• Deep Learning (DL) :

• As mentioned above, it is inspired by the neural architecture in the

human brain which processes and gets insights by passing data
through different layers. DL demands huge data as it has to train
multiple parameters in each layer.
• Deep Learning was dated back to the 1940s with the invention of a
neural network-based model called the “Perceptron”.
• Then why the buzz around it started only in the past few years?
Let’s look at a few reasons behind the buzz.
• Data Growth
• We are living in a digital world where everything is getting digitized.
• You might have observed this shift, where every business is integrating software solutions to ease the
process. Even government policies and schemes are getting launched and implemented online.
• We are witnessing digital payments boom and what about content that is adding up on social media
platforms like Facebook, Insta, Twitter, Whatsapp, etc?
• With approx. 63% of the world's total population having access to the internet and the volume of data
that is getting generated every day is mind-boggling.
• Hardware Advancements
• Back in the 2000’s normal laptops used to come with a RAM capacity of 500MB or so and it used to take
minutes to run even small programs.
• How about now? Everyone is using 16GB RAM and more, which can run complex programs in seconds.
• Also, we are using specialized hardware like GPUs and TPUs. These come up with large processing
power that helps in running parallel jobs and reduces execution time. Nowadays laptops are coming up
with built-in GPUs (Graphic cards).
• Google TPUs (Tensor processing units) are designed specifically to handle neural network loads.
• All these advancements in hardware are helping deep learning models to run complex jobs within less
time.
• Python and Open Source Ecosystem
• Back then we used to do programming using complicated languages like C++, Java, and similar ones which demands high knowledge of
software systems.
• Coming to Python, is one of the most accessible and interpretable languages ever with simple syntax. Anyone can learn it in no time and
start building applications, unlike other languages.
• Even people without computer science backgrounds like mathematicians, statisticians, etc found it simple to learn python and
implement deep learning algorithms with less effort in learning to code.
• These days we are also witnessing the trend of people moving from non-CS backgrounds to the data science domain, thanks to Python
for making the transition easier.
• And the other big thing is the open source ecosystem. Mostly used deep learning frameworks like Tensorflow (by Google), PyTorch (by
Facebook), etc are all open-sourced. Anyone can install the libraries and start using them for free. Isn't it cool?
• Cloud and AI Boom
• Have you heard of something like Azure ML, AWS Sagemaker, and all? These are ML studios in the cloud, even anyone without prior
experience in running ML/DL algorithms can use these platforms to build models and generate some predictions.
• Entry barriers to start experimentation with deep learning techniques are getting removed with the advent of these kinds of cloud
platforms.
• How many of you are using Google Colab notebooks with GPU instances? As running deep learning models becomes speedy and
simple with advanced hardware, even if you don't have that configured in your machine you can simply run them on cloud platforms.
• Also with these platforms like Google colab in place, no need to worry about software installations to make your model up and running.
• We are in an AI Boom, Data is the new fuel and all the companies want to harness some insights from the huge volume of data that is
getting created to better understand their customer's pulse and to stand ahead in the competition.
• Google has announced that it is moving from Mobile first approach to the AI-first approach which itself speaks about AI growth. This
kind of move from one of the tech giants further accelerates the growth of deep learning techniques.
• Deep learning is an artificial intelligence (AI) methodthat teaches
computers to process data in a way inspired by the human brain.
Deep learning models can recognize complex pictures, text,
sounds, and other data patterns to produce accurate insights and
predictions. You can use deep learning methods to automate
tasks that typically require human intelligence, such as describing
images or transcribing a sound file into text.
• Why is deep learning important?
• Deep learning technology drives many artificial intelligence
applications used in everyday products, such as the following:
• Chatbots and code generators
• Digital assistants
• Voice-activated television remotes
• Fraud detection
• Automatic facial recognition
• It is also a critical component of technologies like self-driving cars,
virtual reality, and more. Businesses use deep learning models to
analyze data and make predictions in various applications.
• What are deep learning use cases?
• Deep learning has several use cases in automotive, aerospace, manufacturing, electronics, medical research, and other
fields.
• Self-driving cars use deep learning models for object detection.
• Defense systems use deep learning to flag areas of interest in satellite images.
• Medical image analysis uses deep learning to detect cancer cells for medical diagnosis.
• Factories use deep learning applications to detect when people or objects are within an unsafe distance of machines.
• These various use cases of deep learning can be grouped into five broad categories: computer vision, speech recognition,
natural language processing (NLP), recommendation engines, and generative AI.
• Computer vision
• Computer vision automatically extracts information and insights from images and videos. Deep learning techniques to
comprehend images in the same way that humans do. Computer vision has several applications, such as the following:
• Content moderation to automatically remove unsafe or inappropriate content from image and video archives
• Facial recognition to identify faces and recognize attributes like open eyes, glasses, and facial hair
• Image classification to identify brand logos, clothing, safety gear, and other image details
• Speech recognition
• Deep learning models can analyze human speech despite varying speech patterns, pitch, tone, language, and accent. Virtual assistants such as Amazon Alexa, text-to-speech, and speech-to-text
software use speech recognition to do the following tasks:
• Assist call center agents and automatically classify calls.
• Convert clinical conversations into documentation in real-time.
• Accurately subtitle videos and meeting recordings for a wider content reach.
• Convert scripts to prompts for intelligent voice assistance.
• Natural language processing
• Computers use deep learning algorithms to gather insights and meaning from text data and documents. This ability to process natural, human-created text has several use cases, including:
• Automated virtual agents and chatbots
• Automatic summarization of documents or news articles
• Business intelligence analysis of long-form documents, such as emails and forms
• Indexing of key phrases that indicate sentiment, such as positive and negative comments on social media
• Recommendation engines
• Applications can use deep learning methods to track user activity and develop personalized recommendations. They can analyze users' behavior and help them discover new products or services. For
example,
• Recommend personalized videos and content.
• Recommend customized products and services.
• Filter search results to highlight relevant content based on user location and behavior
• Generative AI
• Generative AI applications can create new content and communicate
with end users more sophisticatedly. They can assist in automating
complex workflows, brainstorming ideas, and intelligent knowledge
searches. For example, with generative AI tools like Amazon Q
Business and Amazon Q Developer, users can
• Ask natural language questions and get summarized answers from
multiple internal knowledge sources.
• Get code suggestions and automatic code scanning and upgrades.
• Create new documents, emails, and other marketing content faster.
• Machine learning
• Deep learning is a subset of machine learning. Deep learning algorithms emerged to make traditional machine learning techniques more efficient. Traditional machine learning methods require significant
human effort to train the software. For example, in animal image recognition, you need to do the following:
• Manually label hundreds of thousands of animal images.
• Make the machine learning algorithms process those images.
• Test those algorithms on a set of unknown images.
• Identify why some results are inaccurate.
• Improve the dataset by labeling new images to improve result accuracy.
• This process is called supervised learning. In supervised learning, result accuracy improves only with a broad and sufficiently varied dataset. For instance, the algorithm might accurately identify black cats
but not white cats because the training dataset had more images of black cats. In that case, you would need more labeled data of white cat images to train the machine learning models again.
• Benefits of deep learning over machine learning
• A deep learning network has the following benefits over traditional machine learning.
• Efficient processing of unstructured data
• Machine learning methods find unstructured data, such as text documents, challenging to process because the training dataset can have infinite variations. On the other hand, deep learning models can
comprehend unstructured data and make general observations without manual feature extraction. For instance, a neural network can recognize that these two different input sentences have the same
meaning:
• Can you tell me how to make the payment?
• How do I transfer money?
• Hidden relationships and pattern discovery
• A deep learning application can analyze large amounts of data more deeply and reveal new insights for which it might not have been trained. For example, consider a deep learning model trained to analyze
consumer purchases. The model has data only for the items you have already purchased. However, the artificial neural network can suggest new items you haven't bought by comparing your buying patterns
to those of similar customers.
• Unsupervised learning
• Deep learning models can learn and improve over time based on user
behavior. They do not require large variations of labeled datasets. For
example, consider a neural network that automatically corrects or
suggests words by analyzing your typing behavior. Let's assume it was
trained in English and can spell-check English words. However, if you
frequently type non-English words, such as danke, the neural network
automatically learns and autocorrects these words too.
• Volatile data processing
• Volatile datasets have large variations. One example is loan repayment
amounts in a bank. A deep learning neural network can categorize and
sort that data by analyzing financial transactions and flagging some for
fraud detection.
• Generative AI
• Generative AI took the neural networks of machine learning and
deep learning to the next level. While machine learning and deep
learning focus on prediction and pattern recognition, generative AI
produces unique outputs based on the patterns it detects.
Generative AI technology is built on transformer architecture that
combines several different neural networks to combine data
patterns in unique ways. Deep learning networks first convert text,
images, and other data into mathematical abstractions and then
reconvert them into meaningful new patterns.
• Snapchat Filters: Convolutional Neural Networks (allegedly)
• Netflix Recommendation System: Restricted Boltzmann Machines
• Google Translate (Machine Translation): Recurrent Neural Networks
• Siri (Personal Assistants): Hidden Markov Models (2011-2014), Long Short
Term Memory Networks (2014+), which is a type of Recurrent Neural
Networks.
• Self-driving cars (Image recognition, Video recognition): Convolutional
Neural Networks (among other things)
• Speech recognition, Hand-writing recognition: Recurrent Neural Networks
• Market Segmentation: k-means clustering
• Google AlphaGo: Convolutional Neural Networks
• Image generation: Generative Adversarial Networks (GANs)
• Most common real world applications are in:
• Customer experience - think chatbots and self-service solutions.
• Self-driving cars - recognizing pedestrians and street signs are
done with deep learning, for example.
• Entertainment - Spotify or Netflix suggestions are done with deep
learning.
• Healthcare - research and medical diagnoses are the main
benefactors, given how deep learning models eat a lot of data and
churn out results much faster than human brains can.
• Hello Everyone, here I am going to discuss how to solve a real time Machine Learning Project
end- to -end.
• Understanding the Real World / Business problem. E.g. - Let’s assume you are a ML
engineer at Flipkart and you want to build a Flipkart Product Recommendation System. So
firstly we have to understand that Flipkart is an website and there are similar products
recommended for people. So what is the problem, how is the data collected ; first
understanding the real world business problem that needs to be solved is very important. In
first part we explain what the problem we are trying to solve .
• Real World/ Business objectives and constraints. For e.g., in the Flipkart’s Product
recommendation system, the business objective could be increasing the sales of items. But
the business constraint here is like suppose if somebody goes to product page you should be
able to give the similar products very very fast because if the page loads slowly the
customers will be dissatisfied, our model should be able to decide the related products very
very fast. This is also called a low- latency system, low latency is the business constraint
here. So from this point we can say that the business objectives and business constraints
help us understand the problem better and model it better using Machine Learning.
• Understating the data available to us - We have to understand the exact data that we have
at our disposal.
• Mapping the real world problem to an ML problem. a) Type of ML Problem- After
understanding the Business problem , we have to then understand what type of ML problem
is this, is it a Classification problem, is it a Regression problem, is it a Recommender System
problem, is it a Clustering problem , or is it a Deep Learning problem, etc. ??!! b) KPI’s and
performance metrices - Next comes, given the real world objective and constraints , what is
the Key Performance Indicator, how do we measure our model !! For e.g. , in the Flipkart
Product Recommendation System problem, your KPI could be how many times have people
clicked and purchased. So you have to come up with a performance metric that correlates
well or correlates very very closely as closely as possible with the business objective. c) Train
and Test Dataset construction - Given our big dataset, how do we break it up into Training
dataset and Test Dataset so that we can build our model on top that should be evaluated. So
these three points mentioned in this section goes from a real world problem to a ML
problem.
• Exploratory Data Analysis (EDA) - a) Feature Engineering (FE), b) Feature
Transformation(FT), c) Understating the data, d) Also documenting failed case of FE and FT
because it’s not enough to document or write about things that work, it’s also important to
document FE and FT steps that did not work.
• Building the simplest model/ Building a fairly quick to build
model- In this stage we are making the simplest model and when
we choose a baseline model we clearly articulate why we are
picking this model and we will implement it and will get results for
it. This model will not be our final complex model and for all the
model where it has failed we will do exploratory data analysis on
the failure cases to understand why this algorithm has failed. We
will use different Data visualization techniques to understand why
this model hasn't worked well because this gives us a much
clearer idea on what other types of models we can employ.
• Building more complex Models- Now once we have the Baseline model
then we will go through more complex models in increasing order of
complexity. For each model again we will repeat the same steps as Step
Number 6 ,i.e., we will discuss why did we pick the models, we will discuss
how to implement those models and how we get the results. we will do Data
Analysis on the failure cases and we understand why a model is failing in
some instances. So we go in increasing order of complexity from baseline
model. We try multiple models, we never solve the problem using only one
model, we typically try and do at least two to three models. At the end of
trying all these models we know which model performs best for our problem.
• Productionizing the model- After getting the best model for our problem the
last stage is to productionize the model for the real world implementation
and measuring the actual real world impact of our models
• Deep Learning: What Makes It Different?
• Traditional ML algorithms such as linear or logistic regression, random
forest, or gradient boosting use underlying statistical techniques that
enable machines to iteratively learn from labeled training data how to
perform specific prediction or clustering tasks. This approach typically
requires data scientists or domain experts to manually extract,
engineer, and select which features (variables) to use in the model,
which can be a time-consuming stage of any ML project.
• Deep learning is a specialized, more modern subset of ML that uses
neural network architectures with many layers — often referred to as
“deep neural networks” — to model complex patterns in data.
Common types of deep neural networks include convolutional neural
networks (CNNs), recurrent neural networks (RNNs), and transformer
networks.
• In contrast with traditional ML models, deep learning models can
automatically learn to extract features directly from raw data,
reducing the need for manual feature engineering and making
them very powerful for applications involving unstructured data
like text, images, video, or audio. Although deep learning models
excel in tasks such as image and speech recognition, natural
language processing (NLP), and complex game playing where they
can outperform traditional ML methods, these models are more
complex and computationally intensive, typically requiring large
amounts of training data and specialized hardware like GPUs.
• Deep Learning vs ML
• With deep learning on the scene, we don’t throw classic ML models out
the window. For many classical prediction tasks that involve only
structured, tabular data, ML approaches still are more suitable than
deep learning because they are simpler and less computationally
intensive.
• For example, here’s a simple illustration: If I want to go from Manhattan
to Brooklyn, should I walk or fly in a plane? It comes down to a simple
tradeoff between cost and time spent versus the desired outcome.
Flying from Manhattan to Brooklyn is the equivalent of using a deep
learning model for a “simple” problem. In other words, for this
particular problem, it’s not efficient at all!