Joint Tensor Alignment and Coupled Factorization

Explore joint tensor alignment and coupled factorization in this 50-minute conference talk from the Chemometrics & Machine Learning in Copenhagen group. Delve into the challenges of analyzing coupled tensor datasets with unknown correspondence between entities. Learn about an algorithm that simultaneously aligns tensors and performs coupled factorization, outperforming multi-stage approaches. Examine two formulations and their trade-offs, and review experimental results demonstrating the effectiveness of this joint treatment. Discover applications in synthetic and real datasets, including clustering and factor matching. Gain insights into the flexibility and uniqueness of the proposed approach, as well as its implications for shifted factor analysis. The talk, presented by Vagelis Papalexakis, is based on research presented at the IEEE International Conference on Data Mining (ICDM) 2022.

Introduction
Motivation
Coupled Factorization
Graph Alignment
General Problem Definition
Loss Functions
Optimization
Experiments
Synthetic Data Sets
Real Data Sets
Clustering
Factor Matching Score
Proposed Approach
Proposed Results
Conclusion
Flexibility
Uniqueness
Results
Shifted Factor Analysis
Wrap Up