Page 80 - Handbook of Deep Learning in Biomedical Engineering Techniques and Applications

68 Chapter 3 Application, algorithm, tools directly related to deep learning

2.4 Torch tool
This tool is implemented with C. Easy-to-write codes for new
layers, computational schemes based on Lua, and pretrained
prototypes are considered to be the main advantages of this
tool. It supports basic features for indexing, slicing, transposing,
type casting, resizing, sharing storage, and cloning [8].
It provides a flexible and affordable N-dimensional array
whose object is used by most other packages, and it forms the
core object of the library. The Tensor also supports all the math-
ematical operations, such as max, min, sum, statistical distribu-
tions, and BLAS operations, such as dot product, matrixevector
multiplication, matrixematrix multiplication, matrixevector
product, and matrix product [9].
The torch also simplifies objects and serialization by providing
various convenient functions that are used throughout its entire
packages. The torch.class (classname, parentclass) function can
be used to create many object factories (classes). When the argu-
ment constructor is called, torch initializes and sets with a Lua ta-
ble, which makes the table as an object.
Objects developed with the torch factory can be initialized,
but they do not contain references to objects that cannot be seri-
alized. However, user data can be serialized if it is wrapped by or
metatable that provides both read( ) and write( ) methods [9].

2.5 Theano
Theano is an open-source platform released under the BSD li-
cense and was invented by the LISA (now MILA) group at the
University of Montreal, Quebec, Canada (home of Yoshua Ben-
gio). It is named after a Greek mathematician [10].
Theano is a compiler for mathematical expressions in Python.
It knows how to take code structures and make them into very
efficient code like NumPy, efficient native libraries like BLAS,
and native code (Cþþ) to run as fast as possible on CPUs or
GPUs. It uses clever code optimizations to squeeze performance
as possible from hardware.The actual syntax of Theano is very
symbolic, which can be easy for beginners used to normal soft-
ware development. Particularly, expressions are defined in the ab-
stract and then compiled and later actually used to make
calculations [11].
It was used to handle the types of computation required for
bulk neural network algorithms used in DL. It was one of the
initial libraries of its development started in 2007. Theano pro-
vides extensive installation instructions for the fundamental
operating systems: Windows, OS X, and Linux [11].