perspective from outside: SISSA-CNSTrieste

Alessandro Treves

Brain-inspired ICT, page II

memoryi.e.,

attractors ?

analogdiscrete

Noam Chomsky Lord Adrian

FACETS (Fast Analog Computing with Emergent Transient States)

Daisy (Neocortical Daisy Architectures and Graphical Models for Context-Dependent Processing)

CILIA - Customized Intelligent Life-inspired Arrays

Integrated Projects funded under the FP6 Bio-i 3 Proactive Initiative

+ Projects funded under FP6 in the area of Neuro-IT…

Joint Activities for future ResearchNeuro-IT.net - Neuro-IT Net: Thematic Network Neuroinformatics for living artefacts projectsAMOUSE - Artificial Mouse ARTESIMIT - Artefact Structural Learning through Imitation INSIGHT 2+ - 3D Shape and material properties and recognition MIRROR - Mirror Neurons based Robot Recognition

POETIC - Reconfigurable Poetic Tissue

Neurosciences (Neuro-IT): Funded projects

Joint Activities for future ResearchNeuro-IT.net - Neuro-IT Net: Thematic Network Neuroinformatics for living artefacts projectsAMOUSE - Artificial Mouse ARTESIMIT - Artefact Structural Learning through Imitation INSIGHT 2+ - 3D Shape and material properties and recognition MIRROR - Mirror Neurons based Robot Recognition POETIC - Reconfigurable Poetic Tissue SIGNAL - Systemic Intelligence for Growingup Artifacts that Live AMOTH - A fleet of articifical chemosensing moths for distributed environmental monitoring BIBA - Bayesian Inspired Brain and Artefacts: Using probabilistic logic to understand brain function and implement life-like behavioural co-ordination ECOVISION - Artificial vision systems based on early coginitive cortical processing HYDRA - Living Building blocks for self-designing artefacts PALOMA - Progressive and adaptive learning of object manipulation: a biologically inspired multi-network architecture Neuroinformatics projects managed in Directorate General ResearchMICROCIRCUITS - Cortical, Cerebellar and Spinal Neuronal Networks - Towards an interface of computational and experimental analysis CEREBELLUM - Computation and Plasticity in the Cerebellar System: Experiments, Modeling and Database FET Projects related to Neuron on SiliconNACHIP  - DeveloPment of A Neuro-semiconductor Interface wiht recombinant sodium CHannels NEUMIC - Neurons adn Modified CCMOS integrated Circuit interfacing INPRO - Information Processing by Natural Neural Networks NEUROBIT - A bioartificial brain with an artificial bosy: traininga cultured neural tissue to support the purposive behavior of an artificual body Neuroinformatics LPS (Life-Like Perception-systems) projectsALAVLSI  Attend-to-learn and learn-to-attend with neuromorphic, analogue VLSI APEREST - Approximately Periodic Representation of Stimuli BIOLOCH - Bio-mimetic Structures for Locomotion in the human body CAVIAR - Convolution AER vision architecture for real-time CICADA - Cricket and spider inspired perception and autonomous decision automata CIRCE - Chiroptera-Inspired Robotic Cephaloid: a Novel Tool for Experiments in Synthetic Biology CYBERHAND - Development of a Cybernetic hand prosthesis LOCUST  - Life-like Object Detection For Collision Avoidance Using Spatio-temporal Image MIRRORBOT - Biomimetic multimodal learning in a mirror neuron-based robot ROSANA - Representation of stimuli as neural activity SENSEMAKER - A multi-sensory, task-specific, adaptable perception system SPIKEFORCE - Real-time spiking networks for robot control

Abeles et al have “seen” attractor states rumbling in monkey recordings

1995 - the theoretical expectation had been laid out in Europe++…

Systems of spin-like elements may dynamically relax

governed by the Hamiltonian

towards increasingly complex “discrete” attractor states

(Ising model) ferromagneticconstantijJ

disordered ijJ (e.g., S.K. model) + spin-glass state

jiijJ (Hopfield model) + memory states

A B C D

ArealizationandMemory in the Cortexmonkey

Main theoretical perspectives:

a) Content-based

b) Hierarchical

c) Statistical/modular

Discrete attractors, with unitsarranged in a cortical network

Object 2 in position 1

Object 1 in position 2

Object 1 in position 1..use the sheetto code position...

Neocortex poses the complication of topographic maps..

a structure which remains stable and self-similar across mammalian species

THE HIPPOCAMPUS

H

H

opossum human

monkey

DG

the (early) David Marr view, the basis for reverse engineering the hippocampus

(diagram by Jaap Murre, 1996)

Discrete attractors, with unitsarranged in a cortical network

Object 2 in position 1

Object 1 in position 2

Object 1 in position 1..use the sheetto code position...

..use multiple chartsto code environments...

position

identity

context

Freedom from topography stimulates creativity

Stefan and Jill Leutgeb (2004) find ideals nearly realized in the real brain

CA3 firing patterns seem to fall into a discrete number of continuous attractors, with minimal overlap

A B C

’global remapping’

Karel Jezek(Moser lab,

SPACEBRAINEU project)

…where the heck am I?

…here ? or maybethere?

AMOUSE

The statistical/modular perspective

The Braitenberg model

N pyramidal cells

√N compartments

√N cells each

A pical synapses

B asal synapses

Striatal Networks

Cerebellar Networks

Climbing fibers:

The one case of an

almost private teacher

in the brain

Who is cutting-edge, in

cerebellar technology?

Spinal cord

Olfactory bulb

Tectum

Cerebellum

++-Expansion recoding,

Private teachers

Basal ganglia

-(-)-Massive funnellingTonic output firing

Hippocampus

(+)n+DG input sparsifierCA1 feed-forward

Neocortex

(+)n+Lamination, Arealization

Computationalparadigms

100’s Myrs old

that we failto understand

mammalian species

106

107

108

109

yrs

A Simplified History of Cortical Complexity

lizard

CA1CA3

DG

platypusechidna

infinite recursion

12

3

Where isIntelligentDesign ?

HippocampusCortex

reptilians

The statistical/modular perspective

The Braitenberg model

N pyramidal cells

√N compartments

√N cells each

A pical synapses

B asal synapses

Simulations which include a model of neuronal fatigueshow that the Braitenberg-Potts semantic networkcan hop from global attractor to global attractor:

Latching dynamics

SimulationsSimulations which include a model of neuronal fatiguethe Braitenberg-Potts semantic network

of

Latching forward and forward…

Systematic simulations indicate a latching phase transition

pl

pl

How might a capacity for indefinite latching have evolved?

pc C S 2

Storage capacity (max p to allow cued retrieval)

a spontaneous transition to infinite recursion?

+L+Lp p

SC

pl S ?

Latching onset (min p to ensure recursive process)

AM AM

long-range conn (local conn )

sem

anti

cs

sem

anti

cs

Syntactic structure

S DP I’ (singular/plural)I’ I VPI (singular or plural form)VP (Neg) (AdvP) V’V’ V DP | V S’ | V PPPP Prep DPS’ Compl SDP Det NP | PropN NP (AdjP) N”N” (AdjP) N’N’ N (PP) | N S’Det Art | Dem

BLISS (M Katran, M Nikam, S Pirmoradian

with guidance by Giuseppe Longobardi)

N boy | girl | cat | dog | tiger | jackal | horse | cow | meat | hay | milk | wood | meadow | stick | fork | bowl | cart | table | house || boys | girls | cats | dogs | tigers | jackals | horses | cows | stables | sticks | forks | bowls | carts | tables | houses

PropN John | Mary || John and Mary

V chases | feeds | sees | hears | walks | lives | eats | dies | kills | brings | pulls | is || chase | feed | see | hear | walk | live | eat | die | kill | bring | pull | are |

Compl that | whether

Prep in | with | to | of | under

Neg does not || do not (note singular negation removes sing inflection of verb)

Art the | a(an)

AdjP red | blue | green | black | brown | white | yellow | slow | fast | rotten | fresh | cold | warm | hot

AdvP slowly | rapidly | close | far

Dem this | that || these | those

+ semantic correlations

to assess how the model can learn, we shall need a toy:a basic language including both syntax and semanticsthat should be scaled down to manageable proportions

VP (Neg) (AdvP) V’V’ V DP | V S’ | V PPPP Prep DP

?≠

MediumTerm:

Not just with language

Francesco Battaglia thinks he has an algorithm..

S1 -> DP11 VP11 | DP12 VP12 probability 0.2 , 0.8VP11 -> VR11 | Neg11 VR12 probability 0.8000 , 0.2000VP12 -> VR12 | Neg12 VR12 probability 0.8000 , 0.2000VR11 -> Vt11 DP | Vi11 | Vi11 PPV | Vtd11 SRd | Vti11 SRi | Vtdtv11 DP PPT probability 0.25, 0.05, 0.15, 0.2, 0.2, 0.15VR12 -> Vt12 DP | Vi12 | Vi12 PPV | Vtd12 SRd | Vti12 SRi | Vtdtv12 DP PPT probability 0.25, 0.05, 0.15, 0.2, 0.2, 0.15PP -> Prep DP11 | Prep DP12 probability 0.5, 0.5PPV -> PrepV DP11 | PrepV DP12 probability 0.5, 0.5PPT -> PrepT DP11 | PrepT DP12 probability 0.5, 0.5SRd -> Conjd S1 probability 1SRi -> Conji S1 probability 1DP -> DP11 | DP12 probability 0.75 , 0.25DP11 -> Det11 NP11 | PropN11 probability 0.8 , 0.2DP12 -> Det12 NP12 | NP12 | PropN12 probability 0.495, 0.5 , 0.005NP11 -> N11 | AdjP N11 | N11 PP | AdjP N11 PP probability 0.4 , 0.2, 0.2, 0.2NP12 -> N12 | AdjP N12 | N12 PP | AdjP N12 PP probability 0.4 , 0.2, 0.2, 0.2Det11 -> Art11 | Dem11 probability 0.83 , 0.17Det12 -> Art12 | Dem12 probability 0.83 , 0.17N11 -> man | woman | boy | girl | child | book | friend | mother | table | house | paper | letter | teacher| parent | window | wife | bed | cow | tree | garden | hotel | lady | cat | dog | horse | brother | husband | daughter | meat | milk | wood | fork | bowl | cart | farm | kitchen probability 0.053851, 0.051611, 0.050752, 0.049736, 0.048145, 0.046406, 0.04494, 0.044608, 0.043687, 0.035414, 0.034214, 0.032266, 0.030239, 0.029286, 0.028074, 0.02795,

.,

Sahar

BLISS: so far, we have implemented only syntax

the red paper sits with hot cats on the strong windowswonderful friends don't sit for a short kitchenthe hot houses on cows don't find Johnteachers find a horsethose rotten pens in the cow live in a meatthe dogs don't wonder whether pens of the meat of that girl of the house walk with long farmsstrong mothers give tables in that black meat to the red cartsgood boys hear Phoebegirls die for a strange dogkitchens with Mary give great husbands to the good housesthese papers in mothers send the mother on the bed to the dogs in housesshort dogs on farms of the hotels give Phoebe to the parentsa cold lady with the women wonders whether Joe returns in a strange woman with the friendsteachers of the man know whether these trees sit in the wonderful cat with tablesthe husbands come in friendsgirls with John find JoeJoe uses a old fork of the great beds of the parents in the strange bed of children of the gardenJohn wonders whether the parents on trees with the bowls don't give a cartstrange wives run in tablesJohn sends Mary to Phoebemen send the hot dog to the horse

How muchconstrainedis syntacticdynamics?

We ask thatwith BLISS..

…we take the same measure from latchingPOTTS nets

creativity

memory

carabinieri