47
Internet of Things and Artificial Intelligence
Deep learning is inspired from the working of the human brain. This mimics the
concepts of identifying a “cat” by a child by identifying the cat’s behavior, shape, the tail,
etc. and then putting them all together for a bigger idea generation as the “cat” itself.
Following the example cited above, deep learning progresses through multiple layers
and divides the intuitive problem into parts, with each part mapped to a specific layer.
The first layer is the input or visible layer where the input is being provided, then a series
of hidden layers selected randomly for specific mapping with input data. In the image processing example, layer-wise information progress is made like: from input, pixels to edge
identification at the first hidden layer, then corners and contours by second hidden layer,
then parts of objects are identified in the third hidden layer, and finally the whole object is
identified at the last and final hidden layer. This is shown in Figure 3.3.
In this chapter, our focus is to answer the following in a IoT scenario based on deep
learning:
• The intuitive deep learning applications in smart city datasets
• The performance metrics used for better prediction that carry an intuitive
component
3.5.2 Complementing Deep Learning Algorithms with IoT datasets
Although extensive research is being carried out in the area of energy load forecasting and its suitability in using neural network (Bhattacharyya and Thanh, 2004;
Rodrigues et al., 2014), deep neural architecture is the most promising in this application
scenario (Marino et al., 2016). Following are some of the emerging strategies/techniques
that may be useful for complementing deep learning algorithms with IoT datasets.
Output
(object identity)
CAR
P ERSON
ANIMAL
ird hidden layer
(object parts)
Second hidden layer
(corners and
contours)
First hidden layer
(edges)
Visible layer
(input pixels)
FIGURE 3.3
Deep learning example. (From Goodfellow, L. et al., Deep Learning, MIT Press, 2016.)
Internet of Things and Artificial Intelligence
Deep learning is inspired from the working of the human brain. This mimics the
concepts of identifying a “cat” by a child by identifying the cat’s behavior, shape, the tail,
etc. and then putting them all together for a bigger idea generation as the “cat” itself.
Following the example cited above, deep learning progresses through multiple layers
and divides the intuitive problem into parts, with each part mapped to a specific layer.
The first layer is the input or visible layer where the input is being provided, then a series
of hidden layers selected randomly for specific mapping with input data. In the image processing example, layer-wise information progress is made like: from input, pixels to edge
identification at the first hidden layer, then corners and contours by second hidden layer,
then parts of objects are identified in the third hidden layer, and finally the whole object is
identified at the last and final hidden layer. This is shown in Figure 3.3.
In this chapter, our focus is to answer the following in a IoT scenario based on deep
learning:
• The intuitive deep learning applications in smart city datasets
• The performance metrics used for better prediction that carry an intuitive
component
3.5.2 Complementing Deep Learning Algorithms with IoT datasets
Although extensive research is being carried out in the area of energy load forecasting and its suitability in using neural network (Bhattacharyya and Thanh, 2004;
Rodrigues et al., 2014), deep neural architecture is the most promising in this application
scenario (Marino et al., 2016). Following are some of the emerging strategies/techniques
that may be useful for complementing deep learning algorithms with IoT datasets.
Output
(object identity)
CAR
P ERSON
ANIMAL
ird hidden layer
(object parts)
Second hidden layer
(corners and
contours)
First hidden layer
(edges)
Visible layer
(input pixels)
FIGURE 3.3
Deep learning example. (From Goodfellow, L. et al., Deep Learning, MIT Press, 2016.)