Dear all,

Neural Computation Unit (Doya Unit) would like to invite you to a online seminar as follows.

Speaker: Alexandru Vasilache (FZI Forschungszentrum Informatik)

Title: Efficient Learning Systems: From Spiking Neural Networks to Reinforcement Learning

Abstract:This seminar presents a research trajectory toward energy-efficient and adaptive learning systems for embodied agents. The first part focuses on software-level sparsity optimization for Spiking Neural Networks (SNNs), introducing methods that reduce activation and connection density through spike encoding, hybrid SNN-ANN architectures, and spatial embedding. These approaches achieve energy reductions of several orders of magnitude while maintaining competitive performance across tasks in predictive maintenance, biomedical signal processing, neural decoding, and robotic control. An evolutionary framework for spatially embedded recurrent SNNs further demonstrates that bio-inspired controllers can achieve up to 1600× parameter reduction without major performance loss.

Building on these results, the second part explores model-based and hierarchical reinforcement learning as a pathway toward continual, on-device learning. The focus lies on enabling agents to represent and plan goals across overlapping and flexible temporal scales, improving sample efficiency and long-horizon credit assignment in continual learning settings.

 

Zoom URL:
https://oist.zoom.us/j/98953563139?pwd=NhDaOJQ92bNShdNyBzbryrnUIKbl1v.1
ID: 989 5356 3139
PW:59168

We hope to see many of you at the seminar.

Sincerely,
Neural Computation Unit
Contact: ncus@oist.jp