Machine Learning Notes

Biological Neuron vs. Artificial Neuron

  • Biological Neuron:

    • Human cortex contains approximately 10 billion neurons.
    • Each neuron connects to thousands of others.
    • Neurons encode activations as brief electrical pulses.
    • Components: Dendrites, Soma, Nucleus, Axon, Axon terminal button.

  • Artificial Neuron:

    • Information-processing system mimicking biological neural networks.
    • Handles input signals, processes them, and outputs results.
    • Terms: connectionist models, parallel distributed processors.
    • Based on assumptions:
      • Information processing at neurons (nodes, units).
      • Signals pass over connection links.
      • Each link has a weight that multiplies the signal.
      • Neurons apply a nonlinear activation function.
      • Learning rule adapts weights for tasks.

  • Components of Artificial Neuron:

    • Nodes: Input, Hidden, Output.
    • Layers: Input, Hidden, Output.
    • Connection Weight: Strength of connection between nodes.

Introduction to Machine Learning

  • Artificial Intelligence (AI):

    • Computer Science area focused on creating intelligent machines that think, work, and react like humans.
    • Includes machine learning, deep learning, natural language processing, expert systems, robotics, and computer vision.

  • Machine Learning (ML):

    • Definition: "Gives computers the ability to learn without being explicitly programmed."
    • Subset of AI that extracts patterns from datasets.
    • Finds rules for optimal behavior and adapts to changes.
    • Evolved from pattern recognition.

  • Examples of Machine Learning:

    • Self-driving cars.
    • Fraud detection.
    • Web search results.
    • Credit scoring.
    • Online recommendations.

Typical Architecture of Machine Learning Algorithm

  • Single Layer Network:

    • Input Layer (Layer 0).
    • Processing Layer (Layer 1).
    • Output Layer.

  • Multi Layer Network:

    • Input Layer (Layer 0).
    • Hidden Layer (Layer 1).
    • Output Layer (Layer 2).

Characteristics of Machine Learning

  • Automated data visualization.

  • Optimized to learn complex patterns.

  • Accounts for interactions and nonlinear relationships.

  • Few assumptions.

  • Black box model interpretation is not straightforward.

  • Criteria for good ML systems:

    • Data preparation capabilities.
    • Basic and advanced algorithms.
    • Automation and iterative processes.
    • Scalability.
    • Ensemble modeling.

Supervised Learning vs. Unsupervised Learning

  • Learning:

    • A process involving changes in knowledge, beliefs, behaviors, or attitudes.

  • Types of Learning:

    • Supervised: Training data includes desired outputs.
    • Unsupervised: Training data does not include desired outputs (e.g., clustering).
    • Semi-Supervised: Training data includes a few desired outputs.
    • Reinforcement: Rewards from a sequence of actions.

  • Supervised Learning:

    • Adjusts weights in a neural net using a learning algorithm.
    • Requires multiple iterations through training data.
    • Performs pattern classification.

  • Unsupervised Learning:

    • Draws inferences from input data without labeled responses.
    • Common method: cluster analysis for exploratory data analysis.
    • Clusters modeled using similarity measures (Euclidean or probabilistic distance).