Program

Posters

(in INI Elk room, at left entrance of INI)

Arianna Rubino, PhD Student at Institute of Neuroinformatics, https://www.ini.uzh.ch

Stochastic dendrites enable online learning in mixed-signal neuromorphic processing systems

The stringent memory and power constraints required in edge-computing applications have made event-driven neuromorphic systems a promising technology. On-chip online learning provides such systems the ability to learn the statistics of the incoming data and to adapt to their changes. Implementing online learning on neuromorphic systems presents some challenges that can be solved in biological neural networks by dendritic compartments of cortical neurons, as shown in recent neuroscience studies. We propose spike-based learning circuits to implement stochastic dendritic online learning. The circuits are embedded in a prototype spiking neural network fabricated using a 180 nm process and we present both circuits and behavioral simulation results of this algorithm-circuits co-design approach.

Ana Stanojevic, PhD Student at IBM Research Zurich, EPFL, https://www.zurich.ibm.com/

Equivalence Of ReLU And Single-spike Neural Networks

Artificial neural networks (ANNs) with rectified linear units (ReLUs) are standard in solving many artificial intelligence (AI) tasks and pretrained weights are often available. Spiking neural networks (SNNs) are biologically inspired models in which neurons communicate through sequences of spikes, i.e. sparse binary sequences. SNNs offer a potential for energy efficient neuromorphic implementation, however the training of such networks is challenging. In this work we show that fully-connected ReLU and single-spike SNNs are equivalent. Therefore the SNN inference can be performed directly with the pretrained weights.

Gala Sofia Sánchez Rodríguez , MSc Student at Institute of Neuroinformatics, https://www.ini.uzh.ch

Target-based localization on mobile robotics using a neuromorphic audio sensor (Plus Live Demo in INI Red Room)

This project developed an audio-based, neuromorphic approach to mobile robot guidance. We localize a speech sound target through the difference in timing of arrival between one ear and the other (also known as Interaural Time Difference, used by birds and mammals). The localization is done through samples from microphones using the cross-correlation algorithm, spikes generated from said microphone samples, and spikes from an artificial cochlea sensor. Both the cross-correlation and artificial cochlea spike methods are implemented on Sunfounder's PiCar-4WD robot, with the spike-based approach yielding faster, real-time sound localization for continuous mobile robot guidance.