Multi-Integration of Labels across Categories for Component Identification (MILCCI)

Area of Science:

• Data Science
• Computational Neuroscience
• Time-Series Analysis

Background:

• Large-scale temporal datasets are common across scientific fields, often linked to metadata variables.
• Analyzing how these labels influence time-series data and disentangling category-specific effects presents a significant challenge.

Purpose of the Study:

• To present MILCCI, a novel data-driven method for analyzing multi-trial temporal data.
• To identify interpretable components, capture cross-trial variability, and integrate label information to understand category representations.

Main Methods:

• MILCCI employs a sparse per-trial decomposition, leveraging label similarities within categories.
• It allows for subtle, label-driven adjustments in component compositions across trials.
• The method learns temporal traces for each component, which evolve within trials and vary across trials.

Main Results:

• MILCCI successfully identifies interpretable components and distinguishes the contribution of each metadata category.
• The method demonstrates effective disentanglement of label effects in complex temporal data.
• Performance is validated using synthetic data and real-world examples from diverse fields.

Conclusions:

• MILCCI offers a powerful approach for analyzing labeled temporal data, enhancing understanding of underlying structures and category-specific influences.
• The method provides a framework for robust time-series analysis in neuroscience, social sciences, and beyond.