Discovering Synchronized Subsets of Sequences: A Large Scale Solution.

Area of Science:

• Computer Vision
• Pattern Recognition
• Computational Biology
• Neuroscience
• Finance

Background:

• Identifying correlated time series is crucial for analyzing complex data across various scientific domains.
• Existing maximal clique algorithms are effective but lack scalability for large datasets.
• The need for efficient methods to detect synchrony in large-scale sequence data is significant.

Purpose of the Study:

• To develop a highly efficient and scalable method for finding the largest subset of correlated sequences.
• To provide a theoretically grounded approach for identifying synchronized time series.
• To offer a faster alternative to maximal clique algorithms for large dataset analysis.

Main Methods:

• Representing the search space as a union of epsilon-expanded clusters.
• Utilizing Jung's theorem by fitting a Euclidean ball on these clusters to find synchronized sets.
• Developing an approximate but efficient and scalable algorithm.

Main Results:

• The proposed method achieves results comparable to maximal clique algorithms.
• The new method is up to 300 times faster than maximal clique algorithms.
• Speed improvements increase exponentially with the number of input sequences.

Conclusions:

• The developed method offers a significant speedup for detecting synchronized sequences in large datasets.
• This approach is validated across diverse domains including behavior analysis, finance, and neuroscience.
• The epsilon-expanded cluster method provides a scalable solution for pattern recognition and time series analysis.