Causal Inference and Causal Learning for AI: Interventions, Identification, Counterfactuals, and Causal Decisions - Brossura

Wang, Guangyu

Brossura

ISBN 13: 9798907070431

Casa editrice: BestAI Press, 2026

Riassunto

Modern AI systems are increasingly asked not only to predict, but to decide what will happen under intervention, which changes actually cause better outcomes, and whether evidence collected in one setting will remain valid in another. Causal Inference and Causal Learning for AI develops the mathematical foundations needed to answer those questions rigorously.

Organized around an identification-first framework, the book unifies potential outcomes and structural causal models through a common chain: causal question, estimand, assumptions, identification, estimation, diagnostics, sensitivity, transport, and decision. It develops randomized and platform experiments, nonparametric identification, doubly robust and orthogonal machine learning, heterogeneous treatment effects, policy learning and off-policy evaluation, instrumental variables and quasi-experiments, partial identification and sensitivity analysis, longitudinal treatment, mediation, transportability, data fusion, causal discovery, and causal representation learning.

Throughout, the mathematics is tied directly to the construction and deployment of modern AI systems: data filtering and synthetic data, post-training interventions, model and reward evaluation, online rollout, model-generated labels and judges, routing policies, long-horizon agents, tool use, and mechanistic interventions inside trained networks. The emphasis is not on a catalogue of algorithms, but on a more durable question: what causal claim is actually determined by the available data and assumptions, and what remains valid when that claim is turned into an AI decision?

Written for graduate students, researchers, and technically oriented practitioners, this volume provides a compact but rigorous bridge between classical causal inference, modern causal machine learning, and the decision problems that arise across the AI lifecycle.

Dettagli dell'edizione

• Editore: BestAI Press
• Data di pubblicazione: 2026
• Lingua: Inglese
• ISBN 13: 9798907070431
• Rilegatura: Paperback
• Numero di pagine: 191
• Contatto del produttore: non disponibile
• Persona responsabile: non disponibile