ELM solutions for event-based systems

Abstract

Whilst most engineered systems use signals that are continuous in time, there is a domain of systems in which signals consist of events, or point processes. Events, like Dirac delta functions, have no meaningful time duration. Many important real-world systems are intrinsically event-based, including the mammalian brain, in which the primary packets of data are spike events, or action potentials. In this domain, signal processing requires responses to spatio-temporal patterns of events. We show that some straightforward modifications to the standard ELM topology produce networks that are able to perform spatio-temporal event processing online with a high degree of accuracy. The modifications involve the re-definition of hidden layer units as synaptic kernels, in which the input delta functions are transformed into continuousvalued signals using a variety of impulse-response functions. This permits the use of linear solution methods in the output layer, which can produce events or point processes as output, if modeled as a classifier with the two output classes being "event" or "no event". We illustrate the method in application to a spike-processing problem.

Skimming Digits: Neuromorphic Classification of Spike-Encoded Images

“…Prior works have explored the theoretical performance of SKIM networks using pre-determined patterns with varying levels of noise and jitter. The authors applied the technique to the Mus Silica dataset in the original SKIM paper Tapson et al ( 2013 ) and then later applied the SKIM to a real-world separation problem Tapson et al ( 2015 ). Others have used the algorithm to determine angle and direction in biological motion estimation, and in gait detection Lee et al ( 2014 ).…”

Event-driven spectrotemporal feature extraction and classification using a silicon cochlea model

“…The FEAST method in Afshar et al (2020c) extracts spatio-temporal features for event-based vision data using real-valued exponentially decaying kernels and 2-D “neurons.” The use of exponentially decaying kernels for event-based processing was described in Tapson et al (2015) and called a “time surface” in Lagorce et al (2015) . The time surface is generated by applying an exponential decay with a time constant τ v on a local (typically square) neighborhood centered on the current event.…”

Hybrid extreme learning machine approach for heterogeneous neural networks

“…TROP-ELM proposed by Miche et al [43] is an improvement of OP-ELM that uses a L 1 regularization penalty to rank the hidden layer neurons followed by a L 2 penalty on the regression weights for numerical stability. The synaptic kernel inverse method (SKIM) by Tapson et al [59] for event-based systems redefines the hidden neurons as synaptic kernels in which the input event based signals are transformed into continuous-valued signals. The advanced ELM ensemble (AELME) from Abuassba et al [1] constructs a network ensemble by training a randomly chosen ELM classifier on a subset of training data selected through random resampling.…”