Mobile sensing allows researchers to collect a lot of different data from a smartphone and its sensors (like the accelerometer or GPS). It also allows for connecting wearable devices (like a heart rate monitor) to the phone via Bluetooth. As such, smartphones are a valuable source for data collection in research studies.
-
In our lab, we have created the CARP Mobile Sensing Framework for running such mobile sensing studies. The configuration of what data to collect, when, and how is done using a rather sophisticated domain model, which is serialised and loaded as JSON. However, it is impossible for non-technical researchers (like medical doctors) to make this configuration file. Moreover, researchers do not know what type of data can be collected and the many different ways sensing can be configured.
-
The goal of the project is to enable a non-technical user to generate a valid JSON protocol for a CARP Mobile Sensing study. In particular, we want to create a CARP large language model (LLM) that can help configure sensing using a chat-based interface. This includes selecting and fine-tuning an LLM model for creating study protocols, validating them according to the existing CARP Core domain model, and creating a chat-based user interface for researchers to work with study protocols in an iterative manner. Optionally, the tool should also allow for simulating a protocol as part of an iterative refinement.
-
The project aims to:
- Create an LLM-based solution for generating CARP study protocols
- Develop an interaction model between the user and the LLM model
- Validate study protocols against the CARP Domain model
- Support the simulation of the execution of study protocols
-
This project will be conducted in the CARP Lab at the Digital Health section at DTU. Since you will be working as part of the CARP team and with the CARP software stack, you must be willing to sign a confidentiality and IPR contract (the so-called “G form” at DTU) to access the software.
-
Thesis Type: Technological.
Technical Skills: Python, ML libs, LLM models, and APIs
Research Skills: Data Science, ML, HCI / UX, JSON, Mobile Sensing
General information on Supervision: See https://www.bardram.net/msc-thesis/