Artificial Intelligence Cheat Sheet

AI Problem-Solving Approaches

Approach	Description	Best For	Characteristics
Uninformed Search	Search without problem-specific knowledge	Simple problems, small state spaces	BFS, DFS, uniform cost, iterative deepening
Informed Search	Search using heuristic functions	Large state spaces with domain knowledge	Greedy best-first, A* search
Local Search	Search in current state space only	Optimization problems, constraint satisfaction	Hill climbing, simulated annealing, genetic algorithms
Adversarial Search	Multi-agent environments with conflicting goals	Game playing, strategic decision making	Minimax, alpha-beta pruning
Constraint Satisfaction	Find values that satisfy constraints	Scheduling, configuration problems	Backtracking, constraint propagation

Machine Learning Types

Type	Description	Examples	Algorithms
Supervised Learning	Learn from labeled training data	Image classification, spam detection	Linear regression, decision trees, SVM, neural networks
Unsupervised Learning	Find patterns in unlabeled data	Customer segmentation, anomaly detection	K-means clustering, PCA, association rules
Reinforcement Learning	Learn through interaction with environment	Game playing, robotics, autonomous driving	Q-learning, deep Q-networks, policy gradients
Semi-supervised Learning	Use both labeled and unlabeled data	Image classification with few labels	Self-training, co-training
Transfer Learning	Apply knowledge from one domain to another	Pre-trained models, domain adaptation	Feature extraction, fine-tuning

Search Algorithms

Algorithm	Type	Time Complexity	Space Complexity	Optimality
Breadth-First Search (BFS)	Uninformed	O(b^d)	O(b^d)	Yes (for unweighted graphs)
Depth-First Search (DFS)	Uninformed	O(b^m)	O(bm)	No
Uniform Cost Search	Uninformed	O(b^(C*/ε))	O(b^(C*/ε))	Yes
Greedy Best-First	Informed	O(b^m)	O(b^m)	No
A* Search	Informed	O(b^d)	O(b^d)	Yes (if heuristic is admissible)
Iterative Deepening	Uninformed	O(b^d)	O(bd)	Yes

Knowledge Representation

Method	Description	Advantages	Disadvantages
Logic-Based	Use logical statements to represent knowledge	Precise, supports inference	Computationally expensive, handles uncertainty poorly
Rule-Based	Represent knowledge as if-then rules	Easy to understand, maintain	Can become large and inconsistent
Semantic Networks	Represent relationships as a graph	Intuitive, visual representation	Not good for complex reasoning
Frames	Structured representation with slots and values	Good for representing objects	Complex to implement and maintain
Ontologies	Formal specification of concepts and relationships	Standardized, machine-readable	Requires domain expertise
Probabilistic Models	Represent uncertainty with probabilities	Handles uncertainty well	Complex to construct and compute

Machine Learning Algorithms

Algorithm	Type	Use Case	Pros	Cons
Linear Regression	Supervised (Regression)	Predicting continuous values	Simple, interpretable, efficient	Assumes linear relationship, sensitive to outliers
Logistic Regression	Supervised (Classification)	Binary classification	Simple, fast, provides probabilities	Assumes linear decision boundary
Decision Trees	Supervised (Classification/Regression)	Classification and regression	Interpretable, handles non-linear relationships	Prone to overfitting, unstable
Random Forest	Supervised (Classification/Regression)	Classification and regression	Reduces overfitting, handles non-linear relationships	Less interpretable, requires more computation
Support Vector Machines (SVM)	Supervised (Classification/Regression)	Classification with high-dimensional data	Effective in high dimensions, memory efficient	Doesn't provide probability estimates, sensitive to kernel choice
K-Means Clustering	Unsupervised (Clustering)	Grouping similar data points	Simple, efficient	Need to specify number of clusters, sensitive to initialization
Neural Networks	Supervised/Unsupervised	Complex pattern recognition	Universal approximators, handle complex relationships	Black box, requires large datasets, computationally expensive

Neural Networks

Type	Architecture	Use Case	Characteristics
Feedforward Networks	Layers of neurons, no cycles	Classification, regression	Simple, fully connected layers
Convolutional Neural Networks (CNN)	Convolutional layers, pooling, fully connected	Image processing, computer vision	Translation invariance, parameter sharing
Recurrent Neural Networks (RNN)	Loops to maintain state	Sequence modeling, NLP	Handles variable-length sequences
Long Short-Term Memory (LSTM)	Special RNN with memory cells	Long sequence modeling	Handles vanishing gradient problem
Transformer Networks	Attention mechanisms, self-attention	NLP, image processing	Parallel processing, captures long-range dependencies
Generative Adversarial Networks (GAN)	Generator and discriminator networks	Image generation, style transfer	Generates realistic data samples

AI Planning

Method	Description	Advantages	Disadvantages
Classical Planning	Find sequence of actions to reach goal	Systematic, optimal solutions	Assumes fully observable, deterministic environment
Probabilistic Planning	Handle uncertainty in actions and states	Realistic, handles uncertainty	Computationally expensive
Hierarchical Planning	Decompose complex tasks into subtasks	Manages complexity, reusable components	Requires domain knowledge
Multi-Agent Planning	Coordinate multiple agents to achieve goals	Handles distributed systems	Complex coordination, communication overhead
Temporal Planning	Consider timing constraints in planning	Realistic scheduling, resource allocation	Increased complexity

Expert Systems

Component	Description	Function	Examples
Knowledge Base	Domain knowledge stored in rules/facts	Store and represent knowledge	Medical diagnosis rules, tax calculation rules
Inference Engine	Reasoning mechanism	Apply rules to facts to derive conclusions	Forward chaining, backward chaining
Explanation Facility	Explain reasoning process	Justify conclusions	Trace of rule applications
Knowledge Acquisition Facility	Tool for adding knowledge	Interface for knowledge engineers	Knowledge acquisition tools
User Interface	Interface for users	Interaction with system	Text-based, graphical interfaces

Natural Language Processing (NLP)

Task	Description	Techniques	Applications
Tokenization	Split text into tokens (words, sentences)	Rule-based, statistical methods	Preprocessing for all NLP tasks
Part-of-Speech Tagging	Label each word with its grammatical role	HMM, CRF, neural networks	Grammar checking, information extraction
Named Entity Recognition (NER)	Identify and classify named entities	Rule-based, machine learning	Information extraction, question answering
Sentiment Analysis	Determine sentiment in text	Lexicon-based, machine learning	Social media monitoring, review analysis
Machine Translation	Translate text from one language to another	Rule-based, statistical, neural	Language translation systems
Text Summarization	Generate concise summary of text	Extractive, abstractive methods	News summarization, document processing
Question Answering	Answer questions based on text	Information retrieval, neural networks	Chatbots, search engines

Ethics and AI Safety

Concern	Description	Examples	Mitigation Strategies
Bias and Fairness	AI systems showing unfair behavior toward groups	Biased hiring algorithms, facial recognition bias	Debiased training data, fairness constraints
Privacy	AI systems compromising individual privacy	Data breaches, surveillance	Differential privacy, federated learning
Transparency	AI systems being "black boxes"	Uninterpretable neural networks	Explainable AI, model interpretability
Accountability	Difficulty assigning responsibility for AI decisions	Autonomous vehicle accidents	Clear governance, audit trails
Job Displacement	AI automating human jobs	Automation of routine tasks	Retraining programs, human-AI collaboration
Security	AI systems being vulnerable to attacks	Adversarial attacks, data poisoning	Robust training, secure implementation