MODULE 3: Convolutional Neural Networks

Convolutional Neural Networks –Architecture, Convolution operation, Motivation,

pooling .Variants of convolution functions, Structured outputs, Data types,

Efficient convolution algorithms, Applications of Convolutional Networks, Pre-

trained convolutional Architectures : AlexNet, ZFNet, VGGnet-19, ResNet 50.

CNN (Convolutional Neural Network)

A convolutional neural network (CNN), is a network architecture for deep

learning which learns directly from data. CNNs are particularly useful for
finding patterns in images to recognize objects. They can also be quite effective
for classifying non-image data such as audio, time series, and signal data.

Kernel or Filter or Feature Detectors

In a convolutional neural network, the kernel is nothing but a filter that is

used to extract the features from the images.

Formula = [i-k]+1

i -> Size of input , K-> Size of kernel

Press enter or click to view image in full size
Stride

Stride is a parameter of the neural network’s filter that modifies the amount of
movement over the image or video. we had stride 1 so it will take one by one. If
we give stride 2 then it will take value by skipping the next 2 pixels.

Formula =[i-k/s]+1

i -> Size of input , K-> Size of kernel, S-> Stride

Padding

Padding is a term relevant to convolutional neural networks as it refers to the

number of pixels added to an image when it is being processed by the kernel of
a CNN. For example, if the padding in a CNN is set to zero, then every pixel
value that is added will be of value zero. When we use the filter or Kernel to
scan the image, the size of the image will go smaller. We have to avoid that
because we wanna preserve the original size of the image to extract some low-
level features. Therefore, we will add some extra pixels outside the image.

Formula =[i-k+2p/s]+1

i -> Size of input , K-> Size of kernel, S-> Stride, p->Padding

Pooling

Pooling in convolutional neural networks is a technique for generalizing

features extracted by convolutional filters and helping the network recognize
features independent of their location in the image.
Flatten

Flattening is used to convert all the resultant 2-Dimensional arrays from

pooled feature maps into a single long continuous linear vector. The flattened
matrix is fed as input to the fully connected layer to classify the image.

Layers used to build CNN

Convolutional neural networks are distinguished from other neural networks

by their superior performance with image, speech, or audio signal inputs. They
have three main types of layers, which are:

 Convolutional layer
 Pooling layer

 Fully-connected (FC) layer

Convolutional layer

first layer that is used to extract the various features from the input images. In
this layer, We use a filter or Kernel method to extract features from the input
image.

Pooling layer

The primary aim of this layer is to decrease the size of the convolved feature
map to reduce computational costs. This is performed by decreasing the
connections between layers and independently operating on each feature map.
Depending upon the method used, there are several types of Pooling
operations. We have Max pooling and average pooling.
Fully-connected layer

The Fully Connected (FC) layer consists of the weights and biases along with
the neurons and is used to connect the neurons between two different layers.
These layers are usually placed before the output layer and form the last few
layers of a CNN Architecture.

Dropout

Another typical characteristic of CNNs is a Dropout layer. The Dropout layer is

a mask that nullifies the contribution of some neurons towards the next layer
and leaves unmodified all others.

Activation Function

An Activation Function decides whether a neuron should be activated or not.

This means that it will decide whether the neuron’s input to the network is
important or not in the process of prediction. There are several commonly used
activation functions such as the ReLU, Softmax, tanH, and the Sigmoid
functions. Each of these functions has a specific usage.

Sigmoid — For a binary classification in the CNN model

tanH - The tanh function is very similar to the sigmoid function. The only
difference is that it is symmetric around the origin. The range of values, in this
case, is from -1 to 1.

Softmax- It is used in multinomial logistic regression and is often used as the

last activation function of a neural network to normalize the output of a
network to a probability distribution over predicted output classes.

RelU- the main advantage of using the ReLU function over other activation
functions is that it does not activate all the neurons at the same time.

Convolution Operation
To apply the convolution:

 Overlay the Kernel on the Image: Start from the top-left corner of the image and place the
kernel so that its center aligns with the current image pixel.
 Element-wise Multiplication: Multiply each element of the kernel with the corresponding
element of the image it covers.
 Summation: Sum up all the products obtained from the element-wise multiplication. This sum
forms a single pixel in the output feature map.
 Continue the Process: Slide the kernel over to the next pixel and repeat the process across the
entire image.

Example of Convolution Operation

Convolution Operation
Key Terms in Convolution Operation

 Kernel Size: The convolution operation uses a filter, also known as a kernel, which is typically a
square matrix. Common kernel sizes are 3×3, 5×5, or even larger. Larger kernels analyze more
context within an image but come at the cost of reduced spatial resolution and increased
computational demands.
 Stride: Stride is the number of pixels by which the kernel moves as it slides over the image. A
stride of 1 means the kernel moves one pixel at a time, leading to a high-resolution output of the
convolution. Increasing the stride reduces the output dimensions, which can help decrease
computational cost and control overfitting but at the loss of some image detail.
 Padding: Padding involves adding an appropriate number of rows and columns (typically of
zeros) to the input image borders. This ensures that the convolution kernel fits perfectly at the
borders, allowing the output image to retain the same size as the input image, which is crucial for
deep networks to allow the stacking of multiple layers.

Types of Convolution Operations (Variants of Convolution)

1D Convolution

1D convolution is similar in principle to 2D convolution used in image processing.

In 1D convolution, a kernel or filter slides along the input data, performing element-wise multiplication
followed by a sum, just as in 2D, but here the data and kernel are vectors instead of matrices.

1D Convolution Operation
Applications:
1D convolution can extract features from various kinds of sequential data, and is especially prevalent in:

 Audio Processing: For tasks such as speech recognition, sound classification, and music analysis,
where it can help identify specific features of audio like pitch or tempo.

 Natural Language Processing (NLP): 1D convolutions can help in tasks such as sentiment
analysis, topic classification, and even in generating text.
 Financial Time Series: For analyzing trends and patterns in financial markets, helping predict
future movements based on past data.
 Sensor Data Analysis: Useful in analyzing sequences of sensor data in IoT applications, for
anomaly detection or predictive maintenance.

3D Convolution

3D convolution extends the concept of 2D convolution by adding a dimension, which is useful for
analyzing volumetric data.

Like 2D convolution, a three-dimensional kernel moves across the data, but it now simultaneously
processes three axes (height, width, and depth).

3D Convolution
Applications:

 AI Video Analytics: Processing video as volumetric data (width, height, time), where
the temporal dimension (frames over time) can be treated similarly to spatial dimensions in
images. The latest video generation model by OpenAI called Sora used 3D CNNs.
 Medical Imaging: Analyzing 3D scans, such as MRI or CT scans, where the additional dimension
represents depth, providing more contextual information.
 Scientific Computing: Where volumetric data representations are common, such as in simulations
of physical phenomena.

Dilated Convolution

A variation of the standard convolution operation, dilated convolution expands the receptive field of the
filter without significantly increasing the number of parameters. It achieves this by introducing gaps, or
“dilations,” between the pixels in the convolution kernel.

In a dilated convolution, spaces are inserted between each element of the kernel to “spread out” the
kernel. The l (dilation rate) controls the stride with which we sample the input data, expanding the
kernel’s reach without adding more weights. For example, if d=2, there is one pixel skipped between each
adjacent kernel element, making the kernel cover a larger area of the input.

Dilated Convolution
Features

 Increased Receptive Field: Dilated convolution allows the receptive field of the network to
grow exponentially with the depth of the network, rather than linearly. This is particularly useful
in dense prediction tasks where contextual information from a larger area is beneficial for making
accurate predictions at a pixel level.
 Preservation of Resolution: Unlike pooling layers, which reduce the spatial dimensions of the
feature maps, dilated convolutions maintain the resolution of the input through the network
layers. This characteristic is crucial for tasks where detailed spatial relationships need to be
preserved, such as in pixel-level predictions.
 Efficiency: Dilated convolutions achieve these benefits without increasing the number of
parameters, hence not increasing the model’s complexity or the computational cost as much as
increasing the kernel size directly would.

Dialated Convolution Operation

Dilated Convolution is applied in various tasks of computer vision. Here are a few of those:

 Semantic Segmentation: In semantic segmentation, the goal is to assign a class label to each pixel
in an image. Dilated convolutions are extensively used in segmentation models like DeepLab,
where capturing broader context without losing detail is crucial. By using dilated convolutions,
these models can efficiently enlarge their receptive fields to incorporate larger contexts,
improving the accuracy of classifying each pixel.
 Semantic Segmentation

 Audio Processing: Dilated convolutions are also used in audio processing tasks, such as in
WaveNet for generating raw audio. Here, dilations help capture information over longer audio
sequences, which is essential when predicting subsequent audio samples.
 Video Processing: In video frame prediction and analysis, dilated convolutions help in
understanding and leveraging the information over extended spatial and temporal contexts, which
is beneficial for tasks like anomaly detection or future frame prediction.

Transposed Convolution

Transposed convolution is primarily used to increase the spatial dimensions of an input tensor. While
standard convolution, by sliding a kernel over it produces a smaller output, a transposed convolution
starts with the input, spreads it out (typically adding zeros in between elements, known as upsampling),
and then applies a kernel to produce a larger output.
Standard convolutions typically extract features and reduce data dimensions, whereas transposed
convolutions generate or expand data dimensions, such as generating higher-resolution images from
lower-resolution ones. Instead of mapping multiple input pixels into one output pixel, transposed
convolution maps one input pixel to multiple outputs.

Unlike standard convolution, where striding controls how far the filter jumps after each operation, in
transposed convolution, the stride value represents the spacing between the inputs. For example, applying
a filter with a stride of 2 to every second pixel in each dimension effectively doubles the dimensions of
the output feature map if no padding is used.

Transposed Convolution Operation

The generator component of Generative Adversarial Networks (GANs) and the decoder part of
an AutoEncoder extensively use transposed convolutions.

In GANs, the generator starts with a random noise vector and applies several layers of transposed
convolution to produce an output that has the same dimension as the desired data (e.g., generating a
64×64 image from a 100-dimensional noise vector). This process involves learning to upsample lower-
dimensional feature representations to a full-resolution image.

Depthwise Separable Convolution

A depthwise convolution, an efficient form of convolution used to reduce computational cost and the
number of parameters while maintaining similar performance, involves convolving each input channel
with a different filter. The convolution takes place in two steps: Depthwise Convolution and then
Pointwise Convolution. Here is how they work:
Depthwise Convolution

 Depthwise Convolution: A single convolutional filter applies separately to each channel of the
input in depthwise convolution. A dedicated kernel convolves each channel. For instance, in an
RGB image with 3 channels, each channel receives its kernel, ensuring that the output retains the
same number of channels as the input.
 Pointwise Convolution: After depthwise convolution, pointwise convolution is applied. This
step uses a 1×1 convolution to combine the outputs of the depthwise convolution across the
channels. This means it takes the depthwise convolved channels and applies a 1×1 convolutional
filter to each pixel, combining information across the different channels. Essentially, this step
integrates the features extracted independently by the depthwise step, creating an aggregated
feature map.

In standard convolutions, the number of parameters quickly escalates with increases in input depth and
output channels due to the full connection between input and output channels. Depthwise separable
convolutions separate this process, drastically reducing the number of parameters by focusing first on
spatial features independently per channel and then combining these features linearly.

For example, if we have the following:

 Input Feature Map: 32 Channels

 Output Feature Map: 64 Channels
 Kernel Size for Convolution: 3 x 3
Standard Convolution:

 Parameters =3×3×32×64
 Total Parameters =18432

Depthwise Separable Convolution:

 Depthwise Convolution:
o Parameters= 3 x 3 x 32
o Parameters=288
 Pointwise Convolution:
o Parameteres= 1 x 1 x32 x 64
o Parameters= 2048
 Total Prameters= 2336

Applications in Mobile and Edge Computing

Depthwise separable convolutions are particularly prominent in models designed for mobile and edge
computing, like the MobileNet architectures. These models are optimized for environments where
computational resources, power, and memory are limited:

 MobileNet Architectures: MobileNet models utilize depthwise separable convolutions

extensively to provide lightweight deep neural networks. These models maintain high accuracy
while being computationally efficient and small in size, making them suitable for running on
mobile devices, embedded systems, or any platform where resources are constrained.
 Suitability for Real-Time Applications: The efficiency of depthwise separable convolutions
makes them ideal for real-time applications on mobile devices, such as real-time image and video
processing, face detection, and AR and VR.

Deformable Convolution

Deformable convolution is an advanced convolution operation that introduces learnable parameters to

adjust the spatial sampling locations in the input feature map. This adaptability allows the convolutional
grid to deform based on the input, making the convolution operation more flexible and better suited to
handle variations in the input data.
 3×3
Deformable Convolution

In traditional convolution, the filter applies over a fixed grid in the input feature map. However,
deformable convolution adds an offset to each spatial sampling location in the grid, learned during the
training process.

Adaptive Receptive Field in Deformable Convolution

These offsets allow the convolutional filter to adapt its shape and size dynamically, focusing more
effectively on relevant features by deforming around them. Additional convolutional layers designed to
predict the best deformation for each specific input learn the offsets.

Standard Convolution vs Deformable Convolution

Deformable convolutions have been successfully integrated into several state-of-the-art object detection
frameworks, such as Faster R-CNN and YOLO, providing improvements in detecting objects with non-
rigid transformations and complex orientations. Here are its applications:

 Image Recognition: It is beneficial in cases where objects can appear in different sizes, shapes, or
orientations.
 Video Analysis: Deformable convolutions can adapt to movements and changes in posture, angle,
or scale within video frames, enhancing the ability of models to track and analyze objects
dynamically.
 Enhancing Model Robustness: By allowing the convolutional operation to adapt to the data,
deformable convolutions can increase the robustness of models against variations in the
appearance of objects, leading to more accurate predictions across a wider range of conditions.
Convolutional Neural Networks (CNNs):
Motivation
Feedforward networks, or Multi-Layer Perceptrons (MLPs), are powerful tools. However,
when dealing with data like images, their standard structure reveals significant
limitations. Consider what happens when you feed an image into an MLP: typically, you
flatten the image matrix (e.g., a 28x28 pixel image becomes a 784-element vector).

This flattening process immediately discards important spatial information. Pixels that
were adjacent in the 2D grid, potentially forming a line or texture, are treated as
independent inputs after flattening. The network loses the inherent structure of the data;
it doesn't inherently understand that pixels (0,0) and (0,1) are closer and likely more
related than pixels (0,0) and (27,27). An MLP would have to learn these spatial
relationships from scratch, which is highly inefficient.

Furthermore, high-resolution images lead to enormous input vectors. A modest 224x224

color image has 224×224×3=150,528224×224×3=150,528 input values. Connecting this
input layer to even a moderately sized first hidden layer in an MLP results in an
explosion of parameters (weights and biases). Training such a network becomes
computationally expensive and requires substantial amounts of data to avoid overfitting,
as the model has excessive capacity relative to the underlying structure it's trying to
learn.

An MLP typically connects every input feature (flattened pixel) to every neuron in the
first hidden layer, ignoring spatial relationships and leading to a large number of
parameters.

Convolutional Neural Networks (CNNs) were developed specifically to address these

shortcomings. They are designed to process data that comes in the form of multiple
arrays, like a color image composed of 3D arrays (height, width, color channels). The
core motivation behind CNNs stems from three main ideas that make them highly
effective for tasks involving grid-like data:
1. Local Receptive Fields: Instead of connecting every input neuron to every hidden
neuron, CNNs employ neurons that only respond to a restricted region of the
input layer. This region is called the local receptive field. Imagine a small window
(e.g., 3x3 or 5x5 pixels) sliding over the input image. Neurons in the next layer
process information only within that window at a time. This directly uses the
spatial locality present in images; nearby pixels are likely related and contribute to
forming elementary visual features like edges or corners.
2. Parameter Sharing: This is a powerful concept. In a CNN, the same set of
weights (often called a filter or kernel) is applied across different locations in the
input image. If a filter is designed to detect a horizontal edge, it slides across the
entire image, detecting that edge wherever it appears. This dramatically reduces
the number of parameters compared to an MLP. Instead of learning separate
weights for detecting a feature at every possible location, the CNN learns a single
set of weights for that feature detector. This makes the network much more
efficient and less prone to overfitting.
3. Translation Invariance (Equivariance): A direct consequence of parameter
sharing is that CNNs possess a degree of translation invariance (more accurately,
equivariance). If the network learns to detect a pattern (e.g., a cat's eye) in one
part of the image using a specific filter, it can detect the same pattern if it appears
in a different location using the same filter. This is highly desirable for object
recognition and other image analysis tasks, as the object's identity doesn't change
based on its position in the frame.
A CNN neuron in a feature map connects only to a local patch (receptive field) of the
input. The weights defining this connection (the filter) are shared across different spatial
locations.

In essence, CNNs are motivated by the need to build models that respect the spatial
hierarchy of image data. They learn features locally, share parameters efficiently, and
build up complex representations from simpler ones, making them the standard
approach for many computer vision tasks. The following sections will detail the specific
operations, like convolution and pooling, that enable these properties.

DATA TYPES
In a Convolutional Neural Network (CNN), data is represented and processed using a
fundamental data structure called a tensor. Tensors are multi-dimensional arrays that
can hold data with different dimensions, shapes, and types. For CNNs, which primarily
handle image and video data, the input, intermediate feature maps, and final outputs are
all structured as tensors.
Input data types
The raw data fed into a CNN, such as an image, first needs to be loaded and pre-
processed into a tensor.

1. Image data (2D grid)

o Representation: A 2D array of pixels for grayscale images, or a 3D array with an

extra dimension for color channels (like RGB).

o Data type: Typically, raw images are stored as uint8 (unsigned 8-bit integer),
where each pixel value ranges from 0 to 255.

o Tensor shape: (Height,Width,Channels)

o Example: An RGB image with 256×256 pixels is a tensor of shape (256,256,3)

.
o Preprocessing: Before feeding into the model, the data is typically converted to
float32 and normalized to a range like [0, 1] or [-1, 1] for better training
stability.

2. Video data (3D grid over time)

o Representation: A sequence of image frames, adding a time or frame dimension

to the image tensor.

o Data type: Often handled as float32 tensors after pre-processing.

o Tensor shape: (Frames,Height,Width,Channels)

o Example: A 10-second video with 30 frames per second and 128×128 color
images would be a tensor of shape (300,128,128,3)

3. Medical imaging data (3D volumetric)

o Representation: Volumetric data from CT scans or MRI, adding a depth

dimension to the standard image format.

o Data type: Usually float32 after converting from specialized medical file
formats.

o Tensor shape: ( ℎ, ℎ, ℎ, ℎ )

o Example: A CT scan image might be a tensor of shape (64,512,512,1)

.
Data types for model training and inference

The choice of data type for model weights and intermediate computations is a critical design
decision that balances computational efficiency, memory usage, and numerical precision.

1. Full precision (float32): This is the standard data type for training most deep learning
models. It uses 32 bits for each value, providing high precision and a wide dynamic
range.

o Advantage: High numerical stability, especially for sensitive backpropagation

calculations, which rely on small gradients.

o Disadvantage: Requires significant memory and computation, making training slower

for very large models.

2. Mixed precision (float16 and bfloat16)

o Description: Many modern deep learning libraries and hardware (e.g., NVIDIA
Tensor Cores) use a mix of float32 and a 16-bit floating-point format to speed
up training.

o float16: Uses 16 bits. Halves memory usage and can significantly increase
training speed. It has a smaller dynamic range, making it potentially unstable for
some operations.

o bfloat16: Also 16 bits, but designed to have the same wide dynamic range as
float32 by trading some precision. This makes it more stable than float16 for
training.

o Method: For mixed-precision training, the model's master weights are typically
kept in float32, while computationally intensive parts of the forward and
backward pass are done in float16 or bfloat16.

3. Quantized data types (int8)

o Description: Quantization is the process of converting a model's weights and

activations from floating-point types (like float32) to low-precision integers
(like int8).

o Application: This is primarily used for inference on edge devices with limited
resources, such as mobile phones or IoT devices. It is less common for training
due to the significant loss of precision.

o Advantage: Drastically reduces model size, memory footprint, and computational

requirements, leading to faster inference and lower power consumption.
 o Process: The trained float32 model is "quantized" for deployment.

Efficient convolution algorithms

Efficient convolution algorithms are crucial in deep learning due to the computational intensity of
convolutional layers. Several approaches aim to optimize these operations:

1. Winograd Convolution:
 This algorithm reduces the number of multiplications required for convolution, especially for
small kernel sizes common in CNNs.
 It achieves this by applying transforms to the input and filter, performing element-wise
multiplication, and then an inverse transform.
 Winograd convolution can be significantly faster than direct or FFT-based methods for certain
configurations.

2. FFT-based Convolution:
 Convolution in the spatial domain is equivalent to element-wise multiplication in the frequency
domain.
 This method leverages the Fast Fourier Transform (FFT) to convert the input and filter to the
frequency domain, multiply them, and then use the Inverse FFT (IFFT) to return to the spatial
domain.
 It is particularly efficient for large kernel sizes and inputs.

3. Im2col / Matrix Multiplication (GEMM):

 This widely used technique converts the convolution operation into a series of General Matrix
Multiplications (GEMM).
 The input image is "unrolled" into a matrix (im2col), and the filter is also reshaped, allowing
highly optimized BLAS libraries to perform the matrix multiplication.
 While efficient in leveraging hardware-optimized GEMM kernels, it can incur significant memory
overhead due to the im2col transformation.

4. Direct Convolution Optimization:

 Instead of transforming the convolution into other domains or matrix multiplications, direct
convolution algorithms aim to optimize the direct computation.
 These methods often focus on efficient memory access patterns and parallelization strategies
for specific hardware architectures (e.g., ARM CPUs).
 They can offer advantages in terms of memory overhead compared to im2col or transform-
based methods.

5. Separable Convolutions:
 If a multi-dimensional kernel can be decomposed into a series of one-dimensional kernels (e.g.,
a 2D kernel into two 1D kernels), the convolution can be performed more efficiently.
 This reduces the number of operations and parameters, especially for higher-dimensional
convolutions.
 Examples include spatial separable convolutions and depthwise separable convolutions.

6. Memory-Efficient Convolution (MEC):

 MEC aims to reduce the memory overhead associated with methods like im2col while still
leveraging efficient matrix multiplication.
 It uses a compact lowering scheme for the input matrix and executes multiple smaller matrix
multiplications in parallel.

The choice of the most efficient algorithm depends on factors such as kernel size, input
size, available hardware, and memory constraints. Often, deep learning frameworks employ a
combination of these techniques and dynamically select the most optimal algorithm for a given
layer.

Applications of Convolutional Neural Networks

Image Classification – Search Engines, Social Media , Recommender
Systems

The major use of convolutional neural networks is image recognition and

classification. It is also the only use case involving the most advanced frameworks

(especially, in the case of medical imaging).

The CNN picture categorization serves the following purposes:

 Deconstruct an image and find its distinguishing feature. The system employs a

supervised machine learning classification algorithm for this purpose.

 Reduces the description of its important credentials. It’s done with the help of an

unsupervised machine learning algorithm.

This method is used in the following fields:

Image tagging

The most basic type of image classification algorithm is image tagging. The image

tag is a term or a phrase that describes the images and makes them easier to find.

This method is used by big companies like Facebook, Google, and Amazon. It is also

one of the fundamental elements of visual search. Tagging involves recognition of

objects and even sentiment analysis of the image tone.

Visual Search

This method involves comparing an input image to the access database.

Furthermore, the visual search evaluates the image and searches for other photos

that have comparable credentials.

Recommender engines

Another field where image classification and object identification can be used is

recommender engines. Amazon, for example, employs CNN image recognition to

make suggestions in the “you might also like” area. The presumption is based on the

user’s expressed behavior. The products are matched based on visual criteria, such

as red shoes and red lipstick for a red outfit. Pinterest employs CNN image
recognition in a novel way. The organization focuses on visual credentials matching,

which results in simple visual matching enhanced by tagging.

Face Recognition RNN Applications include Social Media, Identification,

and Surveillance

Face recognition deserves its own section. This subset of image recognition deals

with more complex images. Such images could include human faces or other living

beings such as animals, fish, and insects.

The distinction between straight image recognition and face recognition is based on

operational complexity — the additional layer of work required.

 The shape of the face and its features are recognized first, followed by basic object

recognition.

 The features of the face are then examined further to determine its essential

credentials. For example, It could be the shape of the nose, the skin tone, and

texture, or the presence of scars, hair, or other surface irregularities.

 The sum of these credentials is then calculated into the image data perception of a

specific human being’s appearance. This procedure entails studying a large number

of samples that each present the subject in a different way. For instance, whether

with or without sunglasses).

 The input image is then compared to the database, and the system recognizes a

specific face.

Face recognition is used in social media platforms such as Facebook for both social

networking and entertainment.

  Face recognition in social networking serves to streamline the often dubious process

of tagging people in photos. This feature is especially useful when you need to tag

through hundreds of images from a conference or when there are far too many faces

to tag. So, if you’re planning to build a social network, keep this feature in mind.

 Facial detection in entertainment lays the groundwork for further transformations and

manipulations. The most notable examples are Facebook Messenger filters and

Snap chat Looksery filters. The filters depart from the face’s auto-generated basic

layout and add new elements or effects.

Facial recognition technology is gaining traction as a viable method of personal

identification.

Face recognition cannot be used to verify a persona in the same way that

fingerprints and legal documents can. In cases where there is limited information,

face recognition can be useful in identifying the person. For instance, from

surveillance camera footage or a covert video recording.

Medical Image Computing – Predictive Analytics, Healthcare Data Science

Healthcare is the industry where all of the cutting-edge technology is put to the test.

If you want to test the usefulness of a certain technology, try employing it in a

healthcare setting. Image recognition is no exception.

The most fascinating image recognition CNN use case is medical image computing.

The medical image includes a whole lot of further data analysis that arises from

initial image recognition.

CNN medical image classification detects anomalies in X-ray and MRI images with

better accuracy than the human eye.

These systems can display the series of photos as well as the differences between

them. This feature lays the groundwork for future predictive analytics.

Medical image classification is based on massive datasets such as Public Health

Records. It serves as a training basis for the algorithms and patients’ confidential

data and test results. They work together to create an analytical platform that

monitors the current status of the patient and forecasts results.

Health Risk Assessment Using Predictive Analytics

In healthcare, saving lives is a top priority. And it is always advantageous to have the

ability to predict the future. Because when it comes to patient care, you must be

prepared for anything. The health risk assessment is an excellent demonstration.

Convolutional Neural Network Predictive Analytics is used in this field.

Working of CNN Health Risk Assessment:

 CNN uses a grid topology approach to process data, which is a set of spatial

correlations between data points. The grid is two-dimensional in the case of images.

The grid is one-dimensional in the case of time series textual data.

 The convolution algorithm is then used to identify some aspects of the input.

 Take into account the variations of input.

 Determine variable interactions that are sparse.

 Use the same settings for all of a model’s functions.

Health Risk Assessment applications

 HRA is a predictive application that computes the likelihood of specific events. Based

on patient data, this use case includes disease progression or complications. It looks

for similar PHR, analyses the patient’s data, looks for patterns, and calculates

potential outcomes. This system can be used for routine health checks.

 The framework can be expanded by including a treatment plan. In this case, the

prediction determines the best way to treat the symptoms.

 The HRA system can also be used to investigate the specific environment and

identify potential hazards for those who work there. This method is used to as sess

dangerous situations. In Australia, for example, officials are studying sun activity to

determine the level of radiation threat.

Drug Discovery Using Predictive Analytics

Another major healthcare field that makes extensive use of CNNs is drug discover y.

It is also one of the most inventive uses of convolutional neural networks in general.

RNN (Recurrent Neural Network) and stock market prediction are examples of pure

data tweaking, whereas drug discovery and CNN are not.

The problem is that drug discovery and development is a time-consuming and costly

process. In drug discovery, scalability and cost-effectiveness are critical.

The process of developing new drugs lends itself well to the implementation of

neural networks. During the development of a new drug, there is a large amount of

data to consider.
The following stages are involved in the drug discovery process:

 This is a clustering and classification problem involving the analysis of observed

medical effects.

 Machine learning anomaly detection may be useful in hit discovery. The algorithm

searches the compound database for new activities that can be used for specific

purposes.

 Then, using the Hit to Lead process, the results are narrowed down to the most

relevant. That’s what dimensionality reduction and regression are all about.

 Then there’s Lead Optimization, which is the process of combining and testing lead

compounds to find the best approaches to them. The stages involve the examination

of the organism’s chemical and physical effects.

Following that, the development shifts to live testing. Machine learning algorithms

were relegated to the background and were used to structure incoming data.

CNN optimizes and streamlines the drug discovery process at critical stages. It

allows for a reduction in the time required to develop cures for emerging diseases.

Precision Medicine Using Predictive Analytics

A similar approach can be used with existing drugs when developing a treatment

plan for patients. Precision medicine aims to find the most effective way to treat a

disease.

Supply chain management, predictive analytics, and user modeling are all part of

precision medicine.

This is how it works:

 From the standpoint of data, the patient is a collection of states that ar e affected by a

variety of factors (symptoms and treatments).

 The addition of variables (treatment types) has specific effects in both the short and

long term.

 Each variable has its own set of statistics regarding its impact on a symptom.

 Data is combined to form an assumption about the best course of action based on

the available information.

 The various outcomes and changes in the patient’s condition are then considered.

This is how the assumption is validated. This stage is handled by recurrent neural

networks because it necessitates the analysis of data point sequences.