a cognitive inspired model for search and retrieval Department of Physics and Applied Mathematics - University of Navarra Department of Neuroscience - Center for Applied Medical Research Net-Works 2008 June 10, 2008
Scheme 1 Introduction
Scheme 1 Introduction 2
Scheme 1 Introduction 2 3
Scheme 1 Introduction 2 3 4
Scheme 1 Introduction 2 3 4 5
Verbal fluency task Semantic memory: the memory subsystem that stores knowledge of concepts or meanings. Verbal fluency task: information retrieval test where fronto-temporal modulation is required. Information retrieval: Clustering, the production of words within semantic subcategories, and switching, the ability to shift between clusters.
Fronto-temporal modulation
Semantic verbal fluency task (animals) dog cat mouse lion tiger elephant whale shark dolphin... (sw) dog (cl) cat (cl) mouse (sw) lion (cl) tiger (cl) elephant (sw) whale (cl) shark (cl) dolphin. switching (sw) initiates a different category exploration while clustering (cl) explores current category.
Left. Model consisting of a switching mechanism (dashed circle) added to a network. Right. A particular simulation of the model, starting on stripped circle and combining topological random walks or clustering (solid arrows) with switching (dashed arrows).
We know the players... what about the game? A. Troyer s definition of human strategy on verbal fluency tasks: production of related words until current category is exhausted and then switching to a new category What is a category? How does switching work and what is its contribution to retrieval processes? Is it possible to move from one category to another in the absence of switching? What is the relevance of different fronto-temporal modulations on the verbal fluency success?
25 0.7 HWR=0.07 HWR=0.18 HWR=0.53 20 0.6 averaged word position 15 10 averaged word frequency 0.5 0.4 0.3 0.2 5 r = 0.80 p < 0.05 0.1 data cubic interp. 0 0 0.2 0.4 0.6 0.8 1 word frequency 0 10 20 30 40 50 word position Left. The large negative correlation found implies that the freq of a word is a good indicator of its expected position in the tests. Right. Three different stages: a decay on saying the most freq words, a plateau region of medium freq words and a final decay with unusual words. HWR stands for heterogeneity word rate.
An unsupervised model for inferring lexical networks Use statistics based on evidences rather than expert judgment. Switching mechanism produce meaningless co-occurrences. Our method: evaluate significance of co-occurrences rate respect to its probability expected by chance.
The basis of GTOM is to take into account the number of m-step neighbors that every pair of nodes share in a normalized fashion. For instance, selecting m = 1 is exactly TOM algorithm which measures the overlap coefficient O TOM for every pair of nodes i and j, O TOM (i,j) = J(i,j) min(k i,k j ), (1) where J(i,j) is the number of neighbors shared and min(k i,k j ) is the minimum degree (i.e. number of neighbors) of both nodes.
Id Description n < f > Most frequent Outliers 1 Farm-big 21 0.16±0.16 horse 1 2 Farm- and forest-small 16 0.15±0.15 hen 2 3 Cervidae 12 0.05±0.05 deer 2 4 Wild birds 23 0.11±0.10 eagle 0 5 Pets and singing birds 11 0.26±0.33 dog 0 6 Crustacean and mollusc 18 0.05±0.03 octopus, crab 2 7 Fish 31 0.11±0.14 whale 1 8 Unclassifiable 2 0.04±0.01 manta ray 2 9 Reptiles 21 0.10±0.11 snake 4 10 Rodents 5 0.21±0.18 mouse 1 11 Sabana and felinae 16 0.29±0.23 lion 0 12 Apes 6 0.12±0.12 monkey 1 13 Australian 5 0.08±0.06 kangaroo 2 14 Bears and Polar 9 0.10±0.11 bear 0 15 Wild Canis 3 0.12±0.09 wolf 0 16 Mammalian burrowers 4 0.05±0.03 platypus 0 17 Insects and Arachnids 32 0.09±0.09 fly 1
weasel swan goose badger magpie2 quail domestic goose coyote fox hen turkey duck squirrelpartridge hedgehog hare donkey rabbit parrot European green finch chamoiscaribou wild boar chicken donkey2 goat nightingalecapercaillie cockatoo mule ferret fallow deer moose Goat-kid cow canary condorbearded vulture small pig cock kite roe deer mare pig cat pigeongolden eagle antelope peacock cormorant barn owl goldfinch bull sheep sparrow hawk owl common kestrel mouflon reindeer pony colt horse dove calf dog vulture harrier male sheep sparrow deer true parrot European crow robin guinea pig seagull buffalo bison lamb ostrich cockroach gazelle bird swallow glow-worm mosquito ox stork mite wasp rat grasshopper eagle butterfly scorpion spider orangutan wolf boa little snake cicada falcon earthworm alligator gecko ladybird budgerigar cricket scorpion2 porcupine monkey gorilla dromedary lynx manatee toucan snake bumblebee jackal camel toad worm horsefly gnu ant bee rhinoceros lizard caterpillar fly macaque leopard crocodile snail elephant mouse piranha wall lizard viper flea platypus tick chameleon rattle snake python dragonfly edible crab centipede chimpanzee giraffe puma praying mantis iguana beetle zebra cobra salamander silkworm louse lion pelican clam otter hippopotamus manta frog ray bear cheetah panther jaguar koala velvet crab mammut tiger hamster sea horse panda mole starfish polar bear hyena crab prawn kiwi jellyfish king prawn kangaroo ray fish European lobster llama turtleswordfish shark walrus cuttlefish anteater spider crab fish oyster Norway lobster echidna grouper killer sole whale penguinoctopus lobster mussel brown barnacle bear whale trout hake carp sea bream barb seal dolphin elephant seal cod squid sea lion tuna turbot monkfish northern flatfish anchovy pike hammer fish sea bass sardine skipjack tuna sperm whale conger salmon herring hilt-head bream Pajek
140 120 dog 14 12 python 100 10 accesibility 80 60 40 mouse horse hen cow elephant whale giraffe tiger lion cat difussivity 8 European lobster 6 4 r= 0.21 p<0.05 20 r=0.91 p<0.05 2 rat hen giraffelephant tiger lion whale dog cat 0 0 0.2 0.4 0.6 0.8 1 frequency 0 0 0.2 0.4 0.6 0.8 1 frequency Left. Accesibility: number of times a word is the starting of a cluster. Right. Diffusivity: average position of a word in participants clusters.
MFPT of SRW Mean first passage time measures (MFPT) measures averaged node-node reachability, i.e. first passage time (FPT). SRW uni : switching uniformly distributed P srw ij = (1 q)p rw ij + qp sw ij = (1 q) A ij K i + q 1 n. (2) SRW freq : switching follows a linear frequency gradient P srw ij = (1 q)p rw ij + qp sw ij = (1 q) A ij + q K n i k=1 f k f j (3) FPT is computed as a ergodic regular Markov chain where next movement of the SRW only depends on current state.
MFPT on the lexical network 750 700 650 SRW uni SRW freq 600 550 MFPT 500 450 400 350 300 250 200 0 0.2 0.4 0.6 0.8 1 psw (switching probability )
MFPT on synthetic networks 900 SRW= SRW+ SRW smallworld modular scalefree random MFPT 500 0 q 1 0 q 1 Left. Switching uniformly random distributed Middle. Switching with a positive degree gradient Right. Switching with a negative degree gradient 0 q 1
Lexical acquisition evolution: giant component dominance and delayed modularity. Presence of a neighborhood deactivation mechanism Verbal fluency impairment in neurodegenerative diseases.
a cognitive inspired model for search and retrieval Department of Physics and Applied Mathematics - University of Navarra Department of Neuroscience - Center for Applied Medical Research Net-Works 2008 June 10, 2008 Thanks for your attention!