IBYME   02675
INSTITUTO DE BIOLOGIA Y MEDICINA EXPERIMENTAL
Unidad Ejecutora - UE
congresos y reuniones científicas
Título:
TOWARDS A PHYSIOLOGICAL THEORY OF ABSTRACT LEARNING
Autor/es:
H. REY, D. A. GUTNISKY, B. S. ZANUTTO
Lugar:
Italia
Reunión:
Congreso; 39th Annual European Brain and Behaviour Society; 2007
Institución organizadora:
European Brain and Behaviour Society
Resumen:
The capacity to conceptualize contingencies abstractly is
an efficient way to save memory resources. In the past,
Same/Different (SD) discrimination, the simplest form of
abstract learning, was thought to have language training as
a necessary condition. However, experimental evidence have
shown that this task cannot only be solved by humans, but
also by chimpanzees, rhesus and capuchin monkeys, and pigeons.
In addition, recent neurophysiological data suggests
that neurons in prefrontal cortex of rhesus monkey code abstract
rules. Despite these evidence, how neural networks are
able to learn abstract rules is largely unknown.
We propose a minimal model that is able to learn the
SD task. The model is composed of an input layer, working
memory cells, SD cells and response neurons. Working
memory cells are connected to SD cells through depressing
synapses. Each different stimulus activates a defined subset
of working memory cells. The synapses impinging on the SD
cells that correspond to activated working memory cells become
depressed. A further presentation of the same stimulus
will elicit a lower response from these cells indicating that
the same stimulus was presented, irrespective to its particular
features. In contrast when the stimulus are different, the
synapses are not depressed, and the SD cells response are not
attenuated. Response neurons pool these neurons to provide
the correct response. The model is minimal in the sense that
all its components are already described in the literature, although
without explicit mention to its relevance for this task.
The model provides also a basic building block upon more
complicated abstract tasks can be accomplished, such as conceptual
matching to sample, learning sequences of stimuli, etc.