Causal factor

As explained in Causal AI

A causal factor is a variable or property that contributes to producing an effect in a system.

A causal factor is a variable or property that contributes to producing an effect. In a causal model, factors are treated as causes rather than merely predictive features. For an image, shape, scale, rotation, and position can independently determine visible properties; in decision systems, economic conditions can affect outcomes beyond the chosen action. section 9.4.1 section 12.1.2

Causal factors support several kinds of reasoning:

• Represent causes explicitly in a causal DAG, separating them from an image variable, utility, or other downstream effect. section 9.4.1 section 12.1.2
• Distinguish intervention questions from observational prediction: asking what would happen under a chosen action or changed factor requires causal reasoning. section 12.1.2
• Encode factors, infer image-specific exogenous variation, change one factor, and decode a counterfactual while preserving the inferred variation. section 9.4.1
• Treat representation learning cautiously: an unsupervised embedding may preserve reconstruction information without isolating causal factors. section 5.2.5

Changing a shape while preserving the scene

A dSprites image encodes shape, scale, rotation, and x- and y-position. Given an image of a square, infer its image-specific exogenous variable, replace the shape value with heart, and decode the intervened factor vector to generate the counterfactual image. section 9.4.1

Figure 9.18 The causal DAG for a dSprites image, displayed as a plate model to highlight the shape of N_I and I. N_i is the exogenous variable for the image. The model is trained with an encoder-decoder framework that uses a 50 × 1 dimensional image encoding to represent N_I.

Use the causal DAG to inspect how causal factors and image-specific exogenous variation combine to produce an image. section 9.4.1

How Causal AI covers it

How deep it goes

Explains factor vectors, DAG roles, encoding, intervention, and decoding, while deferring general identification theory.

Examples

Uses dSprites shape, scale, rotation, and position factors alongside economic conditions affecting investment outcomes.

What it assumes

Assumes basic causal graphs and machine-learning models while introducing factor vectors, one-hot encoding, and exogenous image variables operationally.

What you can do after

Build counterfactual image simulations by encoding factors, intervening on one value, sampling exogenous variation, and decoding the result.

How it connects

Connects causal factors to causal DAGs, interventions, counterfactuals, exogenous variables, utility, and causal representation learning.

liveBookmanning.com

Read Causal AI on liveBook Get the book at manning.com

Related concepts

• digital imageproduces · A digital image is visual content encoded as data, typically as a two-dimensional arrangement of pixels or equivalent visual elements.
• tensor shapeincludes · A tensor shape is an ordered sequence of dimension sizes specifying the extent of a tensor or array along each axis.
• resultproduces · A result is an actual or possible outcome, state, or consequence of an event, action, decision, or process.
• rotationincludes · A rotation is a geometric transformation that turns points or objects about a fixed center, origin, or axis through an angle while preserving distances and angles.
• coefficientincludes · A coefficient is a numerical multiplier that scales a quantity, term, or contribution in a mathematical or computational operation.

This explainer was generated by Manning's knowledge graph from the text of Causal AI. The section links point to the parts of the book it draws on.