Yet another `AI` cheat sheet

As AI (Artificial Intelligence) continues to evolve, I sat down to prepare a structured overview of this topic for me to the have some cheat sheet for looking up related terms and definitions quickly. Of course, as is appropriate for this topic, it is important to note that I wrote this blog using an AI tool: AI types, from supervised Neural Networks to Generative AI like GANs, offer specialized solutions—ranging from classification to creative content generation—tailored for diverse tasks and industries.

Artificial Intelligence (AI) stands as a transformative force in computer science, shaping the landscape of technology with its diverse array of methodologies and applications. Categorizing AI into various types delineates distinct approaches that algorithms and systems use to solve problems and interact with their environments. From Supervised Learning’s pattern recognition to Generative AI’s creative content generation, these classifications not only encapsulate the evolution of AI techniques but also underscore their profound impact on computer science. Understanding these AI types not only aids in comprehending the breadth of AI capabilities but also serves as a compass guiding the direction of technological advancements, fostering innovation across industries and redefining the frontiers of computer science.

AI and Machine Learning (ML) compared

Artificial Intelligence (AI) and Machine Learning (ML) are related but distinct fields within computer science. As they often are used interchangeably, the table below attempts a differentiation between the two fields AI and ML:

Aspect Artificial Intelligence (AI) Machine Learning (ML)
Scope Broad field encompassing tasks that require human intelligence, including reasoning, problem-solving, perception, and more. Subfield of AI focused on developing algorithms that learn from data to make predictions or decisions.
Approach Multiple approaches, including rule-based systems, Expert Systems, Neural Networks, and more. Not limited to one specific technique. Focuses on the development of algorithms that learn patterns and make predictions based on data, using techniques like Supervised Learning, Unsupervised Learning, and Reinforcement Learning.
Goal Create intelligent systems that mimic or replicate human-like cognitive functions and behaviors across various domains. Develop algorithms that can generalize patterns from data to make accurate predictions or decisions when presented with new, unseen data.
Examples Autonomous vehicles, Natural Language Processing, robotics, game-playing AI (e.g., chess or Go), recommendation systems. Image recognition, speech recognition, recommendation systems, fraud detection, autonomous drones, and many more.

In summary, AI is the broader field that encompasses the development of intelligent systems, while ML is a subset of AI that focuses specifically on creating algorithms and models that can learn and improve from data. ML is a key enabler of AI, as it provides the tools and techniques for building intelligent systems that can adapt and improve their performance over time based on experience.

AI categorization

Below find a table outlining various AI types and the corresponding technologies commonly associated with each type, whilst specifying the types of problems that each AI type is commonly used to solve:

AI Type Technology Behind It Problems Solved
Supervised Learning Neural Networks (e.g., CNNs, RNNs, MLPs) Classification, Regression, Pattern Recognition
Unsupervised Learning Clustering (e.g., K-means), Dimensionality Reduction Clustering, Anomaly Detection, Feature Extraction
Reinforcement Learning Q-Learning, Policy Gradient Methods, Deep Q-Networks (DQNs) Game Playing, Robotics, Autonomous Systems
Semi-Supervised Learning Neural Networks, Graph-based Methods Data Labeling, Object Recognition, Document Classification
Transfer Learning Pre-trained Neural Networks, Fine-tuning techniques Domain Adaptation, Image Recognition, Natural Language Processing (NLP) tasks
Symbolic AI / Classical AI Expert Systems, Rule-based Systems, Logic Programming Reasoning, Knowledge Representation, Decision Support Systems
Neural Networks Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), Multilayer Perceptrons (MLPs) Image Recognition, Natural Language Processing, Time Series Prediction
Expert Systems Rule-based Systems, Inference Engines Diagnosis, Consultation Systems, Troubleshooting
Evolutionary Algorithms Genetic Algorithms, Genetic Programming, Evolution Strategies Optimization, Search Problems, Parameter Tuning
Generative AI Generative Adversarial Networks (GANs), Variational Autoencoders (VAEs), Autoregressive Models Image Generation, Text Generation, Content Creation
Edge AI Lightweight Neural Networks, Quantization Techniques IoT Devices, Real-time Processing, Low-latency Applications
Natural Language Processing Transformer Models (e.g., BERT, GPT), LSTM, Word Embeddings Language Translation, Sentiment Analysis, Text Summarization
Hybrid AI Systems Combination of various AI techniques (e.g., Neural-Symbolic Integration) Multimodal Learning, Complex Problem-solving
Reflective AI Self-monitoring, Self-assessment, Self-improvement mechanisms Self-assessment, Self-improvement, Error detection and correction

The table above outlines the various AI types and the technologies behind them. AI techniques can often be applied across multiple problem domains. The listed problems are examples of typical applications associated with each AI type, and these technologies might be utilized for various other tasks and industries based on their adaptability and advancements in the field.

Neural Networks are prevalent in various AI types, not all AI techniques exclusively rely on them. Symbolic AI, Evolutionary Algorithms, Expert Systems, and some Unsupervised Learning methods may not necessarily use Neural Networks as their primary technology.

Also, the technologies mentioned for each AI type are not exhaustive but represent some of the commonly associated techniques or frameworks within each category. AI research and development continuously evolve, leading to new methodologies and advancements in technologies applied to these AI types.

Machine Learning Paradigms

Machine learning paradigms, encompassing various techniques, enable machines to learn from data. Key areas include Supervised Learning, where algorithms learn from labeled data to make predictions, crucial for scenarios with known desired outputs. Unsupervised Learning discovers patterns in unlabeled data, useful in clustering and dimensionality reduction. Reinforcement Learning trains models through reward-based decision sequences, applicable in gaming and navigation. Semi-Supervised Learning combines labeled and unlabeled data, improving learning efficiency, especially when extensive labeled data is unavailable. These paradigms, each addressing unique AI challenges, are fundamental to AI’s diverse applications.

Supervised Learning

Supervised Learning algorithms are a cornerstone of AI, primarily used for classification and regression tasks. They rely on labeled training data, where the algorithm learns to map input data to corresponding output labels. This type of AI is widely employed in applications like image recognition, speech recognition, and spam detection. It’s particularly effective in scenarios where the relationship between the input data and the output is clear, and the goal is to extrapolate from the training data to unseen situations.

Unsupervised Learning

Unsupervised Learning techniques excel at discovering patterns and structures within unlabeled data. Common applications include clustering similar data points, dimensionality reduction for visualization, and anomaly detection, making it valuable for customer segmentation and data exploration. These techniques are key in scenarios where the data doesn’t come with predefined labels, and the goal is to derive insights directly from data without guidance.

Reinforcement Learning

Reinforcement Learning powers autonomous decision-making systems by allowing agents to learn optimal actions through interaction with an environment. Widely applied in robotics, game playing, and self-driving cars, it employs reward signals to guide agents toward achieving goals. This paradigm is unique in its focus on learning from the consequences of actions, rather than from direct data labeling, enabling machines to develop complex behaviors that maximize rewards over time.

Semi-Supervised Learning

This approach harnesses the benefits of both labeled and unlabeled data. By leveraging a small set of labeled data alongside a larger pool of unlabeled data, Semi-Supervised Learning reduces the need for extensive labeling efforts while improving model accuracy. This paradigm is particularly useful in real-world scenarios where obtaining labeled data can be costly or impractical, but there’s a need for more accurate models than what Unsupervised Learning alone can provide.

Neural Networks and Their Types

Neural Networks (NN), inspired by the human brain, consist of interconnected layers of artificial neurons. They are the driving force behind modern AI’s success in computer vision, Natural Language Processing, and speech recognition. These networks are capable of learning complex patterns and relationships in data, making them suitable for a wide range of applications from simple tasks like recognizing handwritten digits to complex ones like understanding human speech.

Convolutional Neural Network (CNN)

Convolutional Neural Networks, or CNNs, are specialized Neural Network architectures designed for processing grid-like data, such as images and videos. They use convolutional layers to automatically learn spatial hierarchies of features, making them highly effective in tasks like image classification, object detection, and facial recognition. CNNs have revolutionized the field of computer vision by enabling machines to interpret and analyze visual information with high accuracy.

Recurrent Neural Network (RNN)

Recurrent Neural Networks, known as RNNs, are designed to handle sequential data, making them suitable for tasks involving time series, natural language, and speech. RNNs are unique in their ability to process sequences of inputs by maintaining an internal state that captures the information about previous elements in the sequence, allowing them to make predictions about future elements.

Multilayer Perceptrons (MLP)

Multilayer Perceptrons, or MLPs, are a type of Neural Network architecture with multiple layers of artificial neurons, including an input layer, hidden layers, and an output layer. They are versatile and widely used in various machine learning tasks, including classification, regression, and data transformation. MLPs are foundational to many neural network structures and serve as a basis for more complex architectures.

Generative Adversarial Network (GAN)

Generative Adversarial Networks, or GANs, consist of two Neural Networks, a generator and a discriminator, engaged in a competitive process. GANs are renowned for their ability to generate realistic data, making them invaluable in image synthesis, art generation, and data augmentation. This innovative architecture allows for the creation of highly realistic synthetic data, which can be used for a variety of purposes including training other machine learning models.

Variational Autoencoder (VAE)

Variational Autoencoders, or VAEs, combine autoencoders with probabilistic modeling. They generate data by mapping it into a latent space, enabling tasks like image generation, data reconstruction, and anomaly detection. VAEs are particularly useful in scenarios where the goal is to learn complex distributions of data and generate new samples that are similar to the original dataset.

Deep Q-Network (DQN)

Deep Q-Networks, or DQNs, combine deep learning with reinforcement learning. They use Neural Networks to approximate the Q-values in Q-Learning, enabling more complex and efficient decision-making. DQNs have been successfully applied in various gaming environments, demonstrating the potential of combining deep learning techniques with reinforcement learning strategies.

Advanced Machine Learning Techniques

Advanced Machine Learning Techniques represent a significant stride beyond basic algorithms, involving sophisticated methods that push the boundaries of AI’s capabilities. These techniques include Transfer Learning, which has become a game-changer in fields like Natural Language Processing (NLP) and computer vision. This approach allows models trained on one dataset or task to be repurposed for another task, dramatically reducing the need for large labeled datasets and extensive computational resources. In decision-making and gaming, Reinforcement Learning techniques such as Policy Gradient Methods and Q-Learning are essential. These methods provide advanced strategies for agents to learn and adapt optimal actions within complex, dynamic environments, enhancing AI’s ability to make intelligent decisions in scenarios ranging from strategic game play to real-time autonomous navigation.

Transfer Learning

Transfer Learning is particularly transformative in NLP and computer vision, where models like BERT and GPT, once trained on vast datasets, can be fine-tuned to perform specific tasks with a fraction of the effort and data traditionally required. This technique not only saves valuable time and resources but also opens up new possibilities in AI applications, making advanced models more accessible and versatile.

Policy Gradient Methods

Policy Gradient Methods stand out in Reinforcement Learning for their direct approach to learning optimal policies. Unlike traditional methods that focus on value estimation, these techniques optimize the policy function directly, leading to more nuanced and effective decision-making strategies. They are particularly well-suited for environments with continuous action spaces, such as robotics and autonomous vehicle navigation.


Q-Learning, a cornerstone of Reinforcement Learning, is essential for training agents in environments where the reward dynamics are complex and multi-dimensional. By learning the value of actions in different states, agents can make informed decisions that maximize long-term rewards. This technique has been fundamental in advancing AI in areas such as robotics, strategic game play, and various autonomous systems.

Natural Language Processing (NLP)

Natural Language Processing stands at the forefront of bridging human communication and AI. This field involves complex techniques that enable machines to interpret, generate, and interact with human language in a meaningful way. NLP applications are diverse, including language translation, sentiment analysis, chatbots, and content summarization, and they are fundamentally transforming how we interact with technology, making these interactions more natural and intuitive.


BERT (Bidirectional Encoder Representations from Transformers) has revolutionized NLP with its deep understanding of context and nuance in language. Unlike previous models, BERT interprets words in relation to all other words in a sentence, leading to a more profound understanding of language. This has significantly improved performance in tasks like question-answering and language inference.


GPT (Generative Pre-trained Transformer) is renowned for its exceptional text generation capabilities. As a state-of-the-art language model, GPT excels in creating coherent and contextually relevant text, pushing the boundaries of AI’s creative and linguistic abilities. This model has wide-ranging applications, from writing assistance to conversation simulation.

Autoregressive Model

Autoregressive models are key in sequential data generation, where each output is conditional on previous outputs. These models are vital in NLP for tasks like language modeling and text prediction, and they also play a crucial role in image generation, providing a framework for understanding and generating sequential data.

Symbolic AI / Classical AI

Symbolic AI, also known as Classical AI, uses logic-based algorithms and knowledge representation to simulate human problem-solving and decision-making processes. This form of AI is fundamental in systems that require clear, explainable reasoning and decision-making, such as Expert Systems used in medical diagnosis, legal advice, and other specialized domains.

Expert Systems

Expert Systems utilize rule-based knowledge to replicate the decision-making ability of human experts. These systems are particularly useful in fields where specialized knowledge is essential, providing consistent, reliable advice and decision support based on a predefined set of rules and knowledge.

Rule-based Systems

Rule-based Systems are a cornerstone of Symbolic AI, using a set of predefined rules to make inferences and decisions. These systems are critical in applications that require logical reasoning and decision-making, such as automated customer service, compliance monitoring, and more.

Logic Programming

Logic Programming, exemplified by languages like Prolog, allows AI systems to represent and process knowledge using formal logic. This approach is integral to tasks that involve complex problem-solving and knowledge manipulation, such as in automated planning, reasoning, and intelligent database querying.

Additional AI Techniques and Applications

Various AI techniques extend beyond core machine learning paradigms, broadening AI’s impact. Evolutionary Algorithms utilize natural selection principles for optimization, crucial in complex scenarios. Generative AI, including GANs and VAEs, innovates in creating realistic media content. Edge AI optimizes data processing near its source, enhancing efficiency and privacy. Hybrid AI Systems merge different AI methods for tackling complex problems. Reflective AI emphasizes self-assessment and adaptation, key for AI’s long-term reliability and adaptability.

Evolutionary Algorithms

Evolutionary Algorithms optimize solutions through natural selection-like processes, including mutation, crossover, and selection. These algorithms excel in solving complex, multi-dimensional optimization problems where traditional approaches might falter, such as in evolutionary art, genetic programming, and automated design.

Generative AI

Generative AI, particularly through GANs and VAEs, is at the forefront of creative AI applications. It’s transforming content creation across various domains, enabling machines to generate new, original content that’s indistinguishable from human-created work, thus opening up new avenues in art, design, and media production.

Edge AI

Edge AI refers to AI algorithms that are processed locally, often on the device where data is collected, like smartphones or IoT devices. This approach reduces latency, conserves bandwidth, and enhances privacy by processing data on-device without needing to send it back to a central server or cloud.

Hybrid AI Systems

Hybrid AI integrates multiple AI techniques, combining the strengths of different approaches, combining their strengths to create more robust, versatile, and efficient solutions. This approach is particularly effective in complex scenarios where a single AI technique may not be sufficient, such as in advanced robotics, personalized medicine, and complex decision-making systems.

Reflective AI

Reflective AI embodies the advanced capability of AI systems to self-monitor, evaluate, and improve their performance. This self-reflective capability is crucial in ensuring AI systems remain effective, accurate, and reliable over time, particularly in dynamic environments where continuous adaptation is key.

Clustering and Dimensionality Reduction

In Unsupervised Learning, Clustering and Dimensionality Reduction are crucial for analyzing data. Clustering, like K-Means, groups data based on similarities, useful in pattern recognition and segmentation. Dimensionality Reduction, using methods like PCA, simplifies datasets by reducing variables, aiding in data visualization and performance enhancement. These techniques are key in extracting valuable insights from large datasets for effective data-driven decisions.


K-Means clustering algorithm is widely used for its simplicity and efficiency in grouping data into distinct clusters based on their features. This method is particularly useful in market segmentation, image compression, and as a pre-processing step in complex data analysis workflows.

Dimensionality Reduction

Dimensionality reduction, with PCA as a notable example, is crucial for managing high-dimensional data. By reducing the number of features while retaining essential information, these techniques simplify data analysis, improve model training efficiency, and enable better data visualization and interpretation.

Transformer Model

Transformer models, characterized by their self-attention mechanisms, have revolutionized NLP by enabling more effective processing of sequential data, particularly in tasks like language translation, text summarization, and context-aware responses. The success of transformers lies in their ability to handle long-range dependencies and parallelize computations, leading to significant improvements in processing speed and model performance.

Word Embeddings

Word embeddings represent a significant advancement in NLP, providing a way for models to understand the contextual meanings of words. Techniques like Word2Vec, GloVe, and FastText transform words into dense vectors, capturing semantic similarities and relationships in a way that enhances the performance of NLP models in tasks like text classification, sentiment analysis, and more.

Combination of Various AI Techniques

The integration of different AI methodologies, seen in Hybrid AI Systems, marks a significant advancement in AI. It combines the strengths of various paradigms, like neural network learning and symbolic AI reasoning. A key example is Neural-Symbolic Integration, which merges the efficiency of neural networks with the clarity of symbolic AI. This synergy enhances AI’s problem-solving capabilities, enabling more sophisticated applications in fields like personalized medicine and complex robotics.

Neural-Symbolic Integration

Neural-Symbolic Integration is an emerging field focused on combining the fluid learning capabilities of neural networks with the structured, logical reasoning of symbolic AI. This integration promises to create AI systems that not only learn from vast amounts of data but also reason and interpret information in a more human-like manner, enhancing AI’s applicability in complex, real-world scenarios.

Self-Monitoring in AI

Self-Monitoring in AI is vital for developing more autonomous and reliable AI systems. It encompasses Self-Assessment and Self-Improvement Mechanisms, enabling AI systems to continuously evaluate and enhance their performance. These features are crucial in areas requiring long-term operation and adaptability, like autonomous vehicles and healthcare diagnostics. Self-Monitoring allows AI to autonomously assess its decisions and refine its algorithms, ensuring accuracy and reliability. Integrating these capabilities is essential for the evolution of AI, enhancing its robustness and effectiveness in diverse applications, from automotive to healthcare.


Self-Assessment mechanisms in AI involve systems critically evaluating their decisions and outcomes, comparing their performance against predefined benchmarks or objectives. This introspective capability is key to maintaining the accuracy, reliability, and trustworthiness of AI systems, especially in critical applications.

Self-Improvement Mechanisms

Self-Improvement Mechanisms enable AI systems to autonomously refine and enhance their algorithms based on performance assessments. This adaptive capability ensures that AI systems can evolve and improve over time, becoming more effective and efficient in their tasks, and adapting to new challenges and environments.

Four primary AI types

The four primary AI types - reactive, limited memory, theory of mind, and self-aware - can be related to the AI types previously discussed in the following manner:

1. Reactive AI

This type of AI operates based on predefined rules and patterns, making decisions and taking actions solely based on the current input without storing or learning from past experiences. It corresponds to some extent with certain AI types like traditional Symbolic AI / Classical AI and specifically Supervised Learning models that operate on present data without retaining memory of past instances.

2. Limited Memory AI

These AI systems have the ability to learn from past experiences to some extent. They use a limited memory of previous data or observations to make decisions. AI types like Reinforcement Learning and some forms of Supervised and Unupervised Learning, which can utilize past experiences to inform current decisions or actions, align to a certain degree with the concept of limited memory AI.

3. Theory of Mind AI

This hypothetical AI type refers to systems that can understand and attribute mental states, beliefs, intentions, and emotions to themselves and others. Current AI technologies primarily focus on tasks like Natural Language Processing (NLP) and sentiment analysis, which provide some level of understanding of human communication and emotions, albeit not at the level of true theory of mind.

4. Self-Aware AI

This type of AI involves systems that possess consciousness, self-awareness, and understanding of their own existence and capabilities. As of now, AI systems do not possess true self-awareness. The closest related concept could be Reflective AI, where systems can assess their own performance, make improvements, or understand their limitations to some extent. However, achieving true self-awareness remains a significant challenge in AI and is more within the realm of philosophical speculation rather than current technological capability.

These primary AI types offer a conceptual framework to understand the development and capabilities of AI systems, while the previously mentioned AI types represent different methodologies, techniques, and applications within the field, each addressing specific problem-solving approaches and domains.

Asimov’s laws of robotics

Artificial Intelligence (AI) and Machine Learning (ML) have emerged as transformative forces in the world of technology. These fields, which strive to replicate and enhance human-like intelligence in machines, have seen remarkable advancements and found applications in various domains. Yet, as AI and ML systems continue to evolve, ethical considerations become increasingly vital. It’s a topic that resonates with science fiction author Isaac Asimov’s enduring and thought-provoking Three Laws of Robotics. These laws, penned decades ago, offer timeless insights into the ethical and societal dimensions of AI, guiding us in ensuring that the machines we create align with our values and principles.

  1. A robot may not injure a human being or, through inaction, allow a human being to come to harm: This fundamental law underscores the paramount importance of ensuring the safety of humans. In AI and ML, this principle manifests in the development of safety measures and ethical guidelines to prevent AI systems from causing harm, whether intentionally or unintentionally.

  2. A robot must obey the orders given it by human beings, except where such orders would conflict with the First Law: The second law emphasizes the importance of human control over AI systems. It prompts us to design AI and ML systems that respect human authority and can be guided by human commands, all while adhering to the primary rule of not causing harm.

  3. A robot must protect its own existence as long as such protection does not conflict with the First or Second Law: Asimov’s third law introduces a sense of self-preservation to robots. In AI terms, this can be likened to ensuring the robustness and reliability of AI systems, so they can fulfill their intended tasks effectively while maintaining safety and security.

These laws, though fictional, serve as a valuable framework for contemplating the ethical implications of AI and ML. They encourage us to prioritize human well-being, emphasize the importance of human control, and underscore the need for resilient and dependable AI systems. As we navigate the future of AI and ML, these principles remain a guiding light, urging us to ensure that our creations align with the highest ethical standards.

See also