Fifth Groningen Spring School on Cognitive Modeling
– ACT-R, Nengo, PRIMs, error-driven learning, and dynamical systems. –
Date: March 30 to April 3, 2020
Location: Groningen, the Netherlands
Fee: € 250 (late fee after February 15 will be € 300)
More information and registration: www.cognitive-modeling.com/springschool <http://www.cognitive-modeling.com/springschool>
We are happy to announce the fifth Groningen Spring School on Cognitive Modeling (March 30 to April 3, 2020). This year, the Spring School will again cover four different modeling paradigms: ACT-R, Nengo, PRIMs, and error-driven learning. It thereby offers a unique opportunity to learn the relative strengths and weaknesses of these approaches.
Moreover, this year we are offering a lecture series on dynamical systems, which should be interesting for anyone looking into modeling cognitive dynamics at some level of abstraction. We recommend this lecture series as an excellent combination with Nengo, for those interested in neuromorphic computing.
The first day will provide an introduction to all five topics. From day two, spring school students will be asked to commit to one topic, for which they will attend lectures as well as tutorials. In addition, students can sign up for a second topic, for which they will attend lectures only. All students are invited to join a series of plenary research talks on the different paradigms.
On the first day, spring school students are asked to introduce themselves and their research interests in a poster session.
Please feel free to forward the information to anyone who might be interested in the Spring School.
We are looking forward to welcoming you in Groningen,
The Spring School team
ACT-R is a high-level cognitive theory and simulation system for developing cognitive models for tasks that vary from simple reaction time experiments to driving a car, learning algebra, and air traffic control. ACT-R can be used to develop process models of a task at a symbolic level. Participants will follow a compressed five-day version of the traditional summer school curriculum. We will also cover the connection between ACT-R and fMRI.
Nengo is a toolkit for converting high-level cognitive theories into low-level spiking neuron implementations. In this way, aspects of model performance such as response accuracy and reaction times emerge as a consequence of neural parameters such as the neurotransmitter time constants. It has been used to model adaptive motor control, visual attention, serial list memory, reinforcement learning, Tower of Hanoi, and fluid intelligence. Participants will learn to construct these kinds of models, starting with generic tasks like representing values and positions, and ending with full production-like systems. There will also be special emphasis on extracting various forms of data out of a model, such that it can be compared to experimental data.
How do people handle and prioritize multiple tasks? How can we learn something in the context of one task, and partially benefit from it in another task? The goal of PRIMs is to cross the artificial boundary that most cognitive architectures have imposed on themselves by studying single tasks. It has mechanisms to model transfer of cognitive skills, and the competition between multiple goals. In the tutorial we will look at how PRIMs can model phenomena of cognitive transfer and cognitive training, and how multiple goals compete for priority in models of distraction.
Teachers: Jacolien van Rij and Dorothée Hoppe (University of Groningen)
Error-driven learning (also called discrimination learning) allows to simulate the time course of learning. It is based on the Rescorla-Wagner model (Rescorla & Wagner, 1972) for animal cognition, which assumes that learning is driven by expectation error, instead of behaviorist association (Rescorla, 1988). The equations formulated by Rescorla and Wagner have been used to investigate different aspects of cognition, including language acquisition (e.g., Hsu, Chater, and Vitányi, 2011; St. Clair, Monaghan, and Ramscar, 2009), second language learning (Ellis, 2006), and reading of complex words (Baayen et al, 2011). Although error-driven learning can be applied for all domains in cognitive science, in this course we will focus on how it could be used for modeling language processing and language learning.
Dynamical Systems: a Navigation Guide
Teacher: Herbert Jaeger (University of Groningen)
This lecture-series gives a broad overview over the zillions of formal models and methods invented by mathematicians and physicists for describing “dynamical systems”. Here is a list of covered items: Finite-state automata with and without input, deterministic and non-deterministic, probabilistic), hidden Markov models and partially observable Markov decision processes, cellular automata, dynamical Bayesian networks, iterated function systems, ordinary differential equations, stochastic differential equations, delay differential equations, partial differential equations, (neural) field equations, Takens’ theorem, the engineering view on “signals”, describing sequential data by grammars, Chomsky hierarchy, exponential and power-law long-range interactions, attractors, structural stability, bifurcations, phase transitions, topological dynamics, nonautonomous attractor concepts. In the lectures, I try to work out the underlying connecting lines between the “dots” listed above.