Classifying artificial-intelligence technologies

[lpetrich]

AI Knowledge Map: how to classify AI technologies – Francesco Corea – Medium

My lists are combinations of the article text and the main diagram.

By problem domain:

• Perception: the ability to transform raw sensorial inputs (e.g., images, sounds, etc.) into usable information
• Reasoning: the capability to solve problems
• Knowledge: the ability to represent and understand the world
• Planning: the capability of setting and achieving goals
• Communication: the ability to understand language and communicate

By paradigm:

• Symbolic
  • Logic-based tools: tools that are used for knowledge representation and problem-solving
  • Knowledge-based tools: tools based on ontologies and huge databases of notions, information, and rules
• Statistical
  • Probabilistic methods: tools that allow agents to act in incomplete information scenarios
  • Machine learning: tools that allow computers to learn from data
    • Supervised
    • Unsupervised
    • Reinforcement learning
• Subsymbolic
  • Embodied intelligence: engineering toolbox, which assumes that a body (or at least a partial set of functions such as movement, perception, interaction, and visualization) is required for higher intelligence
  • Search and optimization: tools that allow intelligently searching through many possible solutions

Author Francesco Corea also notes Pedro Domingos's classification of "AI Tribes", groups of researchers who prefer certain approaches.

• Symbolists: logic-based
• Connectionists: machine learning
• Evolutionaries: search and optimization
• Bayesians: probabilistic methods
• Analogizers: knowledge-based methods

The remaining one is embodied intelligence.

FC classifies a very long list of AI technologies, ranging from expert systems to natural-language processing to artificial neural networks to evolutionary algorithms.

 

[lpetrich]

There is a simpler classification: top-down vs. bottom-up.

• Top-down: explicit inference rules stated by the system designers
• Bottom-up: implicit inference rules worked out by learning

AI started with the top-down approach, and it has had some successes, like high-level programming-language translation, computer algebra, and expert systems.

But such an approach is difficult for many problems, and as computers got more and more powerful, bottom-up approaches like statistical fitting became more and more practical. "Deep Learning" is simple bottom-up AI on a massive scale. But bottom-up AI can have the problem of the resulting AI models being difficult to interpret.

 

[lpetrich]

A common part of AI practice is machine learning, and there are numerous algorithms for that. Which machine learning algorithm should I use? - The SAS Data Science Blog The author states in it that "Several data scientists I talked with said that the only sure way to find the very best algorithm is to try all of them."

It has a flowchart, and I will translate it into text.

Dimension Reduction?
. Yes
. . Unsupervised Learning: Dimension Reduction
. No
. . (can use output of dimension reduction)
. . Have Responses?
. . . Yes
. . . . Predicting Numeric?
. . . . . Yes
. . . . . . Supervised Learning: Regression
. . . . . No
. . . . . . Supervised Learning: Classification
. . . No
. . . . Unsupervised Learning: Clustering

Supervised learning means that inputs have intended outputs.
Unsupervised learning means lack of such outputs. Instead, one wants to sort out one's inputs.

Unsupervised Learning: Dimension Reduction
Topic Modeling?
. Yes
. . Probabilistic?
. . . Yes
. . . . Latent Dirichlet Analysis
. . . No
. . . . Singular Value Decomposition
. No
. . Principal Component Analysis

Unsupervised Learning: Clustering
Hierarchical?
. Yes
. . Hierarchical
. No
. . Need to specify k? (how many clusters)
. . . Yes
. . . . Categorical Variables?
. . . . . Yes
. . . . . . K-modes
. . . . . No
. . . . . . Prefer Probability?
. . . . . . . Yes
. . . . . . . . Gaussian Mixture Model
. . . . . . . No
. . . . . . . . K-means
. . . No
. . . . DBSCAN (density-based spatial clustering)

Supervised Learning: Regression
Speed or Accuracy?
. Speed
. . Decision Tree
. . Linear Regression
. Accuracy
. . Random Forest
. . Neural Network
. . Gradient Rooting Tree

Supervised Learning: Classification
Speed or Accuracy?
. Speed
. . Explainable?
. . . Yes
. . . . Decision Tree
. . . . Logistic Regression
. . . No
. . . . Data Is Too Large?
. . . . . Yes
. . . . . . Naive Bayes
. . . . . No
. . . . . . Linear SVM
. . . . . . Naive Bayes
. Accuracy
. . Kernel SVM
. . Random Forest
. . Neural Network
. . Gradient Rooting Tree

 

[untermensche]

Human intelligence decides what is important.

Machines are tools not a fellow intelligence.