HN Explorer

AI-Based Optimization of Non-Pharmaceutical Interventions for the Covid Pandemic

6 points by thedeepone ↗ HN
The dashboard demonstrates how Evolutionary AI could be used to model the potential effects of non-pharmaceutical intervention (NPI) strategies to contain and mitigate the pandemic. The predictor is trained with historical data on the number of cases and the NPIs over time in various countries, i.e. restrictions on schools and workplaces, public events and gatherings, and transportation. A Pareto front of Prescriptors is then evolved to discover the best tradeoffs between minimizing cases and restrictions. To illustrate this principle, the site includes an interactive demo: you can explore how, given your preferred tradeoff, the pandemic could be contained and mitigated in different countries.

See the demo here: https://evolution.ml/esp/npi/ Download the full paper here: https://arxiv.org/abs/2005.13766

2 comments

[ 3.7 ms ] story [ 13.7 ms ] thread
rpm13
It has been interesting to follow the prescriptions of this demo in the past few weeks. In May, the system was prescribing less stringent NPIs that it nevertheless predicted to contain the pandemic. Now its recommendations are much more stringent---apparently it has learned that people no longer adher to the NPIs the way they used to! This indeed seems to be the case, anecdotally and based on news media, but it is interesting that the AI learned it on its on from the data.
ekmey
Nice thing is, now that this data exists, the models should never need to rediscover this relaxed adherence phenomenon, e.g., moving forward with covid, or for the next pandemic.