Enhancing the Cloud-Readiness of Perceptual Computing Through Data Standardization Software

点击翻译按钮获取中文摘要

ABSTRACT Behavioral science is in the midst of a paradigm shift away fromhuman judgment and toward rigorous perceptual computing (PC) and investment in human predictive analytics. As a part of three separate NIH R01s (including the Parent R01 [MH125958]), our team is developing open-source software tools to exquisitely measure social and emotional behavior, including tools to quantify facial movements, body actions, and language. Unfortunately, the field of perceptual computing currently has no agreed-upon standard for organizing, maintaining, and curating large audio-visual datasets, and the myriad data streams that accompany them (for example, simultaneous psychophysiology recordings from wearable devices). The absence of such a standard is a major impediment to innovation in perceptual computing, as it hinders the development of large -scale computational pipelines and algorithm optimizations that cloud applications require. The purpose of this Supplement is to directly address this problemthough the creation of acommon data structure ready for adoption and iteration by the field of PC, along with a set of basic tools for working with data that follow this structure. The parent R01 for this Supplement is an ideal context in which to develop such a standard, as it involves the collection of more than 3000 individual audio-visual files across two sites (The Children’s Hospital of Philadelphia and Baylor College of Medicine). In other words, the parent R01 is a microcosm of the larger challenge facing the field – how to effectively and seamlessly integrate separate datasets in ways that supports robust and highly replicable analysis paths. This Supplement to our R01 will address this challenge by developing an open-source Sensor Data Structure (SDS) – a data generation, storage, and basic processing standard for use by the perceptual computing community – along with open-source software tools and Container environments to generate and validate data. We propose to parallel the achievements by highly successful NIH -supported Biomedical Imaging Data Structure (BIDS), which was developed to address an analogous problem in the field of brain imaging (e.g., the need to harmonize and curate large multicenter datasets). Although the Parent R01 is staffed for collecting and analyzing data, this Supplement would provide three new deliverables, all implemented via the addition of software engineer with industry experience: 1) creation of the data structure, 2) creation of a Python module and Container environment for implementing the standard, and 3) posting and monitoring these two deliverables publicly on GitHub. This is acritical step toward our ultimate goal of developing PC tools that are “cloud-ready”, and in widespread use by the PC community.