Causal machine learning

As explained in Causal AI

A causal machine learning model combines machine learning with representations of causal relationships to answer intervention or counterfactual questions.

An ad model may predict who will buy after seeing an impression. A causal machine learning model asks a different question: how would changing the impression affect engagement and, in turn, purchases? It combines learned relationships with a representation of what causes what, so predictions can support causal queries rather than only describe observed patterns. section 1.3.3 section 1.5.3

Figure 1.2 A model that captures how ad impressions drive engagement, which in turn drives purchases. This model decomposes into {ad impression, engagement} and {engagement, purchase}.

The ad-impression → engagement → purchase diagram shows how a causal structure separates one prediction problem into linked components. section 1.3.3

A practical workflow starts by stating the causal question, then representing possible causes mathematically, and finally fitting the model to data. A directed acyclic graph (DAG) can supply the structure. Its factors, or causal Markov kernels, can be implemented separately, composed into a probabilistic model, and fitted with standard statistical learning methods. For example, a kernel P(t | o, r) returns a probability for a queried value of T given O and R; the implementation can also generate values. section 1.5.3 section 3.1.7

The causal structure matters beyond answering interventions. Separating direct causes and effects makes the model decomposable; capturing a relationship that remains stable across settings may also help when deployed data drifts. Neither benefit removes the need to check assumptions: a causal answer must be identified by the assumptions and available data. section 1.3.3 section 3.1.7

How Causal AI covers it

How deep it goes

Outlines a question-to-model workflow and explains DAG-based kernel composition, while deferring detailed causal-model development.

Examples

Contrasts ad-impression purchase models, illustrates drift with elevation and temperature, and sketches a transportation-model factorization.

What it assumes

Assumes some familiarity with predictive machine learning while identifying mathematics and machine learning foundations for review.

What you can do after

Build it from scratch by composing causal Markov kernels on a DAG and fitting their parameters to data.

How it connects

Connects causal inference, directed acyclic graphs, probabilistic machine learning, causal invariance, and data drift.

liveBookmanning.com

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

Related concepts

• language modelcontrasts with · A language model is a computational model that represents patterns in linguistic tokens to estimate sequence probabilities and predict or generate language from context.
• counterfactual reasoningimplemented by · Counterfactual reasoning examines hypothetical outcomes under conditions, actions, or policies different from those that actually occurred.
• causal graphical modelconfigured by · A causal graphical model is a probabilistic or structural model whose graph represents causal relationships and whose mechanisms support reasoning about interventions.
• generative modelinguses · Generative modeling is a machine-learning approach that learns patterns or a data distribution to generate new samples resembling the data.

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.