Synthetic Forager Workshop

At SfN 2009, which is taking place from October 17 until 21, 2009, a synthetic forager workshop will be held. The Synthetic Forager (SF) – project is a corporation of Universities with the single overarching goal to identify the neuronal, cognitive and behavioral principles underlying optimal foraging in rodents and to implement these principles in a real world foraging robot named “Synthetic Forager”. The robot uses a biologically based cognitive technology for autonomous exploration and foraging in indoor and outdoor environments. The approach and technologies developed in SF will have long-term implications to areas including cleaning robots, search and rescue systems, terrestrial and planetary systems, and humanitarian de-mining. Behavioral and neurophysiologic studies of rodent foraging as well as the computer-modeling based on these results will be discussed. Further on technologies for behavioral and neurophysiologic research as a video tracking system or an action-potential recording system will be shown.

We are very proud that Dr. John Lisman, Brandeis University, USA gives a guest talk at the workshop:
Cellular and Network Mechanisms for Memory Storage and Recall in the Hippocampus.

Dr. Paul Verschure from UPF, Barcelona, Spain will talk about
The multi-level neuronal organization of perception, cognition and action in a Synthetic Forager

Dr. Christoph Guger and Robert Prückl from g.tec, Austria will explain
RABAMS - Rodent automated behaviour analysis and manipulation system

Dr. Cyriel Pennartz from University of Amsterdam, Netherlands will speak about
Hippocampal and striatal ensemble representations in a foraging task requiring conditioning to cues and contexts.

For more information please visit:

http://www.iua.upf.edu/sf/

http://www.gtec.at

Abstracts

Cellular and Network Mechanisms for Memory Storage and Recall in the Hippocampus
John Lisman

The hippocampus is a memory structure, notably for the storage of sequence information. There is now a great deal of information about the anatomy and cellular properties of the hippocampus that constrain models of how the system works. Our work has analyzed two properties of hippocampal cells, the spatial properties of place cells and the phenomenon of phase precession.

The input to the hippocampus comes from grid cells of layer 2 of the entorhinal cortex. We have found that the transformation of grid cells to place fields can be quantitatively accounted for by a process that does not depend strongly on learning. Rather, the transformation is a simple consequence of the statistical properties of the summation of ~1400 grid cell inputs (of varying synaptic strength), in conjunction with a network-mediated winner-take-all process mediated by gamma frequency feedback inhibition. Our simulations show that the average dentate granule cell (the input cells of the hippocampus) has two place fields, in accord with data. Granule cells, in turn, excite CA3 cells (which also receive input from the grid cells). Our simulations of a second winner-take-all computation show that CA3 cells achieve ultimate spatial selectivity (i.e. have one place field), in accord with data.

In another series of studies we have analyzed the phase precession that can be observed in all hippocampal regions. As a rat traverses a place field, there are ~7-8 cycles of theta frequency oscillation. The phase precession refers to the fact that on each successive theta cycle, firing occurs at an earlier phase of the theta cycle. Moreover, theta is subdivided into discrete subcycles by higher frequency gamma oscillations. Firing tends to occur at a particular phase of gamma. Thus the overall neural code can be described as a discrete theta phase code. Understanding this code is important because it is now clear that the code is used to communicate with other brain regions including the prefrontal cortex and striatum. According to our interpretation of the phase precession, this code is used to read out the predicted sequence of upcoming places, cued by the current location and based on previous learning. Analysis of the wiring diagram of the hippocampus and theoretical considerations regarding how accurate sequence recall can be done suggest the following model. Consider positions A to F on a track. During learning, LTP leads to enhanced synaptic connections between CA3 cells representing position A and dentate granule cells representing position B (this occurs in feedback connections from CA3 and mossy cells to granule cells). These connections thus connect different places in the sequence (heteroassociation). A variety of studies now show that the recurrent connections within CA3 form an autoassociative network. We envision that dentate and CA3 work together to recall a sequence: the cue, A, is slightly corrupted by noise and so is designated A’. This is sent to CA3, where autoassociation converts the representation to its canonical form, A. CA3, through backprojections and strengthened connections from A cells to B cells, evokes B’ in the dentate. This slightly corrupted form is then sent to CA3, where it is converted to its canonical form, B. Through this back and forth process, the entire sequence A-F can be read out accurately during a theta cycle. On the next theta cycle, the cue will be B (because the animal has moved) and this will again evoke sequence recall. However, because the cue is now B, cell C (and other cells in the sequence) will fire with earlier theta phase, thus producing the phase precession.

The multi-level neuronal organization of perception, cognition and action in a Synthetic Forager
Paul Verschure

The single overarching goal of the SF consortium is to identify the neuronal, cognitive and behavioural principles underlying optimal foraging in rodents and to implement these principles in a real-world foraging artefact or the Synthetic Forager (SF.01). SF exploits our growing understanding of exploration and foraging behaviour in rodents, advances current theories of the neuronal and behavioural organization of foraging and transfers this understanding towards the construction of novel realworld synthetic cognitive technologies. The overall integration of the perceptual, cognitive and behavioural control systems of SF will be accomplished using a well established robot based cognitive architecture, called Distributed Adaptive Control (DAC) further informed by the formal analysis of rodent foraging.

In this presentation I will define the Distributed Adaptive Control (DAC) neuromorphic architecture (1,2) that proposes how different levels of the neuraxis - from the brainstem to the cerebral cortex - interact to give rise to perception, cognition and action. I will show how DAC in turn has given rise to novel hypotheses on perception and action. As an example, I will discuss the integration of spatial and cue information in the hippocampus and the acquisition and execution of rules in the pre-frontal cortex.

RABAMS
Christoph Guger, Robert Prückl

This part of the SF workshop will show technology required for place field detection including electrodes and amplifiers, will explain and show spike sorting and cutting principles and will demonstrate how to record spikes. Furthermore the position reconstruction of rats based on place cells will be demonstrated with a setup including a video based tracking system, headstages and biosignal amplifiers and a Simulink model for the real-time analysis. The detection of the rat's home base with EEG and ECG recording and mapping technology will be explained as well as newly developed smart objects for the automation of experiments.

Hippocampal and striatal ensemble representations in a foraging task requiring conditioning to cues and contexts.
Cyriel M.A. Pennartz

Dr. Pennartz will present a foraging task in which rats learn to explore different subspaces of a multipartite environment (Y-maze) and to learn to associate dynamic, local cues with reward. The task is set up so that all three subspaces (chambers) of the environment are visually identical; the animal therefore has to rely on its path-integration and allocentric mapping system, which is thought to involve the hippocampus, to keep track of which subspace he visits at any particular time (cf. Ito et al. J. Neurosci. 28: 6950-6959, 2008). Each of the three subspaces contains three reward sites with a cue light situated above it. In the first task phase, rats learn to approach a reward port once the cue light above this port switches on (cue conditioning phase). In the second task phase, not every cue presentation that is followed by an approach response to the port results in reward anymore; one of the chambers is now coupled to a high reward probability and the other two chambers to a low probability (contextual conditioning phase). We show how this task can be used to study foraging behavior guided by spatial contexts and by time-discrete cues, and how simultaneously recorded groups of hippocampal and ventral striatal neurons represent different aspects of the task setting.