Navigation and service

PGI Colloquium: Prof. Dr. Emre Neftci, Forschungszentrum Jülich (PGI-15), Germany; previously Univ. of California, USA (Online Event)

Online Talk

Please note: You will receive the link to the online talk in the e-mail invitation, usually sent out a few days before the lecture takes place. It is also available on request from the contact person below.

26 Nov 2021 11:00

Brain-inspired learning on neuromorphic substrates


The potential of machine learning to advance artificial intelligence is driving a quest to build dedicated systems that accelerate neural networks under real-world conditions. A natural approach to this is to take inspiration from neuroscience and build neuromorphic systems that emulate the biological processes of the brain using digital, mixed-signal and emerging nano-technologies. 

Machine Learning (ML) concepts can be used as a guiding principle for the development of neuromorphic systems, but brain-like efficiency comes with constraints on communication, dynamics, and reliability that are typically ignored in ML. In this talk, I will describe two examples that can reconcile ML theories with neuromorphic systems' constraints. First, I will present gradient descent approximations that pave the way to local, hardware-friendly synaptic plasticity rules that can achieve Deep Learning-level performance. Second, I will introduce Neural Sampling Machines (NSM), a class of stochastic neural networks that exploit device stochasticity in binary neural networks for probabilistic inference and learning. 

These algorithms and the neuromorphic systems they empower can enable new applications for dynamic real-world problems where on-chip adaptability, latency, or energy efficiency are critical.


Dr. Martina Luysberg
Phone: +49 2461 61-2417
Fax: +49 2461 61-6444