We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

stat.ML

Change to browse by:

References & Citations

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Statistics > Machine Learning

Title: Independent mechanism analysis, a new concept?

Abstract: Independent component analysis provides a principled framework for unsupervised representation learning, with solid theory on the identifiability of the latent code that generated the data, given only observations of mixtures thereof. Unfortunately, when the mixing is nonlinear, the model is provably nonidentifiable, since statistical independence alone does not sufficiently constrain the problem. Identifiability can be recovered in settings where additional, typically observed variables are included in the generative process. We investigate an alternative path and consider instead including assumptions reflecting the principle of independent causal mechanisms exploited in the field of causality. Specifically, our approach is motivated by thinking of each source as independently influencing the mixing process. This gives rise to a framework which we term independent mechanism analysis. We provide theoretical and empirical evidence that our approach circumvents a number of nonidentifiability issues arising in nonlinear blind source separation.
Comments: NeurIPS 2021 final camera-ready version
Subjects: Machine Learning (stat.ML); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
Cite as: arXiv:2106.05200 [stat.ML]
  (or arXiv:2106.05200v3 [stat.ML] for this version)

Submission history

From: Luigi Gresele [view email]
[v1] Wed, 9 Jun 2021 16:45:00 GMT (2542kb,D)
[v2] Thu, 28 Oct 2021 08:55:25 GMT (2613kb,D)
[v3] Wed, 9 Feb 2022 17:42:22 GMT (2638kb,D)

Link back to: arXiv, form interface, contact.