CogSci Here We Come!

We’ve had a raft of papers accepted at the annual conference of the Cognitive Science Society.

Biologically-Based Neural Representations Enable Fast Online Shallow Reinforcement Learning
- M Bartlett, J Orchard, TC Stewart
This paper investigates the use of Spatial Semantic Pointers (SSPs) in reinforcement learning. It shows that RL algorithms like TD(0) and TD(lambda) can more easily solve spatial tasks using SSPs.

A model of path integration that connects neural and symbolic representation
- N Dumont, J Orchard, C Eliasmith
We show that Spatial Semantic Pointers (SSPs) can be used for spatial cognition. Our model uses spiking neurons that exhibit grid-cell and place-cell firing patterns.

Biological Softmax: Demonstrated in Modern Hopfield Networks
- M Snow, J Orchard
Modern Hopfield networks often use the Softmax function, which cannot directly be implemented locally. We present a modern Hopfield network that is more biologically plausible than its predecessors.

Fractional Binding in Vector Symbolic Architectures as Quasi-Probability Statements
- PM Furlong, C Eliasmith
This paper discusses the relationship between fractionally bound vectors in Vector Symbolic Architecture (VSA) and kernel density estimators. We show how statements in VSAs can be considered analogous to probability statements, and present sketches for networks that can not only represent probabilities, but also compute functions on probabilities, like computing entropy or the mutual information between two variables.