Machine Learning
Welcome, everyone, to our introductory session on
Machine Learning. We'll aim to demystify Machine Learning, illustrate it with
simple examples, and delve into some real-world applications. Remember, our
goal is to understand the foundational concepts, not to become experts
overnight. So, let's get started.
What is Machine Learning?
Machine learning is a branch of artificial
intelligence that enables computers to learn and make decisions without being
explicitly programmed & how we make computers to learn – with the help of
data.
It's like teaching an infant to recognize shapes or
fruits or color ultimately data points.
You show the infant various objects and tell them,
"This is a square," or "That's a circle." Over time, the
infant begins to identify these shapes or colors . Similarly, in Machine
Learning, we train computers with data to predict and make decisions.
What problem is machine learning solving ? How things
were happening prior to ML or Benefits of ML
Before machine learning became a significant part of
our technological tools, many tasks were completed manually, using traditional
programming methods. Here are some examples which will help us to understand
the importance of ML or will give us a perspective let’s see them one by one
Data Analysis: Before machine learning,
data analysis was largely a manual process. Analysts would comb through data
sets manually and use basic statistical techniques to uncover insights. This
was a time-consuming process and often only scratched the surface of what the
data could reveal. With machine learning, we can automatically process large
datasets and uncover deeper insights. Machine learning algorithms can spot complex
patterns and trends that humans might miss, helping businesses make more
data-driven decisions.
Recommendation Systems: Before the advent of
machine learning, recommendation systems were rudimentary. For example, an
online store might recommend products based on their popularity or their
relation to a user's previous purchase. Today, with machine learning, these
systems can analyze a user's behavior and preferences in real time, providing
personalized recommendations that increase engagement and sales.
Email Filtering: In the past, spam filters
were rule-based. They used a set of manually created rules to identify spam,
like looking for certain keywords. This approach was not always accurate and
couldn't adapt to new spam tactics. Machine learning changed this by allowing
systems to learn from a large number of examples, improving the accuracy of
spam detection.
Fraud Detection: Fraud detection used to be
a reactive process that happened after the fraudulent activity had taken place,
and it heavily relied on manual investigation. With machine learning,
predictive models can spot unusual patterns in transaction data, allowing
potential fraud to be detected and prevented in real time.
Speech Recognition: Early speech recognition
systems were based on hard-coded rules and were often inaccurate. Machine
learning, particularly deep learning, has greatly improved the accuracy of
speech recognition systems, leading to virtual assistants like Siri, Alexa, and
Google Assistant, which can understand a wide range of natural language
commands.
It enables you to do things that cant be possible
manually will discuss all the examples in latter slides but for now it empowers
you and you see that in your everyday life like spam filtration or fraud
detection.
Machine Learning can be broadly classified into three
types: Supervised Learning, Unsupervised Learning, and Reinforcement Learning.
Let's explore each of these with simple examples:
Supervised Learning: This is like learning
under the guidance of a teacher. In supervised learning, we provide the machine
with labeled input data and the corresponding correct output. The machine
learns the relationship between the input and output during the training
process and uses this learned relationship to predict the output when new input
data is given. It's called "supervised" because the model is learning
under the guidance of the training dataset (similar to a student learning under
a teacher's supervision).
Example: Consider an email spam filter. We could
train a supervised learning model by providing it with many example emails
along with labels indicating whether each email is "spam" or
"not spam". After learning from these examples, the model can then
predict whether a new email is spam.
Unsupervised Learning: In unsupervised learning,
the machine is provided with unlabeled input data. The machine's task is to
learn the underlying structure of the data on its own. In other words, we're
not telling the model what to look for; the model must discover interesting
patterns in the data by itself.
Example: An example of unsupervised learning is
customer segmentation in marketing. Here, the goal might be to divide a
customer base into groups that exhibit similar purchasing behaviors. We don't
tell the model how to separate the customers; it figures out on its own how to
group customers.
Reinforcement Learning: Reinforcement Learning is
about interaction and exploration. The model (often called an
"agent") learns by interacting with its environment, receiving
rewards for correct actions and penalties for incorrect ones. Over time, the
agent learns to make decisions that maximize its total reward.
Example: Reinforcement learning is often used in
training game-playing AI. For example, in a chess game AI, the model explores different
moves and sequences of moves, receiving a reward when it wins a game and a
penalty when it loses. Over time, the model learns to make the moves that are
more likely to lead to winning the game.
These are high-level descriptions and examples. Each
of these types of machine learning can be further divided into subtypes, and
there are also other types of machine learning that combine elements of these
three. The specific type of machine learning that's best for a particular task
depends on the nature of the problem and the available data.
Let's break down the process of machine learning
into its main steps, using an example of a Supervised Learning task where we
aim to predict whether an email is spam or not:
Collecting Data: This is the first step in
the machine learning process, where you gather data relevant to the problem
you're trying to solve. In our spam detection example, the data might be a
collection of emails, each labeled as either "spam" or "not
spam".
Preprocessing Data: Real-world data is often
messy and incomplete. Preprocessing includes cleaning the data (handling
missing data, removing duplicates, etc.), converting categorical data to
numeric data (e.g., "spam" could be 1 and "not spam" could
be 0), normalizing data, and possibly extracting features (like the subject
line, the email's text, or the sender's email address).
Splitting the Data: We usually split our data
into a training set and a test set. The training set is used to train our
machine learning model, and the test set is used to evaluate its performance on
unseen data. A common split might be 80% of the data for training and 20% for
testing.
Selecting a Model: There are many different
types of machine learning models (like decision trees, support vector machines,
neural networks, etc.), and the choice depends on the problem and the data. For
spam detection, you might start with a model like Naive Bayes, which is often
used for text classification tasks.
Training the Model: During training, the model
learns from the training data. It tries to find patterns in the input features
that are related to the target variable (whether the email is spam or not). The
specifics of this process depend on the type of model.
Evaluating the Model: Once the model has been
trained, it's important to test its performance on unseen data to ensure that
it hasn't just memorized the training data (a problem called
"overfitting"). We use the test set (which the model hasn't seen
during training) for this. In our example, we might measure the percentage of
emails the model correctly identifies as spam or not spam.
Tuning Model Parameters: Most machine learning
models have parameters that can be tuned to improve performance. For example, a
neural network has a parameter called the "learning rate" that
determines how quickly it adjusts its internal calculations during training.
Tuning these parameters can lead to better performance, but it can also be a
time-consuming process.
Making Predictions: Once we're satisfied with
our model's performance, we can use it to make predictions on new, unseen data.
In our example, we could now feed new emails into our model, and it would
predict whether each one is likely to be spam or not.
Remember, these steps are a general outline of the
process, and the specifics can vary depending on the problem, the data, and the
type of machine learning being used. But overall, this process forms the
backbone of many machine learning tasks.
