LEARNING

MS. ROOPA
ASST.PROFESSOR
JSSCACS
MYSURU-570025
LEARNING IN AI
 In Artificial Intelligence (AI), "learning" refers to the process by which a system (often a machine or computer
program) improves its performance over time by gaining experience or data.
 Learning in AI is the ability of a machine to improve its performance on a task through experience (data), without
being explicitly programmed for every scenario.
 Learning refers to the process by which a system acquires knowledge or improves its performance based on
experience.
 The idea behind learning is that precepts should be used not only for acting, but also for improving the agent’s
ability to act in the future. Learning takes place as the agent observes its interactions with the world and its
own decision-making processes.
 Learning can range from trivial memorization of experience.
 Learning refers to the process by which a system acquires knowledge or improves its performance based on
experience. Machine learning researchers have come up with a large variety of learning elements.
Learning in AI means teaching a system (or agent) to improve its performance by gaining knowledge through experience.
Instead of programming every action manually, the system learns from past actions and feedback to do better in the future.
Why is Learning Important?
•It helps the agent act better over time.
•The agent learns from what it sees and does in the world.
•Learning can be as simple as remembering things or as complex as making better decisions.
Types of Learning Elements
Machine learning researchers have created many different parts (elements) that help an agent learn. To design a good learning
system, we need to think about:
1.What should be learned?
→ Which parts of the agent's brain or system do we want to improve?
2.How does the agent get feedback?
→ What information does the agent receive to know if it’s doing well or not?
3.How is the information shown or stored?
→ How is learning material (data, knowledge) represented?
Components of a Learning Agent
These are the parts an agent might learn:
1.Rules to take actions based on the current situation.
→ Like: "If it's raining, take an umbrella."
2.Ways to understand the world from what it senses (perceives).
→ Understanding the situation from inputs.
3.Info about changes in the world and what might happen after actions.
→ Predicting future results.
4.How good or bad a situation is (utility).
→ Like scoring different situations.
5.How good or bad an action is.
→ Choosing the best move.
6.Goals to achieve.
→ What the agent is trying to do (like winning a game or solving a task).
Characteristics of Learning:
1.Learning is based on experience
1. The system improves its performance by learning from past actions or data.
2. Example: A robot gets better at walking by practicing over time.
2.Learning improves performance
1. The goal of learning is to make the system perform better at a specific task.
2. Example: A self-driving car learns to drive more safely.
3.Learning is adaptive
1. A good learning system can adjust to changes in the environment.
2. Example: A chatbot can adapt its replies based on the user's tone.
4.Learning involves generalization
1. The system should not only remember past data but also apply it to new situations.
2. Example: After seeing a few types of chairs, the system should recognize a new chair.
5.Learning can be supervised or unsupervised
1. Supervised: Learns from labeled data (with answers).
2. Unsupervised: Finds patterns in data without labels.
6.Learning requires feedback
1. Feedback (like success, failure, rewards, or corrections) helps the system improve over time.
7.Learning reduces the need for programming
1. Instead of writing code for every situation, the agent learns on its own.
Advantages of Learning
1.Improves Performance Over Time
1. The system gets better with more experience.
2. Example: A speech recognition app improves with more usage.
2.Reduces Manual Programming
1. No need to hard-code every rule or behavior.
2. The system learns behaviors automatically.
3.Adapts to New Situations
1. Learning enables the system to handle changes in the environment.
2. Example: A spam filter updates itself as new spam types appear.
4.Discovers Hidden Patterns
1. It can find patterns in large data that humans may miss.
2. Useful in data analysis and predictions.
5.Enables Personalization
1. Systems can learn user preferences and personalize experiences.
2. Example: Netflix recommendations.
Disadvantages of Learning
1.Requires a Lot of Data
1. Learning systems often need huge amounts of data to perform well.
2. Without enough data, learning is inaccurate.
2.High Computational Cost
1. Training models can be time-consuming and resource-heavy.
3.Possibility of Overfitting
1. The system may memorize instead of learning to generalize, leading
to poor performance on new data.
4.Quality of Learning Depends on Data Quality
1. Bad or biased data can lead to wrong learning outcomes.
5.Unpredictable Behavior
1. In some cases, a learning system might make unexpected or
incorrect decisions.
 Forms Of Learning:
Forms of Learning refer to the different ways individuals acquire knowledge, skills, behaviors, or attitudes through
experience, practice, or observation.

Form of Learning Example Goal Type of Label Used

Supervised Learning Email spam detection Predict a label (spam or not spam) Labeled data (input-output pairs)

Customer Group similar items without

Unsupervised Learning No labels
segmentation predefined labels
 Document Improve accuracy using few
Semi-Supervised
classification with few labeled + many unlabeled Few labeled + many unlabeled
Learning
labeled documents examples
Reinforcement Game-playing AI (e.g., Learn actions to maximize
Reward signal (not direct labels)
Learning AlphaGo) reward over time
Predict next word in a
Self-Supervised Learn useful representations Labels generated from data
sentence (e.g., GPT
Learning from raw data itself
pre-training)
Medical diagnosis
Multi-Instance Classify a bag of instances (e.g., Label on group, not individual
based on multiple
Learning all scans per patient) instances
scans
Stock market Continuously learn from
Online Learning Labeled data in sequence
prediction streaming data
Image classification
Selectively query labels for
Active Learning with human in the Few labeled + selected labels
most informative data
Supervised Learning:
Supervised learning, also known as supervised machine learning, is a type of machine learning that trains the model
using labeled datasets to predict outcomes. A Labeled dataset is one that consists of input data (features) along with
corresponding output data (targets).
✅ 1. Labeled Data

 Definition: Training data includes both inputs and correct outputs (labels).

 Example: An image labeled as “cat” or “dog.”

🎯 2. Goal-Oriented Learning

 Purpose: Learn a mapping function from inputs to outputs to make accurate predictions.

 Use Cases: Spam detection, image classification, sales forecasting.

🔄 3. Training Process

 The model is trained using known data so it can generalize to new, unseen data.

 The algorithm iteratively adjusts to minimize error between predicted and actual outputs.

📊 4. Evaluation is Straightforward

 Since you have labels, you can directly compare predictions to actual outcomes using metrics like accuracy, precision, recall, or MSE.
🧠 5. Types of Problems
 Classification: Output is categorical (e.g., yes/no, red/blue/green)
 Regression: Output is continuous (e.g., house prices, temperature)
🧪 6. Requires Data Splitting
 Typically split into:
o Training set – to train the model
o Test/Validation set – to evaluate generalization
Types of Supervised Learning Algorithm
1. Classification:The task is to predict a discrete label or category.
Categorical (discrete values).
Ex: Disease diagnosis (e.g., cancer vs. no cancer)
2. Regression:The task is to predict a continuous output.
Continuous (numerical values).
Ex: Predicting temperature.
Type Output Examples Algorithms
Email spam detection, Logistic Regression,
Classification Categorical (discrete) Handwritten digit Decision Trees, SVM,
recognition Random Forest
Linear Regression, Ridge
House price prediction,
Regression Continuous (numerical) Regression, SVR, Neural
Stock price forecasting
Networks
Classification Regression
 Logistic Regression  Linear Regression
 Decision Trees  Polynomial Regression
 Random Forest  Ridge/Lasso Regression
 Support Vector Machines (SVM)  Support Vector Regression (SVR)
 K-Nearest Neighbors (KNN)  Decision Trees (Regression Trees)
 Naive Bayes  Neural Networks (for more complex tasks)
 Neural Networks (for complex tasks)
Unsupervised learning: Is a type of machine learning where the algorithm is given data without explicit labels (i.e., no "correct
answers"). The goal is to find hidden patterns or structures in the data.
Unlike supervised learning, where we have labeled data to guide the learning process, unsupervised learning algorithms must
infer the structure of the data on their own.
Type Description Examples Algorithms

Group similar data points Customer segmentation, Image K-Means, DBSCAN, Hierarchical
Clustering
together segmentation Clustering, GMM

Reinforcement Learning (RL) :Is a type of machine learning where an agent learns to make decisions by interacting
with an environment.
 The agent's goal is to maximize some notion of cumulative reward through trial and error.
 In RL, the agent takes actions in the environment, receives feedback in the form of rewards or penalties, and uses
that feedback to improve future decisions
 Reinforcement learning has been successfully applied to various fields like robotics, gaming (e.g., AlphaGo),
autonomous vehicles, and recommendation systems.
Example:Robotics ,finance, Game Playing
Self-supervised learning is a deep learning methodology where a model is pre-trained using unlabelled data and
the data labels are generated automatically, which are further used in subsequent iterations as ground truths.
 Machine Learning (ML) is a branch of Artificial Intelligence (AI) that works on algorithm developments and statistical
models that allow computers to learn from data and make predictions or decisions without being explicitly programmed.
How does Machine Learning Work?
Machine Learning process includes Project Setup, Data Preparation, Modeling and Deployment. The following figure
demonstrates the common working process of Machine Learning. It follows some set of steps to do the task; a sequential
process of its workflow is as follows:
Characteristics of Machine learning:

 Learns from Data: Machine learning systems improve by analyzing data instead of being programmed with specific

instructions.

 Improves Over Time: The more data it gets, the better the system becomes at making predictions or decisions.

 Makes Predictions: It can predict future outcomes based on patterns found in past data.

 Works Automatically: Once trained, machine learning models can work without needing constant human input.
A Decision Tree in AI is a tree-like model used for decision-making, where each internal node represents a "decision" or a

feature, each branch represents an outcome of that decision, and each leaf node represents a final decision or classification.

 Simple Structure: It’s easy to visualize and understand as it breaks down decisions into a sequence of questions.

 Recursive Partitioning: The tree is built by splitting data based on feature values, aiming to reduce uncertainty or entropy.

 Works for Classification & Regression: Can be used for both classification (categorizing data) and regression (predicting

continuous values).

 Easy to Interpret: Decision trees are transparent, so it’s easy to understand why a decision was made.
Example: Loan Approval
Imagine a bank wants to decide whether to approve a loan for a customer. The decision tree could look at factors such as:
1.Income Level: Is the income above a certain threshold?
1. Yes: Go to the next question.
2. No: Reject the loan.
2.Credit Score: Is the credit score above 650?
1. Yes: Go to the next question.
2. No: Reject the loan.
3.Employment Status: Is the person employed full-time?
1. Yes: Approve the loan.
2. No: Reject the loan.
Each node in the decision tree represents a decision based on specific criteria (like income, credit score, or employment status),
and by following the branches, the tree leads to a decision (approve or reject the loan).
 [Income Level]
/ \
> 50k <= 50k
/ \
[Credit Score] Reject
/ \
> 650 <= 650
/ \
[Employment Status] Reject
/ \
Full-time Part-time
/ \
Approve Reject