GlucoSynth: Generating Differentially-Private Synthetic Glucose Traces
Authors: Josephine Lamp, Mark Derdzinski, Christopher Hannemann, Joost van der Linden, Lu Feng, Tianhao Wang, David Evans
NeurIPS 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We provide a comprehensive evaluation on the real-world utility of the data using 1.2 million glucose traces; Gluco Synth outperforms all previous methods in its ability to generate high-quality synthetic glucose traces with strong privacy guarantees. |
| Researcher Affiliation | Collaboration | 1University of Virginia, Charlottesville, VA, USA; 2Dexcom, USA jl4rj@virginia.edu; {mark.derdzinski; christopher.hannemann; joost.vanderlinden}@dexcom.com; {lu.feng; tianhao; evans}@virginia.edu |
| Pseudocode | No | The paper describes methods in prose and with diagrams (e.g., Figure 3, Figure 4) but does not include a formally labeled 'Pseudocode' or 'Algorithm' block. |
| Open Source Code | No | The paper provides links to implementations of benchmark models (e.g., 'Time GAN [4] is implemented from www.github.com/jsyoon0823/ Time GAN'), but does not explicitly state that the source code for Gluco Synth, the methodology described in this paper, is publicly available. |
| Open Datasets | No | In order to train on a huge set of glucose traces, we used a private dataset, not publicly available (one of the motivations for this project was actually to share a synthetic version of these traces). |
| Dataset Splits | No | The paper mentions, 'We use a separate validation dataset (not the set of original training traces) for all experimental results,' but does not provide specific details on the split percentages, sample counts, or methodology for creating this split from the 1.2 million traces mentioned. |
| Hardware Specification | Yes | Our experiments were completed in the Google Cloud platform on an Intel Skylake 96-core cpu with 360 GB of memory. |
| Software Dependencies | No | The paper mentions software like 'Tensorflow Privacy functions [24]' and 'Tensorflow DP Keras Adam Optimizer,' but does not specify their version numbers. |
| Experiment Setup | Yes | Throughout all our experiments we use Gluco Synth model parameters of α = 0.1 and η = 10 and a motif tolerance of σ = 2 mg/d L and motif length τ = 48. |