|Date:||2018, April 25|
|Author:||Mossakowski, Till and Glauer, Martin|
|Title:||Fuzzy Logic for Conceptors|
Conceptors are an approach to neuro-symbolic integration based on recurrent neural networks. Conceptors can learn and classify input signals that vary over time. The signal is fed into a recurrent neural network, and the resulting sequence of states is characterised to live in a higher-dimensional ellipsoid using conceptors.
We discuss Herbert Jaeger's two-valued logic for conceptors and argue that conceptors are intrinsically fuzzy in nature. Therefore, a fuzzy logic for neural conceptors seems to be more suitable than a two-valued one. We present such a fuzzy logic, which forms a fuzzy institution.