Quantitative Predictive Monitoring and Control for Safe Human-Machine Interaction
Authors: Shuyang Dong, Meiyi Ma, Josephine Lamp, Sebastian Elbaum, Matthew B. Dwyer, Lu Feng
AAAI 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We apply the proposed approach to two case studies: Type 1 Diabetes management and semi-autonomous driving. Experiments show that the proposed approach improves safety and effectiveness in both case studies. |
| Researcher Affiliation | Collaboration | Shuyang Dong1, Meiyi Ma2, Josephine Lamp3, Sebastian Elbaum1, Matthew B. Dwyer1, Lu Feng1 1University of Virginia 2Vanderbilt University 3Dex Com EMAIL, EMAIL, EMAIL |
| Pseudocode | Yes | Algorithm 1 STL-U quantitative monitoring algorithm Algorithm 2 Adapting a Basal-Bolus Controller |
| Open Source Code | No | The paper mentions using third-party simulators like the UVA/PADOVA T1D Patient Simulator (Man et al. 2014) and the CARLA simulator (CARLATeam 2023), but it does not provide any explicit statement or link for the source code of the methodology described in this paper. |
| Open Datasets | No | We evaluate the proposed approach via experiments using the UVA/PADOVA T1D Patient Simulator (Man et al. 2014)... We use the simulator to generate data based on 30 virtual patient profiles including: 10 adults, 10 adolescents and 10 children. The paper describes generating data from a simulator rather than using a publicly available dataset, and does not provide access to the generated data. |
| Dataset Splits | Yes | We use the simulator to generate data based on 30 virtual patient profiles including: 10 adults, 10 adolescents and 10 children. Each patient is simulated for an 85-day period, with each time step in the simulation representing 3 minutes. We use 70-day, 5-day, and 10-day data for training, validation, and testing, respectively. We segment the data into samples of 20 steps (1 hour) by sliding windows, and obtain about 413,326 samples in total for each patient population. |
| Hardware Specification | Yes | The experiments were run on a machine with 2.1GHz CPU, Nvidia Quadro RTX5000 GPU, 128GB memory, and Cent OS 7 operating system. |
| Software Dependencies | No | The paper mentions "Cent OS 7 operating system" but does not provide specific version numbers for other key software components, such as programming languages, libraries (e.g., PyTorch, TensorFlow), or frameworks used for the implementation. |
| Experiment Setup | Yes | Each model is trained for 50 epochs. Given a chosen SRT and dropout rate, we repeat Bayesian LSTM predictions for 30 times using the Monte Carlo method. We estimate Gaussian distributions using these predictions and obtain flowpipes under a 95% confidence level. For the LSTM models of adults, Bernoulli drop Connect with a dropout rate of 0.8 is the best choice. For adolescents and children, Bernoulli drop Connect with a rate of 0.9 and Gaussian Dropout with a rate of 0.9 were found to be best, respectively. We set these BG thresholds and basal change percentages following medical guidelines (American Diabetes Association 2022). |