Connectionists: Hippocampal formation: information-theoretical model

[Andras Lorincz]

Dear All:

 

I would like to draw your attention to a recent information-theoretically
motivated model of the entorhinal-hippocampal (ECHC) loop [1].  I also
present a general representational architecture that links the computational
functions of the ECHC loop in association with sensory processing and
reinforcement learning [2]. Related publications can be found here
http://nipg.inf.elte.hu

 

[1]
@article{lorincz09here,
  author = {A. Lorincz and G. Szirtes},
  title = {Here and now: how time segments may become events in the
hippocampus},
  journal = {Neural Networks},
  year = {2009},
  volume = {22},
  pages = {738-747},
  url =
{http://nipg.inf.elte.hu/index.php?option=com_remository&Itemid=27&func=file
info&id=179},
}
 
Abstract: The hippocampal formation is believed to play a central role in
memory functions related to the representation of events. Events are usually
considered as temporally bounded processes, in contrast to the continuous
nature of sensory signal flow they originate from. Events are then organized
and stored according to behavioral relevance and are used to facilitate
prediction of similar events. In this paper we are interested in the kind of
representation of sensory signals that allows for detecting and/or
predicting events. Based on new results on the identification problem of
linear hidden processes, we propose a connectionist network with
biologically sound parameter tuning that can represent causal relationships
and define events. Interestingly, the wiring diagram of our architecture not
only resembles the gross anatomy of the hippocampal formation (including the
entorhinal cortex), but it also features similar spatial distribution
functions of activity (localized and periodic, 'grid-like' patterns) as
found in the different parts of the hippocampal formation. We shortly
discuss how our model corresponds to different theories on the role of the
hippocampal formation in forming episodic memories or supporting spatial
navigation. We speculate that our approach may constitute a step toward a
unified theory about the functional role of the hippocampus and the
structure of memory representations.
 
[2]
@incollection{lorincz09learning,
  author = {A. Lorincz},
  title = {Learning and Representation: From Compressive Sampling to the
'Symbol Learning Problem'},
  booktitle = {Handbook of Large-Scale Random Networks},
  year = {2009},
  editor = {B. Bollobas and R. Kozma and M. Dezso},
  volume = {18},
  series = {Bolyai Society Mathematical Studies},
  pages = {445-488},
  address = {Berlin, Germany},
  publisher = {Springer},
  url =
{http://nipg.inf.elte.hu/index.php?option=com_remository&Itemid=27&func=file
info&id=176},
}
 
Abstract: In this paper a novel approach to neurocognitive modeling is
proposed in which the central constraints are provided by the theory of
reinforcement learning. In this formulation learning is (1) exploiting the
statistical properties of the system's environment, (2) constrained by
biologically inspired Hebbian interactions and (3) based only on algorithms
which are consistent and stable. In the resulting model some of the most
enigmatic problems of artificial intelligence have to be addressed. In
particular, considerations on combinatorial explosion lead to constraints on
the concepts of state-action pairs: these concepts have the peculiar flavor
of determinism in a partially observed and thus highly uncertain world. We
will argue that these concepts of factored reinforcement learning result in
an intriguing learning task that we call the symbol learning problem. For
this task we sketch an information theoretic framework and point towards a
possible resolution. 

 

 

Andras Lorincz

ECCAI Fellow

Department of Software Technology and Methodology

Eötvös Loránd University

Email: andras.lorincz at elte.hu

Group: http://nipg.inf.elte.hu 

Personal: http://people.inf.elte.hu/lorincz

 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: https://mailman.srv.cs.cmu.edu/mailman/private/connectionists/attachments/20090830/0c4b2c4e/attachment.html