Convolutional Neural Networks with TensorFlow

TENSORFLOW VS
PYTORCH 1
TENSORFLOW VS PYTORCH
Authors Name/s per 1st Affiliation (Author)
Dept. name of the organization
Name of organization, acronyms acceptable
City, Country
e-mail address
Authors Name/s per 2nd Affiliation (Author)
Dept. name of the organization
Name of organization, acronyms acceptable
City, Country
e-mail address

TENSORFLOW VS
PYTORCH 2
TensorFlow
TensorFlow (Google ́s TensorFlow) is
currently one of the most common global
deep learning library, Google employs
machine learning for translation, search
engine and image capturing. Because of this
deep learning library, the users of Google
can experience a faster and more §liable
search which is done with Artificial
Intelligence [1]. Through this library,
Google can provide a recommendation of
the texts for example, when one types
convolut... Google will suggest the other
remaining words like convolutional neural
network as seen in the image below;
Figure 1: Showing Google suggestion of the
next words in a sentence through
TensorFlow
[1]
TensorFlow was developed by the brain of
Google team to help accelerate deep neural
and machine learning research, it was built
with a world massive computer. This library
was developed to run on several GPCs,
mobile operating system and CPUs.
TensorFlow has multiple wrappers which
operate in several languages like the Java,
C++ and Python.
Working principle of TensorFlow
TensorFlow enables the designers to
develop dataflow graphs (these are
structures which elaborate how data flows
via a series or a graph of processing node).
Every graph ́s node signifies a mathematical
operation and every edge or connection
between the nodes is an array of
multidimensional data. The programmer can
realize all these through the use of Python
language. It is easier to study and to deal
with Python language as opposed to other
substitute programming languages like C++
and Java. Python also provides offers

TENSORFLOW VS
PYTORCH 3
appropriate ways of expressing how high-
level abstraction is possible to be connected
together. The tensors and nodes in
TensorFlow are objects of Pythons while the
applications of the TensorFlow are
applications of Python language.
The real mathematical operations
are not performed using python language.
The transformation library which is present
in TensorFlow is coded through C++ of
higher performance binaries. The python
language is hence employed to direct the
traffic between the pieces and also to give
higher-level programming abstractions to
connect them. The usage of TensorFlow can
run on any target which is suitable for
example android devices, GPUs, CPUs, iOS,
local machine and cluster in the cloud. In
case one uses Google ́s own cloud he will be
able to run TensorFlow on a Google ́s
custom TPU (TensorFlow Processing Unit.
The resultant models developed through
TensorFlow can be employed on several
devices where they are employed to operate
to serve predictions. Distributed training is
simple to run because of the new API and
due to the use of TensorFlow lite which
makes it probable to use models on several
platforms. Nevertheless, the written codes
for the easier version of TensorFlow need to
be rewritten (sometimes significantly while
sometimes slightly).
For the below example
import numpy as np
import tensorflow as tf
The TensorFlow will be imported as
tf , with Python language, it is a very
common practice to employ short name for a
library. This is beneficial as it avoids typing
the whole name of the library when one
needs to use it just as elaborated above. For
example, one can import TensorFlow as tf
then call tf when he wants to use it as a
function of TensorFlow. When we can

TENSORFLOW VS
PYTORCH 4
multiply X_1 and X_2, the TensorFlow will
develop a node to help for the connection of
the operation. For this example, this is called
multiplication. If the graph is determined,
then the computational engine of
TensorFlow will multiply together X_2 and
X_1 as illustrated in the diagram below;
Figure 2: Showing multiplication of X_2 and
X_1
[2]
TensorFlow is currently used vastly by
startups, business firms and companies to
develop new systems as well as automate
things. This framework draws a reputation
from its distributed training support,
development options, support several
gadgets like Android and scalable
production.
PyTorch
This is an open-source learning machine
library which is based on Torch library, this
is used for computer vision and natural
language processing applications. This
framework is basically created by Facebook
́s artificial intelligence research lab. This is
an open and free source software which was
unleashed under a modified BSD license.
Operation of PyTourch
PyTourch is developed with aims that
make it different from other frameworks of
deep learning. Since PyTourch is the first
Python framework, it took a big leap over
other types of the framework which is
executed through a Python wrapper on a C /
C++ monolithic engine [2]. Through
choosing several configurations PyTorch
site tool shows that one needs the latest