Analysis Primitives

Analysis primitives are operators applied across layers to extract meaningful patterns from features.
They summarize, transform, or model data at various temporal and spatial scales.

Example of Analysis Primitives

Primitive Type	Description	Implemented
Statistical Moments	Unary operators summarizing distributions (mean, variance, skewness, kurtosis).
Shape Descriptors	Peaks, slopes, valleys in time-series; geometric descriptors of movement curves.
Entropy & Complexity	Approximate/sample entropy, recurrence analysis; quantify predictability or irregularity.
Time-Frequency Transforms	Fourier or wavelet transforms to detect rhythm, periodicity, or temporal structures.
Symmetry [^1]	Unary/binary operators measuring geometric or dynamic balance (e.g., left vs. right entropy or energy).
Synchronization [^2]	Binary/n-ary operators measuring alignment of signals (cross-correlation, phase-locking, group entrainment).
Causality	Directional relationships (e.g., Granger causality, transfer entropy) to detect leader–follower dynamics.
Clusterability ¹	Measures the tendency of data points to form clusters by means of the Hopkins statistics.
Predictive Models	Hidden Markov Models, classifiers, neural networks; used for gesture segmentation or quality inference.
Physical Models	Biomechanical analogies (mass–spring–damper systems) to capture dynamics such as Fluidity.
Saliency / Rarity ²	Detecting unusual occurrences in movement with respect to most frequent patterns.

References

Corbellini, N., Ceccaldi, E., Varni, G., & Volpe, G. (2022, August). An exploratory study on group potency classification from non-verbal social behaviours. In International Conference on Pattern Recognition (pp. 240-255). Cham: Springer Nature Switzerland. ↩
Niewiadomski, R., Mancini, M., Cera, A., Piana, S., Canepa, C., & Camurri, A. (2019). Does embodied training improve the recognition of mid-level expressive movement qualities sonification?. Journal on Multimodal User Interfaces, 13, 191-203. ↩