Inversion-free method evaluated for high-dimensional image counterfactuals
Original title: Inversion-Free High-Dimensional Counterfactual Inference
On CelebA and ADNI, the authors compare inversion-based and inversion-free methods on effectiveness and identity preservation.
Counterfactual inference answers what-if queries from observational data, a causal graph, and modeling assumptions. Extending scalar methods to images usually inverts a factual image to a latent state, updates attributes, and renders a counterfactual. The authors question whether inversion is necessary and evaluate an inversion-free method adapted from image editing on CelebA and ADNI, reporting tradeoffs in effectiveness and identity preservation.
- Conventional pipelines invert the factual image, update attributes, then render.
- The inversion-free counterpart is repurposed from the image-editing literature.
- Benchmarks use the CelebA and ADNI image datasets.
- Effectiveness measures faithful rendering of the counterfactual attribute.
- Identity preservation checks that the same unit is rendered.
From the original article
1 min read at arxiv.orgCounterfactual inference is the task of answering hypothetical "what-if" queries about a unit, given an observational dataset, modeling assumptions, and a causal graph. This enables a modeler to reason about hypothetical scenarios that cannot otherwise be answered without assumptions or a causal graph.
By Saketh Vishnubhatla · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.