If you want to attend this years CSCA symposium on Distributed Processing Models, please register before friday the 20th of June at www.csca.nl -> symposium.

Below you can find more information about this years symposium.

Date: 27 June 2008, 9.30 - 17.00 PM

Place: Tinbergenzaal KNAW, het Trippenhuis (Kloveniersburgwal 29, Amsterdam)

A central issue in cognitive science is the nature of mental representations and how these might be represented in the brain. A highly influential approach to this problem is provided by neural network models. A distinguishing characteristic of all of these models is that they take neurobiological constraints into account, i.e. the question of how the brain provides the computational machinery that enables complex cognitive functions such as memory and language. One of the best known approaches is the connectionist modeling approach as described in the 1986 2-volume book Parallel Distributed Processing by James McClelland and David Rumelhart. To date, PDP and other connectionist models are probably the most influential computational approach in Cognitive Science.

On June 27 2008, the Cognitive Science Center Amsterdam of the University of Amsterdam will organize a symposium chaired by visiting professor McClelland in which a number of prominent European researchers will give presentations on their work within the general area of connectionist modeling. Speakers will include Eric Postma, Axel Cleeremans, Mike Page, Bob French and Jaap Murre. At the end of the symposium, professor McClelland will comment on the various lectures. The symposium will an excellent overview of the current status of connectionist modeling.

For more information, please contact Joost van der Meer, email: J.J.W.vanderMeer@uva.nl.

Program:

09:30 - 10:00 Opening

10:00 - 10:45 Eric Postma - Modeling the perceptual front-end for models of vision

Connectionist models are weak on their representation of perceptual inputs. Often, these inputs are represented by arbitrary real-valued or binary vectors. In this lecture, a biologically-inspired model of a perceptual front-end for connectionist models will be outlined. Each component of the front-end will be motivated by biological and psychological findings. In addition, the successful application of the front-end to models of active and passive visual recognition will be demonstrated.

10:45 - 11:00 Break

11:00 - 11:45 Axel Cleeremans - Connectionism & implicit learning
When one is conscious of something, one is also conscious that one is conscious. Higher-Order Thought Theory (Rosenthal, 1997) takes it that it is in virtue of the fact that one is conscious that one is conscious that one is conscious! Here, we ask what the computational mechanisms may be that implement this intuition. Our starting point is Clark and Karmiloff-Smith’s (1993) point that knowledge acquired by a connectionist network always remains knowledge in the network rather than knowledge for the network: While such networks may become exquisitely sensitive to regularities contained in their input-output environment, they never exhibit the ability to access and manipulate this knowledge as knowledge. Instead, knowledge can only be expressed through performing the trained tasks and remains forever embedded in the causal pathways that developed as a result of learning. To address this issue, we present simulations in which two networks interact. The states of a first-order network trained to perform a simple categorization task become input to a second-order network trained trained either as an encoder or on another categorization task. Thus, the second-order network observes the states of the first-order network and and has, in the first case, to reproduce these states on its output units, and in the second case, to use the states as cues in order to solve the secondary task. This implements a limited form of metarepresentation, to the extent that the second-order network’s internal representations become re-representations of the first-order network’s internal states. We explore how well this mechanism accounts for observed dissociations between performance and report in the different situations recently explored by Persaud et al. (2007). We conclude that this mechanism forms the basis of mental attitudes, that is, a cognitive system¹s understanding of the manner in which its first-order knowledge is held (belief, hope, fear, etc.). Consciousness, in this light, involves knowledge of the geography of one own¹s internal representations, a geography that is itself learned over time as the results of an agents attributing value to the various experiences it enjoys through interaction with itself, the world, and others.

11:45 - 12:30 Mike Page - Connectionist modeling of order memory

I shall briefly survey some of the major connectionist models of memory for serial order, with particular emphasis on models of the immediate serial recall (ISR) task. ISR involves the recall of a list of items in the correct order and most commonly employs verbal materials. In this form, it has been extensively researched and there are a number of competing modelling approaches. I shall describe various classes of model, including associative-chaining models, positional models, and ordinal models. The ability to perform ISR for verbal materials correlates with measures of word-learning, such as vocabulary size. I will briefly discuss possible mechanisms underlying this correlation, and extend the discussion to the relationship between short- and long-term memory for serial order in a number of domains.

Lunch + Poster presentation

14:00 - 14:45 Bob French - Associative and rule-based cognition

I will present a new neural network model of category learning developped by Rosemary Cowell and myself that addresses the question of how rules for category membership are acquired. A rule for category membership is said to have emerged when the observer disregards a significant subset of an object’s features and focuses only on a reduced subset in order to determine the object’s category. The architecture of the model comprises a statistical-learning (Kohonen) network and a rule network whose weights, crucially, emerge from the statistical network. The statistical-learning network is implemented with a neurobiologically plausible Hebbian learning mechanism and forms category representations on the basis of perceptual similarity. The rule network extracts rules from the statistical-learning network by exploiting noise in the system to discover which of the stimulus features are sufficient to determine category membership. Thus, rules emerge during training. These rules are weightings of individual features; weights are stronger for features that convey more information about category membership. We demonstrate that this model predicts a cognitive advantage in classifying perceptually ambiguous stimuli over a system that relies only on perceptual similarity. In addition, it produces reaction time data that reflect the level of agreement between the statistical- and rule- learning components. Finally, the model demonstrates that occasional feedback greatly enhances the categorization performance of the system, which has implications for the “poverty of the stimulus” debate.

14:45 - 15:00 Break

15:00 - 15:45 Jaap Murre - Modeling cortical-hippocampal interactions

15:45 - 16:15 James McClelland - Commentary and concluding comments

16:15 - 16:30 Discussion

16:30 Drinks