Page 42 - EETimes Europe June 2021

42 EE|Times EUROPE

Europe Can Lead the Way in Regulation of AI

on transparency, accuracy, and the elimination of algorithmic bias. Lack of a decisive stand on regulation now could pave the way for
Industry lobbyists may argue that tight regulation will risk stifling other models of regulation from around the world which may not give
innovation or effectively limit the market to tech giants that have the such high priority to these protections. By acting swiftly, the EU can
deep pockets needed to achieve compliance. But there is a lot at stake lead the way on this important subject once again. ■
here when it comes to the protection of people’s safety and funda-
mental rights. Sally Ward-Foxton is editor-in-chief of EE Times Weekend.

MARKET & TECHNOLOGY TRENDS
EU Project Looks to Mimic Biological

Neural-System Processing

By Maurizio Di Paolo Emilio

rtificial intelligence is consid-
ered the computational enabler
for technological innovation in
A the coming years. The internet
of things already makes extensive use of
deep-learning computational paradigms to
offer services for searching for information
on the web or for recognizing audiovisual
information, while the emerging internet
of everything (IoE) will manage and deliver Figure 1: Spiking convolutional neural network (CNN) 1
services that process data from billions of
networked sensors.
CEA-Leti announced its participation in program director at CEA-Leti, pointed out a self-driving car? What if the same is true of
the EU’s new MeM-Scales project, which aims in an interview with EE Times Europe. your results — and what if the results are also
to develop a class of algorithms, devices, and “Memory of this interaction forms in time- time-based, such as instructions given to a
circuits that will mimic the multi-timescale scales ranging from milliseconds (short-term self-driving car on when to turn and when to
processing of biological neural systems. memory) to months and years (long-term increase or decrease speed?
The results will be used to build neuro- structural changes),” said Vianello. “To Spiking neural networks (SNNs) are a
design systems that interact with the real solution to this problem (Figure 1). They can
The MeM-Scales project world, neuromorphic circuits need to mimic accept time-based inputs and produce time-
the multi-timescale processing of the brain.
based outputs. Instead of ordered layers, they
aims to develop algorithms, Therefore, these circuits are the critical ele- have more complex structures within them for
devices, and circuits that ments in the processing pipeline.” passing data between neurons, such as loops
or multidirectional connections. Because they
will mimic the brain’s NEURAL NETWORKS are more complicated, they require different
types of training and learning algorithms,
In a standard neural network (NN) model,
multi-timescale processing input data is first sent to the input neurons such as making changes to backward-
to enable both learning and and is then passed through hidden layers of propagation–like approaches to adapt to
other neurons via connections called syn-
spike behavior.
inference at the edge. apses. The data is transformed at each step, In general, SNNs are neural network
and the output from one layer is used as input paradigms that implement the biological
for the next layer. neuron by emulating the natural signals of
morphic computing systems that can effi- The data eventually arrives at the final the nervous system (spikes) and the pro-
ciently process real-world sensory signals output layer, which provides the prediction cessing mechanism of the natural neuron’s
and natural-time–series data in real time and — for example, a category classification or a spikes (mechanism of action). The pecu-
to demonstrate the concepts with a practical numerical value in a regression. There is no liarity of SSNs lies in the internal way of
laboratory prototype. Targeted applica- real-time element here; the input data is all processing information, i.e., as a sequence
tions include high-dimensional distributed transmitted at the same time, passes through of spikes (impulses).
environmental monitoring, implantable each hidden layer in order, and is output
microchips for medical diagnosis, wearable all at once. SIMULATING A NEURAL SYSTEM
electronics, and human-computer interfaces. But what if your input data doesn’t all Processing on multiple timescales is inspired
To interact with the real world, brains arrive at the same time cleanly — what if it’s by neural processing in the nervous system,
process and perceptualize the sensory signals time-series or time-related data in some other which occurs naturally on timescales ranging
in multiple timescales, Elisa Vianello, edge AI way, such as real-time input from sensors on from milliseconds (axonal transmission) to

JUNE 2021 | www.eetimes.eu