Biologically Plausible Large Language Models - Prof. Dr. Radu Grosu (TU Wien)
Veranstaltungsort
Hörsaal AH IV, Informatik-Zentrum Hauptbau Hörn, Eingang AhornstraßeBeschreibung
In this talk I will present our journey at TUW towards constructing a biologically plausible
Large Language Model.
I will start with the Electrical-Equivalent-Circuit model used in Neuroscience, and discuss how
this model can be restated as a NeuralODE, in the form of a Liquid Time Constant Neural Network (LTC).
Using LTCs I will demonstrate how one can solve various robotic tasks, including autonomous lane keeping
with only 19 neurons. I will then show how to dramatically speed up convergence and accuracy,
in the setting of Liquid- Resistance Liquid-Capacitance Neural Networks (LRCs).
Finally, I will show how to scale up any recurrent nonlinear neural network, including our LRCs,
to a Large Language Model, in form of a Structured State Space Model (LrcSSM).
About the speaker
The research interests of Radu Grosu include the modeling, the analysis and the control of cyber-physical systems and of biological systems.
The applications focus of Radu Grosu includes smart-mobility, Industry 4.0, smart-buildings, smart-agriculture, smart-healthcare, smart-cities, IoT, cardiac and neural networks,
and genetic regulatory networks.
Radu Grosu among others is recipient of the NSF Career Award, and the ACM Service Award. He is an elected member of the IFIP, Working Group 2.2.
Invitation of Informatik RWTH, RIA and Regina
For this talk, there is no video available.