The colloquium will also be streamed online.
Prof. Wolfgang Losert, University of Maryland
Invited by Prof. Joachim Rädler and Prof. Friedhelm Serwane
This presentation will review the potential role of astrocytes in neural network intelligence, exploring potential impact on both neuroscience and computing through a combination of experimental studies on living neural networks, and through the development of algorithms. Our focus is on the multimodal character of information, assessing the potential that information may be processed through non-electrical signals. Non-electrical signals are based on the concept of mechanochemical excitability, which enables rhythmic intracellular oscillations and waves on the tens of seconds to minute timescales, which in turn drive both mechanical and biochemical actions. With such oscillations, astrocytes become potential active contributors to brain computing. I will introduce a novel Machine Learning algorithm that harnesses these oscillations, which we named “rhythmic sharing.” The algorithm excels at anomaly detection and decreases the tendency of convolutional neural networks to forget prior knowledge when trained on new data.