IEEE - Aerospace and Electronic Systems - July 2021 - 75

Munir et al.
Figure 4.
Energy reduction imparted by data fusion and AI as compared to
AI without data fusion.
respectively. For JTX2 CPU, AI with data fusion provides
a speedup of 3.8 and 9.8 for LeNet and AlexNet,
respectively, as compared to AI without data fusion. The
data fusion in JTX2 CPU and CNN model execution in
JTX2 GPU lead to a speedup of 9.6 and 9.3 for LeNet
and AlexNet, respectively.
Since GPUs serve as a decent platform for acceleration
of deep learning models, offloading the CNN inference
tasks to the JTX2 GPU engenders a speedup of 2.8 to
2.9 for AlexNet as compared to only using the JTX2
CPU. However, experimental results indicate that the
LeNet execution on GPUs results in lower performance as
compared to LeNet execution on CPU. This is due to the
small size of LeNet model (i.e., LeNet model consists of
only60 000 parameters), which can be efficiently run on
CPU without offloading. For LeNet inference on GPU, the
data transfer between CPU, memory, and GPU incurs
nonnegligible latency overhead that causes higher overall
execution time ofLeNet inference on GPU as compared to
CPU. Conversely, since AlexNet is a large CNN model
with60 million parameters, AlexNet inference on GPU
provides better performance than CPU because the computation
time for large data dominates the data transfer time.
ENERGY CONSUMPTION FOR COMBINED DATA FUSION
AND AI AT THE EDGE
Data fusion also imparts energy benefits to AI. Figure 4
depicts the energy reduction furnished by data fusion and
AI as compared to AI without data fusion across three edge
computing platforms and the two CNN models. Results
verify that the data fusion leads to huge energy savings for
AI. For Xeon CPU and JTX2 CPU, the energy reductions
furnished by data fusion and AI over AI without data fusion
are 56.11%-88.32% and 70.88%-88.42%, respectively.
For the data fusion and AI execution on GPU along with
the CPU in the JTX2 platform, energy reduction is
88.00%-88.52% as compared to AI without data fusion.
These energy savings are attained because with data fusion,
JULY 2021
Figure 5.
Accuracy of AI with data fusion as compared to AI without data
fusion.
same amount of input data can be processed much quicker
by the AI models as compared to the case without data
fusion (see section " Average Latency for Combined Data
Fusion and AI at the Edge " ). This lower latency ofAI with
data fusion also translates to lower energy consumption as
compared to AI without data fusion.
ACCURACY FOR COMBINED DATA FUSION AND AI AT
THE EDGE
AI with data fusion can obviously result in improved performance
as illustrated in section " Average Latency for
Combined Data Fusion and AI at the Edge " because inference
is performed on reduced data size. However, there
may be a concern regarding accuracy of combined data
fusion and AI because the fused data are different from
the original raw data. Hence, we measure the impact of
data fusion on accuracy using average softmax probability
as a metric by which the CNN model predicts a class. For
accuracy evaluations, we use Darknet framework [35]
with LeNet CNN model. The experimental setup is the
same as described in section " Experimental Setup. " We
generate one image with fusing 10 images by averaging
the pixel values, resulting in data size reduction by 90%.
For inference with data fusion, we use 100 fused input
images, whereas we use 1 000 original images in the case
of inference without data fusion.
Figure 5 shows data fusion impact on softmax probability
for CNN inference. Our CNN model has 10 output
classes C_i where i; i 2 0; 1; 2; 3; .. . ; 9.C_i denotes
class i that corresponds to the output for digit i (e.g., C_0
corresponds to digit 0 and C_9 corresponds to digit 9).
Results indicate that data fusion does not adversely affect
the softmax probability though there is a fluctuation in
softmax probability depending on the digit class. Results
show that data fusion imparts higher accuracy to AI on
average. For example, data fusion results in CNN inference
accuracy of 88.53% as compared to the accuracy of
87.51% without data fusion for all digit classes. Results
IEEE A&E SYSTEMS MAGAZINE
75
IEEE - Aerospace and Electronic Systems - July 2021

Table of Contents for the Digital Edition of IEEE - Aerospace and Electronic Systems - July 2021
