Introduction to Artificial Intelligence

Introduction to AI

• Artificial Intelligence (AI) is the simulation of human intelligence in machines programmed to think, learn, and make decisions.
• AI involves creating algorithms and systems capable of performing tasks traditionally requiring human cognitive functions, like problem-solving, reasoning, and language understanding.
• AI is a multidisciplinary field drawing on computer science, mathematics, psychology, linguistics, and philosophy.

AI History

• 1940-1950: Early days of AI
• 1943: McCulloch & Pitts created a Boolean circuit model of the brain.
• 1950: Turing's "Computing Machinery and Intelligence" was published.
• 1950-1970: Excitement and early AI programs
• 1950s: Early AI programs included Samuel's checkers program, Newell & Simon's Logic Theorist, and Gelernter's Geometry Engine.
• 1956: The Dartmouth meeting adopted the term "Artificial Intelligence."
• 1965: Robinson's complete algorithm for logical reasoning.
• 1970-1990: Knowledge-based approaches
• 1969-79: Early development of knowledge-based systems.
• 1980-88: Expert systems industry booms.
• 1988-93: Expert systems industry busts, known as "AI Winter."
• 1990s: Statistical approaches
• Resurgence of probability and focus on uncertainty.
• General increase in technical depth.
• Emergence of agents and learning systems.
• 2000s: Continues with ongoing advancements.

Defining Intelligence

• Intelligence is defined as the capacity to acquire and apply knowledge
• Intelligence is defined as the faculty of thought and reason.

Defining Artificial Intelligence

• AI is the study of systems that appear intelligent to an observer.
• AI uses methods based on the intelligent behavior of humans and animals to solve complex problems.
• AI deals with difficult, real-world problems and requires complex reasoning and knowledge.

Approaches to AI

• Think like people: Designing systems that are as intelligent as humans, emulating the human thought process.
• Act like people: Creating machines that perform functions requiring intelligence when performed by people, exemplified by the Turing Test.
• Think rationally: Focuses on the study of computations that enable systems to perceive, reason, and act.
• Act rationally: Focuses on rational behavior doing the right thing to maximize goal achievement, given available information.

The Turing Test

• 1950: Alan Turing proposed an operational definition of intelligence.
• The Turing test consists of an interrogator, a computer, a person answering questions, and a separator.
• The interrogator asks questions to both the computer and the person.
• The interrogator then identifies which one is the computer.
• To pass the Turing test, the machine must convince the interrogator into believing that it is human.
• Natural language processing is needed to communicate successfully.
• Knowledge representation is needed to store knowledge.
• Automated reasoning is needed to answer questions and draw conclusions.
• Machine learning is needed to adapt to new circumstances and detect patterns.
• Computer vision is needed to perceive objects.
• Robotics are needed to manipulate objects and move.

Rational Decisions

• Rationality is defined here as maximally achieving pre-defined goals.
• Rationality only concerns what decisions are made, not the thought process behind them.
• Goals are expressed in terms of the utility of outcomes.
• Being rational means maximizing expected utility.

Goals of AI

• Engineering: Solve real-world problems using knowledge and reasoning, creating new opportunities in business, engineering, and other applications.
• Scientific: Use computers as a platform for studying intelligence, designing theories, hypothesizing aspects of intelligence, and implementing these theories on a computer.
• AI researchers focus on automating intelligence as AI technology becomes integrated into everyday life.

AI Application Systems

• Expert Systems
• Diagnostic Systems
• Financial Decision Making
• Configuring Hardware and Software
• Robotics
• Web search Engines

Problem Solving in AI

• Problem-solving utilizes algorithms and procedures to solve problems, make decisions, or learn from experience.

Types of Problems in AI

• Deterministic problems: Al systems that produce the same output for a particular input, following predefined rules.
• Non-deterministic problems: Al systems that may produce different outputs for the same input due to inherent randomness, handling uncertainty and adapting.
• Discrete problems: Problems with a limited number of distinct states and actions.
• Continuous problems: AI states and actions can take on any value within a certain range.

Data Mining

• Data Mining (DM) is defined as the nontrivial extraction of implicit, previously unknown, and potentially useful information from data. (Piatetsky - Shapiro)

Piatetsky-Shapiro View of Data Mining Steps

• Start with Initial Data.
• Select relevant data to get Target Data.
• Clean data to get Preprocessed Data.
• Convert data into usable format to get Transformed Data.
• Discover Data Mining.
• Data is used to build Data Model.
• Data Model is interpreted to gain Knowledge.

DM Goals

• Prediction: predict future/unknown values of variables/patterns utilizing Machine learning and Pattern recognition
• Description: analyse given datasets, pattern mining, associative rule mining and clustering

Types of Data Mining Tasks

• Predictive: Includes classification (e.g., k-Means, PCA, ...)
• Descriptive: Includes clustering (e.g., k-Means, PCA, ...)
• Clustering: unsupervised learning to group objects so objects are similar to each other in the same cluster
• Classification: supervised learning approach to predict categorical class labels of new instances

Data Representation

• Discrete Categories: (colour of a flower petal)
• Numerical Data: integer values (number of petals in a flower), real values (length of a petal)
• String/Textual Data: words in a document, time series data, continuous, chronological, flows in one direction

Data Pre-processing

• Data pre-processing improves the quality of data, and obtained results from data analysis
• Data Cleaning: remove noise and correcting inconsistencies
• Data Integration: combining data from different sources and providing users with a unified view of these data
• Data Transformation: normalizing and aggregating data so that it can be brought into a standard format. Scales the data within a small, specified range
• Data Discretization: Converting continuous data into discrete buckets or intervals and improves the efficiency of certain algorithms
• Data Reduction: To reduce the volume and achieve the same or similar analytical results with includes binning, histograms, clustering, and Principal Component Analysis (PCA)
• Feature Selection: selecting the most important features to reduce the number of input variables

Information Retrieval

• Information retrieval (IR) is the process of obtaining relevant information from large repositories of data.
• It is a key component of search engines and database systems.
• IR deals with the retrieval of data and information resources (web pages, catalogs, multimedia, etc).

Knowledge Representation

• Knowledge Representation (KR) is the way we organize and structure information in artificial intelligence.
• It's how an Al system understands and uses information to solve problems
• KR Category:
  • Symbolic Representation
  • Sub-symbolic Representation
  • Ontologies

Uncertainty in AI

• Refers to situations where the state of the world is not completely known, or the outcomes of actions are not completely predictable.
• Can be due to inherent randomness, incomplete or noisy data, or the complexity of the world.
• Aleatory uncertainty (inherent randomness);
• Epistemic uncertainty (lack of knowledge).
• Methods for dealing with uncertainty:
  • Probabilistic reasoning
  • Fuzzy logic
  • Bayesian networks

Learning in AI

• A process that allows artificial intelligence systems to improve their performance on specific tasks over time.
• Achieved by modifying the system's internal parameters based on past experiences and feedback.
• Three Types of Learning
  • Supervised
  • Unsupervised
  • Reinforcement

Types of Learning

• Supervised Learning: The Al is provided with input-output pairs, where the output is the 'correct' answer for each input. The Al learns a function that maps inputs to correct outputs.
• Unsupervised Learning: The Al is given inputs but no explicit outputs. The goal is to find structure in the inputs, such as clustering or dimensionality reduction.
• Reinforcement Learning: The Al learns how to perform actions based on reward feedback. It's about making a sequence of decisions, with the goal of maximizing a reward signal.

Importance of Learning

• To allows Al systems to adapt to new circumstances, generalize from previous experiences, and improve over time.

What is Machine Learning (ML)

• ML is a subset of artificial intelligence involving the creation of algorithms that can modify themselves without human intervention to produce desired outputs by feeding on data.
• The goal of ML is enabling computers to learn from data and make decisions based on data.

Importance of ML

• ML enables computers to handle new situations via analysis, self-training, observation, and experience.
• ML turns a deluge of data into knowledge, predictions, and decisions.

Key Concepts of ML

• Data: The raw information from which machines learn
• Model: A representation of what is being learned
• Training: The process of learning from data
• Inference: Making predictions on new data based on the trained model

Applications of Machine Learning

• Image Recognition
• Speech Recognition
• Automatic Language Translation
• Medical Diagnosis
• Stock Market trading
• Online Fraud Detection
• Virtual Personal Assistant
• Product Recommendation
• Self-driving cars
• Email Spam and Malware Filtering

Types of ML Algorithms

• Supervised Learning: Learning from labeled data.
• Unsupervised Learning: Finding structure in unlabeled data.
• Reinforcement Learning: Learning from actions and rewards.
• Depending on the nature of the data and the problem at hand.
• Each type has its own set of algorithms, challenges, and use cases.

Supervised Learning Concepts

• The output we want our model to predict is the Label.
• Input variables used to make the prediction are Feaures.
• The dataset from which the machine learns is the Training set.
• Supervised Learning is about Classification and Regression.
• Classification (predict categorical values)
• Regression (predict real value)

Supervised Learning Common Algorithms

• Linear Regression
• Logistic Regression
• Support Vector Machines
• Decision Trees and Random Forests
• Neural Networks

Supervised Learning Applications

• Email spam filters
• Credit scoring
• Medical diagnosis
• Fraud detection
• Recommendation systems
• Supervised learning is fundamental for applications where one can provide the machine with examples of correct behavior.

Unsupervised Learning COncepts

• Unsupervised Learning identifies patterns in data without any labels, often used to find groups or patterns within the data
• Cluster: A group of data points aggregated because of certain similarities.
• Dimensionality Reduction: The process of reducing the number of variables under consideration.

Unsupervised Learning Applications

• Common Algorithms
• K-Means Clustering
• Hierarchical Clustering
• Principal Component Analysis (PCA)
• Autoencoders
• Applications
• Market segmentation
• Organizing large libraries of media
• Anomaly detection
• Unsupervised learning helps to understand the underlying structure of data and can reveal hidden patterns that are not immediately apparent.

Reinforcement Learning

• RL algorithm learns by taking actions in an environment to maximize some notion of cumulative reward.
• Key Concepts: Agent: The learner or decision-maker. Environment: What the agent interacts with. Reward: Feedback from the environment to assess the actions of the agent. Policy: The strategy that the agent employs to make decisions.
• RL Common Algorithms:
• Q-Learning
• Deep Q Network (DQN)
• Policy Gradients
• Actor-Critic Methods
• RL Applications Include: Playing video games Autonomous vehicles Robotics
• Reinforcement learning is at the cutting edge of Al, enabling machines to make decisions with long-term outcomes in mind, often in complex environments.

Deep learning

• Is a branch of machine learning that is particularly well suited for modeling complex patterns in data.
• Utilizes neural networks with multiple layers (hence "deep") to learn representations of data in an incremental manner.
• Models handle large-scale, high-dimensional data and perform tasks such as image recognition, natural language processing, and time series analysis with state-of-the-art accuracy.
• Neural Networks: These networks consist of interconnected nodes, or "neurons," which process information in a layered structure.
• Layers: Deep neural networks comprise input, hidden, and output layers. The input layer receives the raw data, the hidden layers process the data through various transformations, and the output layer provides the final decision or prediction.
• Activation Functions: They introduce non-linearity into the network, allowing the model to learn more complex patterns. Common activation functions include ReLU (Rectified Linear Unit), sigmoid, and tanh.
• Backpropagation: involves a forward pass through the network to compute the output and loss, followed by a backward pass to update the weights of the network using gradient descent.
• Gradient Descent: technique to minimize the loss function by iteratively updating the parameters of the network in the direction that reduces the loss.
• Loss Functions: measure how well the model's predictions match the actual data. Common loss functions include mean squared error for regression tasks and cross-entropy for classification tasks.
• Regularization: Techniques like dropout, L1/L2 regularization are used to prevent overfitting, where the model performs well on training data but poorly on unseen data.

Types of Deep Learning Models

• Types of Deep Learning Models:
• Convolutional Neural Networks (CNNs): (image processing and computer vision tasks)
• Recurrent Neural Networks (RNNs): (sequential data and natural language
• Long Short-Term Memory (LSTM) and Gated Recurrent Units (GRUs): These are (long sequence data)
• Autoencoders: (dimensionality reduction or feature learning
• Generative Adversarial Networks (GANs): (generating new data that is similar to the training data

Deep Learning Applications

• Image and Video Recognition: facial recognition to self-driving cars
• Natural Language Processing (NLP): Deep learning has led to significant advances in machine translation, sentiment analysis, and question-answering systems.
• Speech Recognition: virtual assistants and voice-controlled devices, accurately transcribe and understand spoken language.
• Recommendation Systems: power recommendation engines of many online services, personalized content to users.
• Drug Discovery and Genomics: life sciences, predict the effects of molecules and assist in the design of new drugs.

ML Challenges Include:

• Overfitting: A model learns the training data too well and performs poorly on unseen data
• Underfitting: A model is too simple to capture the underlying pattern of the data
• Addressing the issues of overfitting and underfitting
  • Cross-validation
  • Regularization
  • Ensemble methods
  • Can mitigate overfitting and underfitting

ML in Industry

• Healthcare: Predictive analytics for patient care, drug discovery, and medical imaging.
• Finance: Fraud detection, algorithmic trading, and credit scoring.
• Retail: Personalized recommendations, inventory management, and customer service automation.