Privacy-and-Utility-Aware Publishing of Schedules
Authors: Maike Basmer, Stephan A. Fahrenkrog-Petersen, Ali Kaan Tutak, Arik Senderovich, Matthias Weidlich
AAAI 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experiments with synthetic and real-world schedules demonstrate the feasibility, robustness, and effectiveness of our mechanism. We experimentally demonstrate the feasibility of our perturbation functions |
| Researcher Affiliation | Academia | 1Humboldt-Universit at zu Berlin, Unter den Linden 6, 10117 Berlin, Germany 2Weizenbaum Institute, Hardenbergstraßße 32, 10623 Berlin, Germany 3York University, 4700 Keele St, Toronto, ON M3J 1P3, Canada |
| Pseudocode | No | The paper describes algorithms and problem formulations (e.g., CSP in Section 3.3 and 6) but does not present them in a structured pseudocode or algorithm block format. |
| Open Source Code | Yes | The code is publicly available on Git Hub1. 1https://github.com/hu-dbis/privacy-scheduling-tools |
| Open Datasets | No | We rely on two datasets: (i) a synthetically generated set of schedules in a controlled setup, and (ii) a dataset drawn from real-world historical schedules of a health facility. The dataset bases upon jobs extracted from real patient scheduling data of an outpatient cancer hospital in the United States. No concrete access information is provided for either. |
| Dataset Splits | No | A dataset consisting of 1000 schedules was created by using randomly drawn scheduling parameters to construct new schedules. In total, we considered 5 days (1 working week in April 2021) and 6 medical floors, which amounts to 30 historical schedules in total. No specific training/test/validation splits or percentages are provided. |
| Hardware Specification | Yes | The experiments with synthetic data ran on a Dell R920 server with 1TB RAM, 4/60/120 CPUs/cores/threads at 2.5GHz, running open SUSE 15.3. The real-world data experiments ran on a laptop with an i712800H 2.40 GHz processor and 64.0GB RAM. |
| Software Dependencies | Yes | We used Python 3.10 along with several libraries, most importantly or-tools 9.4.1874 (Perron and Furnon 2022) for the ISA in the privacy loss computation. |
| Experiment Setup | Yes | We consider the values {0.01, 0.5} for the privacy parameter ϵ and {0.005, 0.02} for the utility threshold δ. For each utility function and each parameter combination (h, ϵ, δ), the PUP is applied to each schedule with the respective configuration. Each PUP instance is allocated a time budget of 5min. |