Predicting Infectiousness for Proactive Contact Tracing
Authors: Yoshua Bengio, Prateek Gupta, Tegan Maharaj, Nasim Rahaman, Martin Weiss, Tristan Deleu, Eilif Benjamin Muller, Meng Qu, victor schmidt, Pierre-Luc St-Charles, hannah alsdurf, Olexa Bilaniuk, david buckeridge, gaetan caron, pierre luc carrier, Joumana Ghosn, satya ortiz gagne, Christopher Pal, Irina Rish, Bernhard Schölkopf, abhinav sharma, Jian Tang, andrew williams
ICLR 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | In this paper, we use a recently-proposed COVID-19 epidemiological simulator to develop and test methods that can be deployed to a smartphone to locally and proactively predict an individual’s infectiousness (risk of infecting others) based on their contact history and other information, while respecting strong privacy constraints. Predictions are used to provide personalized recommendations to the individual via an app, as well as to send anonymized messages to the individual’s contacts, who use this information to better predict their own infectiousness, an approach we call proactive contact tracing (PCT). Similarly to other works, we find that compared to no tracing, all DCT methods tested are able to reduce spread of the disease and thus save lives, even at low adoption rates, strongly supporting a role for DCT methods in managing the pandemic. Further, we find a deep-learning based PCT method which improves over BCT for equivalent average mobility, suggesting PCT could help in safe re-opening and second-wave prevention. 4 EXPERIMENTS |
| Researcher Affiliation | Academia | Yoshua Bengio*af Prateek Gupta*abh Tegan Maharaj*ag Nasim Rahaman*ac Martin Weiss*ag Tristan Deleu af Eilif Muller af Meng Qu ai Victor Schmidt a Pierre-Luc St-Charles a Hannah Alsdurf d Olexa Bilanuik a David Buckeridge e Gáetan Marceau Caron a Pierre-Luc Carrier a Joumana Ghosn a Satya Ortiz-Gagne a Chris Pal a Irina Rish af Bernhard Schölkopf c Abhinav Sharma e Jian Tang ai Andrew Williams a (*,, Equal contributions, alphabetically sorted; a Mila, Québec; b University of Oxford; c Max-Planck Institute for Intelligent Systems Tübingen; d University of Ottawa; e Mc Gill University; f Université de Montreal; gÉcole Polytechnique de Montreal; h The Alan Turing Institute; i HEC Montréal) |
| Pseudocode | No | The paper includes a 'PCT model architecture' diagram (Figure 2) that visually represents the data flow and components, but it does not contain an explicit pseudocode block or a clearly labeled algorithm block. |
| Open Source Code | No | Details on the simulator can be found in Gupta et al. (2020), and the code is open-source at https://github.com/mila-iqia/COVI-Agent Sim. This refers to the COVI-Agent Sim simulator used in the paper, not the specific implementation of the Proactive Contact Tracing (PCT) deep learning models developed by the authors. |
| Open Datasets | Yes | We opt to use COVI-Agent Sim (Gupta et al., 2020), which features an agent specific virology model together with realistic agent behaviour, mobility, contact and app-usage patterns. With the simulator, we generate large datasets of O(10^7) samples comprising the input and target variables defined in sections 2.1 and 3.1. |
| Dataset Splits | Yes | We use 200 runs for training and the remaining 40 for validation (full training and other reproducibility details are in Appendix 5). |
| Hardware Specification | No | Our ML experiments use approximately 250 days training time on GPUs while simulations required approximately 41 days of CPU time. All CPU time was run on compute using renewable resources. This project could not have been completed without the resources of MPI-IS cluster, Compute Canada & Calcul Quebec, in particular the Beluga cluster. While general hardware types (GPUs, CPUs) and computing resources (MPI-IS cluster, Compute Canada, Calcul Quebec, Beluga cluster) are mentioned, specific models or detailed specifications are not provided. |
| Software Dependencies | No | The paper mentions the use of 'COVI-Agent Sim' but does not provide specific version numbers for any software dependencies or libraries used in their implementation or experiments. |
| Experiment Setup | Yes | The model was trained for 160 epochs on a domain randomized dataset (see below) comprising 10^7 samples. We used a batch-size of 1024, resulting in 80k training steps. The learning rate schedule is such that the first 2.5k steps are used for linear learning-rate warmup, wherein the learning rate is linearly increased from 0 to 2e-4, followed by a cosine annealing schedule that decays the learning rate from 2e-4 to 8e-6 in 50k steps. (Section 5.2 also details domain randomization parameters: Adoption rate [30 60], Carefulness [0.5 0.8], Initial proportion of exposed people [0.002, 0.006], Oracle additive noise [0.05 0.15], Oracle multiplicative noise [0.2 0.8], Global mobility scaling factor [0.3 0.9], Symptoms dropout, Symptoms drop-in, Quarantine dropout (test), Quarantine dropout (household), All-levels dropout) |