VDO uses virtual weights to speed up convergence to game equilibria
What happened
Ruilan Wang proposes the Virtual Double Oracle (VDO) method: using a bilinear identity, each newly added strategy is initialized with the MWU weights it would have had "if it had existed from the start," rather than being reset to a uniform prior. The author proves that, provided the cumulative cost of newly added strategies grows at most linearly with the number of strategies k (a condition supported by the empirical data), VDO achieves O(√(k/T)) exploitability as the strategy population grows. Building on this, the author also proposes a discounted dVDO. Experiments on matrix games, poker, and Goofspiel show that with a suitable discount, dVDO significantly reduces finite-horizon exploitability compared with VDO and with ODO using uniform reset, and improves the discovered strategy populations. These results all come from the paper's own experimental reports.
AI-written from the coverage · updated 4 d ago
Coverage timeline
Follow the coverage to see the story from different sides.
- arXiv Board & Classic Game AIVirtual Double Oracle: Faster Convergence by Leveraging Strategies That Were Watching All Along
The study proposes Virtual Double Oracle (VDO), which uses a bilinear identity to initialize each newly added strategy with the MWU weights it would have had "if it had existed from the start," rather than resetting to a uniform prior, and proves that it achieves O(√(k/T)) exploitability as the strategy population grows.
Heat trend for this story
- Current (comparable)
- 3
- Peak (comparable)
- 7Nov 18 13:00
- 24-hour change
- -50%
The trend only compares participants observed continuously and completely, so it may cover fewer than the current heat count. Hover over or tap the chart to see each hour's heat; with a keyboard, use the left and right arrow keys.