Page 23 - Handbook of Deep Learning in Biomedical Engineering Techniques and Applications

Chapter 1 Congruence of deep learning in biomedical engineering 11

Figure 1.10 Applying 2D Haar in the vertical direction.

• Model size reduction is much higher than solitary worth dete-
rioration (SVD), network pruning, and deep compression.
• Applying deep compression with 8-bit quantization, Squeeze-
Net yields a 0.66 MB model (363 smaller than 32-bit AlexNet)
with equivalent accuracy to AlexNet. Furthermore, applying
deep compression with 6-bit quantization and 33% sparsity
on SqueezeNet produces a 0.47 MB model (510 smaller
than 32-bit AlexNet) with equivalent accuracy. SqueezeNet is
indeed amenable to compression.

3. Background study

3.1 Need of security

Presently, transferring information over the web causes secu-
rity issues, so the information should be kept secure, safe, and
only accessible by the approved client. The need is to send the
right information stealthily. This means that the beneficiary
should have the option of comprehending the message. There
are two well-known ways to do this: cryptography and
steganography.

3.1.1 Types of security methods
There are different kinds of security strategies for data
concealing, for example, steganography, watermarking, and
cryptography.

3.1.1.1 Steganography
Steganography implies secret correspondence. The message is
implanted inside another item, known as spread work. Steganog-
raphy is characterized as a procedure of concealing messages by
utilizing a method that nobody else is aware of. Steganography is