mindmap
root((Machine Learning))
Supervised Learning
Regression
Simple Linear Regression
Multiple Regression
Polynomial Regression
Support Vector Regression
Random Forest Regression
Classification
Logistic Regression
Decision Tree
Random Forest
SVM
Neural Networks
Naive Bayes
k-Nearest Neighbors
Unsupervised Learning
Clustering
K-means
Hierarchical Clustering
DBSCAN
Association Rule
Apriori / FP-Growth
Dimensionality Reduction
PCA
t-SNE
Autoencoders
Reinforcement Learning
Q-Learning
Deep Q-Networks
Policy Gradient Methods
Introduction to Machine Learning
September 11, 2025
Human Progress: From Tools to AI
1950 – Alan Turing
- Publishes “Computing Machinery and Intelligence”
Turing concluded that instead of debating the abstract and philosophical question “Can machines think?”, it is more productive to reformulate the question into an operational test of intelligence. What he called the Imitation Game , now known as the
Turing Test
1956 – John McCarthy
Named the field — Artificial Intelligence.
Founded the first AI conference (Dartmouth, 1956).
Provided core tools (Lisp language) that drove AI research for decades.
- Shaped AI as a discipline with both vision and technical contributions.
1957 – Frank Rosenblatt
First neural network model with learning ability (Perceptron, 1957).
Demonstrated machine learning from data in practice.
Inspired later generations — although perceptrons had limitations (proved by Minsky & Papert in 1969), the concept
evolved into multilayer networks with backpropagation, which is the essence of modern deep learning.
1959 – Arthur Samuel
Coined the term “Machine Learning” (1959)
Defined it as:
“The field of study that gives computers the ability to learn without being explicitly programmed.”
1990s – Vladimir Vapnik & Alexey Chervonenkis
Develop Support Vector Machines (SVMs)
Establish theoretical foundations for statistical learning
Deep Blue
May 1997: Deep Blue defeated Garry Kasparov, the reigning World Chess Champion, in a six-game match (2 wins, 1 loss, 3 draws).
This was the first time a computer defeated a World Champion in a classical chess match under standard tournament time controls.
2016 – Demis Hassabis & DeepMind
Develop AlphaGo, defeating world champion Lee Sedol
Marks a major success in deep reinforcement learning
How Do Humans, Animals, and Machines Learn?
- Learn via experience, teaching, language
- Reasoning and abstraction
- Can generalize knowledge
- Creative and flexible
- Learn via trial & error and imitation
- Instinctive survival learning
- Fast in specific contexts
- Limited abstract reasoning
- Learn via data + algorithms
- Supervised, unsupervised, reinforcement learning
- Detect hidden patterns
- Extremely fast, but no true understanding
AI vs ML vs DL
Definition: The science of making machines “intelligent” — able to perform tasks that normally require human intelligence.
Scope: Includes symbolic logic, expert systems, search algorithms, robotics, natural language processing, and machine learning.
Key idea: AI is about creating intelligent behavior, not restricted to data-driven learning.
Definition: A field within AI that focuses on building algorithms that learn from data instead of relying only on hand-coded rules.
Why subset:
ML is one approach to achieve AI.
Not all AI is ML → e.g., early rule-based expert systems were AI but not ML.
Examples: Decision trees, Support Vector Machines (SVM), k-means clustering, regression.
Definition: A specialized branch of ML that uses artificial neural networks with many layers to automatically learn complex patterns.
Why subset:
DL is a specific technique inside ML.
All DL is ML, but not all ML is DL.
Classic ML (e.g., SVM, decision trees) does not use deep neural networks.
Examples: Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), Transformers (GPT, BERT).
Machine Learning Process
Data Collection: This is the process of gathering raw data from various sources such as databases, sensors, APIs, surveys, or files. The quality and quantity of collected data strongly influence the performance of the model.
Data Preparation: In this stage, the raw data is cleaned, transformed, and organized into a suitable format for training. Tasks include handling missing values, normalizing or standardizing data, feature selection, and splitting data into training, validation, and test sets.
Model Selection: Different algorithms (e.g., Decision Trees, Logistic Regression, Neural Networks) are considered. The goal is to choose a model type that best fits the problem domain, data structure, and business requirements.
Training: The chosen model is trained using the training dataset. During training, the model learns patterns and relationships by adjusting internal parameters (such as weights in neural networks) to minimize errors.
Hyperparameter Tuning: Unlike parameters learned during training, hyperparameters are set before training (e.g., learning rate, number of layers, max depth). Tuning involves systematically adjusting these values to improve model accuracy, efficiency, and generalization.
Evaluation: The trained model is tested using validation or test datasets to measure performance. Common metrics include accuracy, precision, recall, F1-score, or RMSE, depending on the type of problem (classification or regression).
Monitoring & Maintenance: Once deployed, the model is continuously monitored to ensure it performs well with new incoming data. Maintenance may include retraining with updated data, detecting model drift, and applying patches or improvements as needed.