EAISI PEASI Lab

The Eindhoven AI Systems Institute (EAISI) aims to consolidate all AI research at TUE. Within that institute, we run a lab for Personal Informatics (P…I) Experiments At Scale (…EAS…). The letters of those lab characteristics lead to the name “PEASI”. The aim of the lab is to enable large scale experiments by providing a pooled set of personal informatics tools such as pedometers, heart rate monitors, and ESM tools. Physical devices are be pooled across a multitude of TUE projects and their data is be stored exclusively on TUE-managed server resources.

Rationale: No AI without Data!

Before EAISI PEASI, most European knowledge institutes approached data collection with wearables quite ad-hoc, keeping some tens of devices of different types stored on a project-by-project basis.

At the same time:

  • big vendors of affordable wearables partnered with knowledge institutions on inspiring initiatives like “All of Us“, pooling consumer data for research purposes. Such approaches make undesirable compromises with regard to inclusiveness at the level of the study participants (i.e., how to involve also those who don’t own wearables (yet)?) as well as at the level of scientific openness (i.e., how to enable open sharing of raw data, how to customize prompts on the wearables, how to run experimental AI on the devices?);
  • select vendors offered research toolkits for high-end devices, offering rich data access and customization options, but at price ranges that would make experimentation at large scale rather unrealistic.

At TUE, we envisioned creating a bottom-up European alternative marrying the best of both worlds. Therefore, we kick-started EAISI PEASI with a pool of 200 affordable wearables from which we could still extract the raw data and experiment with custom AI-powered prompting strategies. We always envisioned adding other types of personal informatics tools (e.g., smart toilets) later, but for now, the focus on wearables gives good momentum.

History

The lab was launched officially at an EAISI Café session in June 2020. From the start, the lab has been supervised by a steering committee consisting of senior scholars from multiple TUE departments. Operationally, however, it was initially supported by only one senior academic and one PhD candidate. Over time, EAISI PEASI was aligned better with other TUE labs. In particular, bookkeeping watch availability, storing wearables in between wearables, etc., is carried out by dedicated lab personnel from the department of Industrial Engineering. Until Today, we are trying to optimize our processes to let scientists and support staff focus on what they are best at.

Design Goals of Data Infrastructure

The following ten requirements reflect roughly what the lab aims to deliver from the data management side:

  1. a stable back-end should support any type of data (GPS, PPG, steps, …
  2. a stable back-end should support any type of device (mobile, wearable, …)
  3. the back-end should have interfaces to web APIs of big vendors (Fitbit, Strava, …)
  4. the lab managers should have no monopoly on integration development (instead, the infrastructure should offer portals, APIs, or SDKs)
  5. the infrastructure should support radical experimentation (research!), but safely (responsible!)
  6. data should be handled as much as possible nationally or within the EU
  7. the infrastructure should showcase citizen-centric principles beyond GDPR rights
  8. the infrastructure should support standards-based exports (e.g., HL7 FHIR)
  9. the infrastructure should demonstrate novel mechanisms for supporting open science even when original data is confidential by nature (e.g., via synthetic data generation methods)
  10. device logistics should be managed with workflow support, for the sake of scaleability.

These goals are partially fulfilled by relying on TUE’s GameBus technology, and more precisely by the open-source Experiencer framework.

Current State

The lab has supported dozens of studies, generating more than a million events, each with multiple event characteristics (e.g., timestamp, accelerometer and PPG-based time series, self-reported mood, etc.). The PhD manuscript of Alireza Khanshan provides various usage charts for EAISI PEASI-based studies until 2024.

Our current engineering and budget challenges relate to migrating the open source Experiencer framework to Wear OS, since the original Tizen-based wearables had become obsolete.

Advisory Committee

  • Prof. dr. Ph. A.E.G. (Philippe) Delespaul, full professor in Innovation in Mental Health Care at the department of Psychiatry and Neuropsychology at the University of Maastricht;
  • Prof. dr. Panos Markopoulos, full professor of Design for Behavior Change at the department of Industrial Design at Eindhoven University of Technology;
  • Prof. dr. Wim Nuijten, full professor of AI and Operations Research at the department of Mathematics and Computer Science, and Scientific Director of Eindhoven Artificial Intelligence Systems, at Eindhoven University of Technology;
  • Prof. dr. Natal van Riel, full professor of Biomedical Systems Biology at the department of Biomedical Engineering at Eindhoven University of Technology.
Top of page