In August 2022 at the conference on Uncertainty in Artificial Intelligence (UAI 2022), Prime Video presented a paper on Clustering a union of linear subspaces via matrix factorization and innovation search. The paper shows that the innovation pursuit clustering algorithm, which remarkably outperforms the former approaches, can be simplified into an efficient, closed-form, and non-iterative matrix factorization-based method.
In addition, it’s proven that in sharp contrast to most other methods, the performance of the simple matrix factorization-based method doesn’t primarily depend on the distance or incoherency between the span of clusters. It is shown through several theoretical results that the performance of the algorithm depends on the innovative components of the linear clusters that could be notably incoherent with each other even if the span of clusters is close.
UAI is one of the top international artificial intelligence and machine learning (AI/ML) conferences, with a focus on research related to knowledge representation, learning and reasoning in the presence of uncertainty, optimization, ML applications, and online learning.
For more information, see our Clustering a union of linear subspaces via matrix factorization and innovation search paper on the Amazon Science website.