Artificial Intelligence

Artificial intelligence (AI) is the intelligence by machines or software.

Subfields of AI

Subfields of AI:

  • Knowledge Represntation and Reasoning (KR&R)
    • Symbolic AI
  • Machine Learning (ML)
    • Parametric AI
    • Natural Language Processing (NLP)
    • Deep Learning
  • Large Language Model (LLM)

Knowledge Representation and Reasoning (KR&R)

Symbolic AI

Symbolic AI is a subpart of KR&R. It is based on how human intelligence is managed symbollically. It is an alternative to parametric AI.

Part of symbolic learning are:

  • Semantic Networks
  • Frames
  • Rules
  • Restrictions

Aproximated reasoning: fuzzy logic, MYCIN, Dempster-Shafer

MI, OPS-5, ART-IM, RETE algorithm

Semantic Networks

Semantic networks were developed by Quillian. You can read more about them at Wikipedia.

Frames

Frames were proposed by Minsky in 1976. You can read more about them at Wikipedia.

Rules

Restrictions

Machine Learning

Federated learning or collaborative learning is a machine learning technique in a setting where multiple entities (often called clients) collaboratively train a model while keeping their data decentralized, rather than centrally stored.

MLOps ease the implementation and machine learning model management in production.

You can find this post about machine learning frameworks.

Reinforcement Learning from Human Feedback (RLHF).

Data mining is quite related to artificial intelligence. You can read this post about data mining.

It can be:

  • Supervised
  • Unsupervised

Foundational model is an evolution of multiple models, where there is a single model for machine learning.

Data augmentation is adding scenarios.

An ensemble is having different models to solve a problem. Then the answer that is proposed that most models is the one used.

A model card is a document that explains a model and their risks.

Predictive AI

Predictive AI is…

Parametric AI

Parametric AI is based on how human intelligence works physically.

Artificial Neural Networks

Artificial neural networks can be considered a data mining method.

You can read this post about artificial neural networks.

Genetic Algorithms

Deep Learning

Deep learning is the use of multilayered artificial neural networks to learn from data.

It is a subset of machine learning.

It has two steps:

  1. Training
  2. Inference

The training part is based on a model with random weights, that receives and input and creates an output. The deviation in the output is called loss, that is used to update weights.

The inference is the same as traditional software, with an input that comes into a model and creates a model.

Software based on deep learning was referred as software 2.0 by Andrej Karpathy in 2017.

Deep learning can be considered that started by 2012.

GenIA

Generative IA (GenIA) can be considered a subpart of deep learning. It is an alternative to the more traditional Predictive IA.

GenIA consist of generating data.

It became popular in 2021 after the release of Generative Pre-trained Transformer 3 (GPT-3).

When an output is being generated, there are different possible options, where each

A log probability (logprob) is the probability to use an output using logarithmic functions. Logarithmic functions are used instead of percentages because they are more easily computed.

Recover, Amplify, Generate (RAG) is…

Stable Diffusion is a GenIA algorithm to generate images.

Stable Diffusion at Wikipedia

Large Language Model (LLM)

Large Language Models (LLM) is a language model notable for its ability to achieve general-purpose language understanding and generation.

You can read more about it on this post.

Agentic IA

An IA agent is a system that receives information and takes decision autonomously.

Agentic IA implies the use of IA agents in system.

Distributed AI

Distributed AI goals:

  • Ensuring agents acts coherently when taking decision or making actions.
  • Enabling agent reasoning about other agents actions and plans.
  • Developing platforms for multiagent systems and development methodologies.

Natural Language Processing (NLP)

Natural Language Processing (NLP) helps computers communicate with humans in their own language and scales other language-related tasks.

Task of NLP include text classification, sentiment analysis, machine translation, named entity recognition, part-of-speech tagging and question answering.

There are different approaches to NLP. The earliest ones (1970s) were rule-based or statistical-based, and are used for parsing, tokenization, or syntactic analysis. The more modern are based on machine learning and LLMs.

Types of AI

Types of AI:

  • Decision support systems
  • Expert systems
  • Knowledge bank
  • Neural networks

Artificial Intelligence Security

There are some security concerns related to AI.

You can read more about artificial intelligence security on this post.

Artificial Intelligence Standards

Artifical Intelligence standards:

  • ISO/IEC 42001

ISO/IEC 42001

ISO/IEC 42001 has the title “Information Technology – Artificial Intelligence – Management System”.

As of 2024, latest version is ISO/IEC 42001:2023.

OECD IA

OECD IA is an intergovernmental standard.

OECD IA official website

Artificial Intelligence Organizations

Agencia Española de Supervisión de Inteligencia Artificial (AESIA) is the public national agency on IA of Spain.

Artificial Intelligence Certifications

Artificicial Intelligence Certfications for professionals:

  • IBM AI Engineering Professional Certificate
  • CertNexus Certified Artificial Intelligence Practitioner Professional Certificate

Artificial Intelligence Courses

UNED’s Máster Universitario de Investigación en Inteligencia Artificial

UNED’s Máster Universitario de Investigación en Inteligencia Artificial

AI Model Repositories

AI model repositories featured on this post:

  • Hugging Face

Hugging Face

Hugging Face is a website where AI models can be shared.

Hugging Face official website

You might also be interested in…

Leave a Reply

Your email address will not be published. Required fields are marked *