Tapping into the Temporal Structure of Developmental Data using Burstiness Analysis

• In Event: 1-044 - Finding Structure in Time: Computational Methods for Characterizing the Micro-dynamics of Interaction

Integrative Statement

From the production of a parent’s vocalizations to the visual experiences generated from an infant who just made their first unassisted steps, the environmental properties human infants perceive and the actions they produce encompass a rich collection of patterns and regularities that unfold over time. With the increasing ease to which developmental scientists are able to collect massive amounts of rich data from multiple information sources, comes the need for new methods to analyze and characterize the patterns and regularities of human behavior.

In the current talk, I will introduce methodological advancements for how developmental scientists can apply simple distributional analyses to time series data. Specifically, I will review analyses first introduced in physics (Goh & Barabasi, 2008), that characterize spike trains of human behavior in the dimension of burstiness of the spike train of interest. Burstiness is a distributional measure that provides an estimate of a system’s activity patterns spanning from periodic (B=-1), to random (B=0), to theoretically maximal burstiness (B=1). An increasing list of human phenomena has been observed to be characterized as more bursty relative to periodic or random, and recent work in developmental science has started to uncover what the consequences of perceiving and acting in an environment with bursty properties are for the learning and attentional systems of the developing human.

I will focus on the key principles of applying the burstiness analysis to developmental data and will walk the audience through open-access code written in R and Matlab using simple examples. Several applications of the burstiness analysis on developmental data will be presented. In the first example, I will apply the burstiness analysis to a dataset of ego-centric images generated from infant head cameras to show how the burstiness metric can characterize the temporal structure of visual experiences of early infancy. In the second example, I will show that the temporal structure that the burstiness analysis measures has consequences for human learning, by applying the analysis to a dataset of parental utterances in naturalistic infant-parent toy play sessions. By the end of the presentation, the audience will have the necessary background to motivate using the burstiness analysis on their own datasets and also the practical knowledge of how to apply the open-access computer code to implement the analyses.

Author

• Drew Abney, Indiana University
  Presenting Author