{"pk":27160,"title":"TRACX2: a RAAM-like autoencoder modeling\ngraded chunking in infant visual-sequence learning","subtitle":null,"abstract":"Even newborn infants are able to extract structure from a\nstream of sensory inputs and yet, how this is achieved remains\nlargely a mystery. We present a connectionist autoencoder\nmodel, TRACX2, that learns to extract sequence structure by\ngradually constructing chunks, storing these chunks in a\ndistributed manner across its synaptic weights, and\nrecognizing these chunks when they re-occur in the input\nstream. Chunks are graded rather than all-or-none in nature.\nAs chunks are learned their component parts become more\nand more tightly bound together. TRACX2 successfully\nmodels the data from four experiments from the infant visual\nstatistical-learning literature, including tasks involving low-\nsalience embedded chunk items, part-sequences, and illusory\nitems. The model also captures performance differences\nacross ages through the tuning of a single learning rate\nparameter. These results suggest that infant statistical learning\nis underpinned by the same domain general learning\nmechanism that operates in auditory statistical learning and,\npotentially, in adult artificial grammar learning.1","language":"eng","license":{"name":"","short_name":"","text":null,"url":""},"keywords":[],"section":"Posters: Papers","is_remote":true,"remote_url":"https://escholarship.org/uc/item/776340ph","frozenauthors":[{"first_name":"Robert","middle_name":"M.","last_name":"French","name_suffix":"","institution":"University Bourgogne Franche-Comté","department":""},{"first_name":"Denis","middle_name":"","last_name":"Mareschal","name_suffix":"","institution":"Birkbeck University of London","department":""}],"date_submitted":null,"date_accepted":null,"date_published":"2017-01-02T05:00:00+11:00","render_galley":null,"galleys":[{"label":"PDF","type":"pdf","path":"https://journalpub.escholarship.org/cognitivesciencesociety/article/27160/galley/16796/download/"}]}