Instrumentation & Measurement Magazine 23-6 - 14

Fig. 5. Typical architecture of a CNN.

Both convolution and subsampling layers are used for feature extraction. Specifically, as shown in Fig. 5, convolution
layers produce feature maps while subsampling layers reduce
the sizes of these maps without losing the key information in
them. The final output of these layers is fed into the fully connected layer which handles the task of classification. CNNs are
typically used in computer vision and speech recognition applications [11].

Deep Boltzmann Machines and Deep Belief
Networks
In a Boltzmann machine, nodes are fully connected to each
other using undirected edges, as shown in Fig. 6a. The nodes
are divided into visible and hidden nodes, with no output
layer of nodes. In restricted Boltzmann machines (RBMs),
however, the visible nodes are only connected to hidden
nodes, and vice versa. Deep Boltzmann machines (DBMs) are
formed by stacking RBMs on top of each other. The hidden

nodes form full connectivity across subsequent layers.
The visible nodes are the ones
that receive the training set of
data. Compared to other DL
networks, DBMs are unique
in that the input visible nodes
are connected to each other.
The ultimate goal of a DBM
is to detect the distribution of
the given input training dataset. Typical applications of
DBMs are speech and object
recognition [11].
Deep belief networks
(DBNs) are based on stacking RBM layers on top of each other.
However, while top layers have undirected edges, lower layers have directed ones, as shown in Fig. 6b.

Generative Adversarial Networks
Generative adversarial networks (GANs) suggest an approach
for maximum likelihood estimation and employ two neural networks that compete against each other in a zero-sum game [16].
Basically, the GAN architecture utilizes a generator model and a
discriminator model. The generator is the system that is trained
to generate images, while the discriminator is the system that
classifies these images as accurate or not. This process of image
generation and classification is repeated until the generator is able
to produce accurate results. Examples of applications of GANs
include game development and artificial video generation [16].

Autoencoder
Autoencoding is a DL algorithm that efficiently compresses information and
learns how to reproduce accurate approximation of the
original information from
the compressed data. Since
they are used for data compression, autoencoders are
efficient tools for dimensionality reduction. Typical where
autoencoders are applied include data denoising and
dimensionality reduction for
data visualization [14].

Deep Learning in
Instrumentation
and Measurement

Fig. 6. a) Deep Boltzmann machine (DBM). The black nodes are hidden neurons while the white nodes are visible
neurons. b) Architecture of the deep belief network (DBN). The black nodes are hidden neurons and organized into
several layers while the white nodes are visible neurons. Lower layers use directed edges as opposed to fully
undirected edges in DBMs.
14

IEEE Instrumentation & Measurement Magazine

Here, we describe examples
of how DL is being used in
I&M literature, with a note
that our search was inexhaustive and mostly limited
to DL papers appearing in
September 2020

Instrumentation & Measurement Magazine 23-6

Table of Contents for the Digital Edition of Instrumentation & Measurement Magazine 23-6