Reinforcement Learning Basics

Study Notes

A machine learning technique that trains software to optimize decision-making through a trial-and-error learning process.
Positive reinforcement increases the strength and frequency of behavior by associating it with favorable outcomes.
Negative reinforcement strengthens behavior by removing or avoiding negative conditions.
Policy determines actions based on the current state and rules set by the agent.
Value function estimates the potential benefit of being in a specific state.
A model can mimic the environment, predicting future states and rewards based on agent actions.

Deep learning, a subset of machine learning, uses multilayered neural networks to simulate human decision-making.
Neural networks operate similarly to biological neurons, processing inputs and arriving at conclusions.
Applications of deep learning are prevalent in NLP, enhancing interactions in search engines, chatbots, GPS, and digital assistants.
NLP combines computational linguistics with machine learning to enable computers to understand and generate human language, facilitating generative AI.

Field of AI focused on enabling systems to interpret and derive information from visual inputs.
Human vision provides innate understanding of context, while machine learning mimics this perception using cameras and algorithms.
Applications include:
- Autonomous vehicles using sensors for navigation.
- Identifying patterns and abnormalities in medical images, aiding in disease detection.
- AI in radiology optimizes workflows by prioritizing cases and speeding up diagnoses.

AI enhances fraud detection, algorithmic trading, and risk assessment within the finance sector.
Fraud detection involves:
- Anomaly detection to identify unusual patterns in transactions.
- Predictive analysis to prevent potential fraud based on historical data.
- Enhanced security with strong protocols like biometric authentication.
Algorithmic trading includes:
- Quantitative trading utilizing data models for large transactions.
- High-frequency trading that analyzes real-time data for rapid decision-making.
- Arbitrage trading to exploit price differences across markets.
Risk assessment benefits from AI’s ability to analyze extensive customer data to accurately predict creditworthiness.

Bias in AI entails discrimination against individuals or groups based on attributes like race or gender.
Fairness ensures equitable decision-making without disadvantaging any group.
Sources of bias include:
- Data bias arising from non-representative training data.
- Algorithmic bias from inherent biases in design and implementation.
- Human bias introduced by developers.
Types of bias include:
- Selection bias from non-representative training data.
- Measurement bias due to inaccuracies in data collection.
Fairness can be viewed from frameworks like distributive fairness, which focuses on equitable resource distribution, and procedural fairness, which emphasizes unbiased decision-making processes.