D.Y.
Patil College of Engineering, Akurdi, Pune-44
Department of Computer Engineering
Unit 4: Introduction to Machine Learning and Case Studies on
Advanced Machine Learning Techniques
Artificial Intelligence (AI), Machine Learning (ML), and Deep Learning (DL):
Artificial Intelligence (AI): AI refers to the simulation of human intelligence in machines
programmed to think, reason, and learn. It encompasses various techniques and systems
that enable machines to perform tasks requiring human-like cognition, such as problem-
solving, decision-making, and language understanding. AI can be categorized into:
o Narrow AI: Focused on specific tasks (e.g., voice assistants like Siri).
o General AI: Hypothetical systems capable of performing any intellectual task that
humans can do.
o Superintelligence: A level of AI that surpasses human intelligence (currently
theoretical).
Machine Learning (ML): ML is a subset of AI that focuses on enabling machines to learn
patterns and make predictions or decisions without being explicitly programmed. ML
systems improve over time as they are exposed to more data. ML applications include:
o Predictive analytics (e.g., forecasting sales).
o Automation of repetitive tasks (e.g., data entry).
o Recommendation systems (e.g., Netflix, Amazon).
Deep Learning (DL): DL is a subset of ML based on neural networks with many layers
(hence "deep"). It excels at processing large datasets and solving complex tasks such as
image recognition, natural language processing (NLP), and autonomous driving. Key
components of DL include:
o Artificial Neural Networks (ANNs): The backbone of DL.
o Training on large datasets with GPUs or TPUs for computational efficiency.
o Applications in fields like healthcare (diagnostics) and finance (fraud detection).
Types of Learning:
1. Supervised Learning:
o Definition: Involves training a model on labeled data, where input-output pairs are
provided.
o Goal: Learn a mapping function from inputs to outputs.
o Techniques:
Classification: Assigning categories to data points (e.g., spam detection).
Regression: Predicting continuous values (e.g., stock price prediction).
o Examples:
Image classification (e.g., identifying cats vs. dogs).
Predicting house prices based on features like size and location.
2. Unsupervised Learning:
o Definition: Works on unlabeled data, discovering patterns or structures.
o Goal: Group or simplify data.
o Techniques:
Clustering: Grouping similar data points (e.g., customer segmentation).
Dimensionality Reduction: Reducing the number of features while retaining
information (e.g., Principal Component Analysis - PCA).
o Examples:
Grouping customers by purchasing behavior.
Detecting anomalies in network traffic.
3. Semi-supervised Learning:
o Definition: Uses a combination of labeled and unlabeled data, leveraging the vast
amount of unlabeled data to improve learning.
o Goal: Achieve better accuracy with minimal labeled data.
o Applications:
Text classification with a small labeled dataset and a large unlabeled corpus.
Image recognition when only a few images are labeled.
4. Reinforcement Learning (RL):
o Definition: Agents learn by interacting with an environment to maximize
cumulative rewards.
o Key Concepts:
States: The current situation of the agent.
Actions: Choices available to the agent.
Rewards: Feedback received after an action.
o Techniques:
Q-Learning: A value-based approach.
Deep Q-Networks (DQN): Combines Q-Learning with deep learning.
o Applications:
Robotics: Learning optimal movement patterns.
Game playing (e.g., AlphaGo beating human players).
Models of Machine Learning:
1. Geometric Models:
o Represent data and decision boundaries geometrically in feature space.
o Example: Support Vector Machines (SVMs) use hyperplanes to separate classes.
o Applications: Binary classification tasks like spam detection.
2. Probabilistic Models:
o Utilize probability theory to model uncertainty and relationships in data.
o Examples:
Naïve Bayes: Assumes independence among features.
Hidden Markov Models (HMMs): Used for sequential data like speech
recognition.
o Applications: Text classification, sentiment analysis.
3. Logical Models:
o Based on logical rules and decision trees.
o Examples:
Decision Trees: Hierarchical structures for decision-making.
Rule-based Systems: Use "if-then" rules for predictions.
o Applications: Medical diagnostics, expert systems.
4. Grouping and Grading Models:
o Focus on clustering and ranking data.
o Examples:
K-means Clustering: Groups data points into clusters based on similarity.
Ranking Algorithms: Used in search engines to rank results.
o Applications: Customer segmentation, recommendation systems.
5. Parametric Models:
o Have a fixed number of parameters; assume a specific functional form.
o Examples:
Linear Regression: Models relationships between variables linearly.
Logistic Regression: Used for binary classification problems.
o Applications: Predicting sales, disease progression.
6. Non-parametric Models:
o Do not assume a fixed functional form; can grow in complexity with data.
o Examples:
K-Nearest Neighbors (KNN): Classifies data points based on nearest
neighbors.
Decision Trees: Grows dynamically with data.
o Applications: Flexible modeling for diverse datasets.
Case Studies:
1. Profiling the Online Storefronts of Counterfeit Merchandise:
o Objective: Identify patterns and features of online counterfeit operations.
o Techniques:
NLP: Analyzing product descriptions for signs of counterfeit claims.
Image Recognition: Identifying misuse of logos or trademarks.
o Outcome: Improved detection of fraudulent e-commerce activities.
2. Detecting Malicious Websites in Adversarial Classification:
o Objective: Build classifiers robust to adversarial examples from malicious entities.
o Techniques:
Ensemble Models: Combining multiple classifiers for better robustness.
Adversarial Training: Training models to recognize and counter adversarial
patterns.
o Outcome: Enhanced security in web browsing and online transactions.
3. Credit Card Fraud Detection:
o Objective: Detect fraudulent transactions using financial datasets.
o Techniques:
Supervised Learning: Logistic Regression, Random Forests.
Anomaly Detection: Identifying outlier transactions.
o Outcome: Reduced financial losses due to fraud.
4. Topic Models of the Underground Internet Economy:
o Objective: Analyze forums and marketplaces on the dark web.
o Techniques:
Latent Dirichlet Allocation (LDA): Extracting topics from unstructured text.
Sentiment Analysis: Understanding user sentiments in discussions.
o Outcome: Insights into underground economic activities.
5. Learning to Rate Vulnerabilities and Predict Exploits:
o Objective: Assess software vulnerabilities and predict exploitation likelihood.
o Techniques:
Probabilistic Models: Scoring vulnerabilities based on historical data.
Supervised Learning: Training on labeled datasets of known vulnerabilities
(CVEs).
o Outcome: Better prioritization and mitigation of security risks.
5 Marks Question Bank
Artificial Intelligence, Machine Learning, and Deep Learning:
1. Define Artificial Intelligence and categorize it into its major types.
2. Differentiate between Machine Learning and Deep Learning with examples.
Types of Learning:
1. Explain supervised learning with an example of a classification task.
2. Compare unsupervised learning and reinforcement learning with suitable examples.
Models of Machine Learning:
1. Briefly describe geometric models with an example.
2. What are the advantages of non-parametric models compared to parametric models?
Case Studies:
1. What are the objectives of detecting malicious websites using adversarial classification?
2. Mention key techniques used in profiling counterfeit merchandise storefronts.
10 Marks Question Bank
Artificial Intelligence, Machine Learning, and Deep Learning:
1. Explain the relationship between AI, ML, and DL with examples.
2. Discuss the applications of Deep Learning in different domains.
Types of Learning:
1. Explain reinforcement learning with an example of its application.
2. Discuss the role of semi-supervised learning in modern machine learning systems.
Models of Machine Learning:
1. Explain probabilistic models with examples of their applications.
2. Discuss the differences between grouping models and logical models with examples.
Case Studies:
1. Discuss the role of topic models in analyzing the underground Internet economy.
2. Explain the techniques and outcomes in credit card fraud detection case studies.
(Dr. Dipalee Divakar Rane)
Course Coordinator