Payoff-Aware Prediction of Population Game Dynamics
Organizations: School of Computing and Artificial Intelligence, Shanghai University of Finance and Economics · MoE Key Laboratory of Interdisciplinary Research of Computation and Economics, Shanghai University of Finance and Economics · School of Economics and Management, Tsinghua University
Abstract
Population game dynamics describe how aggregate behavior evolves in response to payoff signals. In many applications, the payoff function may change, and the goal is to predict the population trajectory induced by a new payoff design before trajectory data under that design are available. We study this problem as . Standard dynamics learning methods fit the motion observed under training payoffs, but do not separate the payoff-dependent incentive from the payoff-independent response structure needed for prediction under new payoffs. Building on the existing framework of Riemannian game dynamics, we propose PA-RmD, a payoff-aware learning method that keeps payoff information explicit and learns the payoff-independent response structure from data. Theoretically, we show that the learned model preserves fixed-payoff Nash equilibria and derive finite-time one-step prediction bounds under shared-response and payoff-coverage conditions. Empirically, we evaluate PA-RmD on synthetic zero-sum and potential games, as well as a YouTube Trending multi-country dataset, showing that our method performs well under changing payoff designs.
Figures & tables
| Game | Method | ||||||
|---|---|---|---|---|---|---|---|
| One-step ( ) | Rollout RMSE ( ) | Rollout@ ( ) | One-step ( ) | Rollout RMSE ( ) | Rollout@ ( ) | ||
| Zero-sum cyclic | SINDy | (4.05 0.00)E-2 | (4.04 0.00)E-1 | (5.23 0.00)E-1 | (1.09 0.00)E-1 | (2.88 0.00)E-1 | (3.07 0.00)E-1 |
| SIAR | (5.03 0.00)E-2 | (3.05 0.00)E-1 | (3.99 0.00)E-1 | (5.84 0.00)E-2 | (2.11 0.00)E-1 | (2.07 0.00)E-1 | |
| Neural ODE | (1.45 0.05)E-2 | (1.49 0.16)E-1 | (2.09 0.28)E-1 | (7.07 0.20)E-3 | (8.72 0.62)E-2 | (1.20 0.07)E-1 | |
| LEADS | (1.67 0.18)E-2 | (1.75 0.30)E-1 | (2.62 0.56)E-1 | (7.80 1.24)E-3 | (9.71 1.56)E-2 | (1.31 0.16)E-1 | |
| CoDA- | (1.43 0.02)E-2 | (1.42 0.13)E-1 | (2.05 0.29)E-1 | (7.36 0.98)E-3 | (9.38 0.94)E-2 | (1.26 0.15)E-1 | |
Appendix figures & tables22 assets
Supplementary material from the paper’s appendix.
Appendix
| Game | Method | ||||||
|---|---|---|---|---|---|---|---|
| One-step ( ) | Rollout RMSE ( ) | Rollout@ ( ) | One-step ( ) | Rollout RMSE ( ) | Rollout@ ( ) | ||
| Zero-sum cyclic | SINDy | (3.85 0.00)E-1 | (6.83 0.00)E-1 | (7.83 0.00)E-1 | (8.38 0.00)E-2 | (4.65 0.00)E-1 | (6.04 0.00)E-1 |
| SIAR | (7.68 0.00)E-1 | (6.11 0.00)E-1 | (7.16 0.00)E-1 | (1.50 0.00)E-1 | (2.77 0.00)E-1 | (2.79 0.00)E-1 | |
| Neural ODE | (3.19 0.41)E-2 | (3.55 0.71)E-1 | (5.58 1.12)E-1 | (1.36 0.20)E-2 | (1.83 0.29)E-1 | (2.55 0.43)E-1 | |
| LEADS | (3.53 0.30)E-2 | (4.41 0.74)E-1 | (6.88 1.08)E-1 | (1.24 0.12)E-2 | (1.60 0.21)E-1 | (2.24 0.31)E-1 | |
| CoDA- | (3.12 0.38)E-2 | (3.66 0.41)E-1 | (5.88 0.56)E-1 | (1.30 0.09)E-2 | (1.70 0.14)E-1 | (2.37 0.20)E-1 | |
| Game | Method | ||||||
|---|---|---|---|---|---|---|---|
| One-step ( ) | Rollout RMSE ( ) | Rollout@ ( ) | One-step ( ) | Rollout RMSE ( ) | Rollout@ ( ) | ||
| Zero-sum cyclic | SINDy | (1.45 0.00)E-1 | (5.43 0.00)E-1 | (5.78 0.00)E-1 | (2.00 0.00)E-1 | (3.96 0.00)E-1 | (4.83 0.00)E-1 |
| SIAR | (1.24 0.00)E-1 | (3.77 0.00)E-1 | (4.28 0.00)E-1 | (1.30 0.00)E-1 | (2.66 0.00)E-1 | (2.83 0.00)E-1 | |
| Neural ODE | (3.09 0.35)E-2 | (3.02 0.77)E-1 | (4.57 1.54)E-1 | (1.52 0.05)E-2 | (1.79 0.10)E-1 | (2.55 0.17)E-1 | |
| LEADS | (3.28 0.32)E-2 | (3.29 0.69)E-1 | (5.00 1.21)E-1 | (1.58 0.17)E-2 | (1.73 0.12)E-1 | (2.42 0.15)E-1 | |
| CoDA- | (2.94 0.12)E-2 | (2.66 0.58)E-1 | (4.01 1.15)E-1 | (1.49 0.18)E-2 | (1.66 0.23)E-1 | (2.35 0.36)E-1 | |
| Game | Method | ||||||
|---|---|---|---|---|---|---|---|
| One-step ( ) | Rollout RMSE ( ) | Rollout@ ( ) | One-step ( ) | Rollout RMSE ( ) | Rollout@ ( ) | ||
| Zero-sum cyclic | SINDy | (2.02 0.00)E-2 | (2.02 0.00)E-1 | (2.62 0.00)E-1 | (7.28 0.00)E-2 | (1.92 0.00)E-1 | (2.05 0.00)E-1 |
| SIAR | (2.52 0.00)E-2 | (1.52 0.00)E-1 | (2.00 0.00)E-1 | (3.90 0.00)E-2 | (1.41 0.00)E-1 | (1.39 0.00)E-1 | |
| Neural ODE | (7.44 0.19)E-3 | (7.70 0.79)E-2 | (1.09 0.14)E-1 | (4.85 0.12)E-3 | (6.04 0.44)E-2 | (8.27 0.48)E-2 | |
| LEADS | (8.52 0.91)E-3 | (9.01 1.52)E-2 | (1.35 0.28)E-1 | (5.36 0.82)E-3 | (6.77 1.04)E-2 | (9.11 1.06)E-2 | |
| CoDA- | (7.29 0.09)E-3 | (7.28 0.67)E-2 | (1.05 0.15)E-1 | (5.07 0.64)E-3 | (6.52 0.60)E-2 | (8.75 0.99)E-2 | |
| Game | Method | ||||||
|---|---|---|---|---|---|---|---|
| One-step ( ) | Rollout RMSE ( ) | Rollout@ ( ) | One-step ( ) | Rollout RMSE ( ) | Rollout@ ( ) | ||
| Zero-sum cyclic | SINDy | (1.94 0.00)E-1 | (3.50 0.00)E-1 | (4.04 0.00)E-1 | (5.73 0.00)E-2 | (3.22 0.00)E-1 | (4.16 0.00)E-1 |
| SIAR | (3.88 0.00)E-1 | (3.30 0.00)E-1 | (3.89 0.00)E-1 | (1.02 0.00)E-1 | (1.99 0.00)E-1 | (2.04 0.00)E-1 | |
| Neural ODE | (1.61 0.21)E-2 | (1.79 0.36)E-1 | (2.81 0.56)E-1 | (9.51 1.31)E-3 | (1.26 0.19)E-1 | (1.76 0.28)E-1 | |
| LEADS | (1.80 0.15)E-2 | (2.23 0.37)E-1 | (3.47 0.54)E-1 | (8.62 0.80)E-3 | (1.10 0.14)E-1 | (1.55 0.21)E-1 | |
| CoDA- | (1.57 0.19)E-2 | (1.84 0.21)E-1 | (2.95 0.28)E-1 | (9.10 0.59)E-3 | (1.17 0.09)E-1 | (1.64 0.14)E-1 | |
| Game | Method | ||||||
|---|---|---|---|---|---|---|---|
| One-step ( ) | Rollout RMSE ( ) | Rollout@ ( ) | One-step ( ) | Rollout RMSE ( ) | Rollout@ ( ) | ||
| Zero-sum cyclic | SINDy | (7.24 0.00)E-2 | (2.72 0.00)E-1 | (2.91 0.00)E-1 | (1.34 0.00)E-1 | (2.65 0.00)E-1 | (3.23 0.00)E-1 |
| SIAR | (6.21 0.00)E-2 | (1.91 0.00)E-1 | (2.18 0.00)E-1 | (8.65 0.00)E-2 | (1.79 0.00)E-1 | (1.91 0.00)E-1 | |
| Neural ODE | (1.57 0.17)E-2 | (1.55 0.37)E-1 | (2.35 0.76)E-1 | (1.04 0.04)E-2 | (1.24 0.07)E-1 | (1.76 0.12)E-1 | |
| LEADS | (1.67 0.17)E-2 | (1.69 0.35)E-1 | (2.57 0.61)E-1 | (1.08 0.12)E-2 | (1.20 0.08)E-1 | (1.67 0.10)E-1 | |
| CoDA- | (1.49 0.06)E-2 | (1.36 0.29)E-1 | (2.05 0.57)E-1 | (1.02 0.12)E-2 | (1.16 0.15)E-1 | (1.63 0.24)E-1 | |
| Game | Method | ||||||
|---|---|---|---|---|---|---|---|
| One-step ( ) | Rollout RMSE ( ) | Rollout@ ( ) | One-step ( ) | Rollout RMSE ( ) | Rollout@ ( ) | ||
| Zero-sum cyclic | SINDy | (2.81 0.00)E-6 | (3.78 0.00)E-5 | (4.67 0.00)E-5 | (6.15 0.00)E-5 | (9.58 0.00)E-4 | (1.21 0.00)E-3 |
| SIAR | (4.06 0.00)E-6 | (5.55 0.00)E-5 | (7.46 0.00)E-5 | (1.23 0.00)E-4 | (2.04 0.00)E-3 | (2.56 0.00)E-3 | |
| Neural ODE | (3.84 1.11)E-4 | (5.10 1.58)E-3 | (8.13 2.93)E-3 | (4.19 0.46)E-4 | (6.76 1.45)E-3 | (9.14 1.43)E-3 | |
| LEADS | (3.35 0.93)E-4 | (5.28 1.22)E-3 | (7.55 2.57)E-3 | (4.78 0.78)E-4 | (8.71 1.61)E-3 | (1.08 0.19)E-2 | |
| CoDA- | (2.80 0.68)E-4 | (3.99 1.03)E-3 | (5.94 1.49)E-3 | (4.91 1.18)E-4 | (7.85 1.57)E-3 | (1.10 0.28)E-2 | |
| Game | Method | ||||||
|---|---|---|---|---|---|---|---|
| One-step ( ) | Rollout RMSE ( ) | Rollout@ ( ) | One-step ( ) | Rollout RMSE ( ) | Rollout@ ( ) | ||
| Zero-sum cyclic | SINDy | (2.55 0.00)E-3 | (1.74 0.00)E-2 | (2.47 0.00)E-2 | (4.20 0.00)E-3 | (3.55 0.00)E-2 | (3.86 0.00)E-2 |
| SIAR | (8.69 0.00)E-3 | (4.88 0.00)E-2 | (6.09 0.00)E-2 | (7.68 0.00)E-3 | (4.43 0.00)E-2 | (5.54 0.00)E-2 | |
| Neural ODE | (3.68 1.19)E-4 | (2.91 0.95)E-3 | (3.84 1.30)E-3 | (1.26 0.13)E-3 | (1.32 0.14)E-2 | (1.68 0.18)E-2 | |
| LEADS | (5.95 2.30)E-4 | (4.75 1.95)E-3 | (6.32 2.69)E-3 | (1.07 0.09)E-3 | (1.13 0.09)E-2 | (1.48 0.16)E-2 | |
| CoDA- | (2.63 1.01)E-4 | (1.98 0.85)E-3 | (2.70 1.03)E-3 | (1.24 0.05)E-3 | (1.32 0.07)E-2 | (1.65 0.11)E-2 | |
| Game | Method | ||||||
|---|---|---|---|---|---|---|---|
| One-step ( ) | Rollout RMSE ( ) | Rollout@ ( ) | One-step ( ) | Rollout RMSE ( ) | Rollout@ ( ) | ||
| Zero-sum cyclic | SINDy | (1.99 0.00)E-4 | (1.78 0.00)E-3 | (2.83 0.00)E-3 | (1.06 0.00)E-4 | (2.16 0.00)E-3 | (2.35 0.00)E-3 |
| SIAR | (3.84 0.00)E-4 | (5.28 0.00)E-3 | (7.54 0.00)E-3 | (2.98 0.00)E-4 | (5.81 0.00)E-3 | (6.51 0.00)E-3 | |
| Neural ODE | (5.11 2.34)E-4 | (7.77 4.44)E-3 | (1.23 0.73)E-2 | (6.76 0.61)E-4 | (1.36 0.17)E-2 | (1.60 0.17)E-2 | |
| LEADS | (5.31 1.72)E-4 | (8.38 3.61)E-3 | (1.34 0.64)E-2 | (7.89 0.53)E-4 | (1.47 0.12)E-2 | (1.78 0.15)E-2 | |
| CoDA- | (4.42 1.25)E-4 | (6.54 3.24)E-3 | (9.59 4.90)E-3 | (8.20 1.94)E-4 | (1.52 0.26)E-2 | (1.92 0.42)E-2 | |
| Game | Method | Replicator | Projection | MWU | |||
|---|---|---|---|---|---|---|---|
| Zero-sum cyclic | LEADS (few-shot) | (2.91 0.37)E-1 | (1.12 0.12)E-1 | (3.87 2.07)E-2 | (4.53 0.45)E-2 | (5.84 0.31)E-1 | (2.14 0.15)E-1 |
| CoDA- (few-shot) | (1.42 0.13)E-1 | (9.22 1.19)E-2 | (3.66 0.41)E-1 | (1.66 0.21)E-1 | (2.58 0.58)E-1 | (1.61 0.23)E-1 | |
| CoDA- (few-shot) | (1.41 0.12)E-1 | (9.34 0.99)E-2 | (3.47 1.09)E-1 | (1.68 0.21)E-1 | (2.58 0.58)E-1 | (1.59 0.22)E-1 | |
| PA-RmD (zero-shot) | (2.12 0.73)E-2 | (1.73 0.09)E-2 | (1.37 0.06)E-1 | (1.08 0.06)E-1 | (3.04 0.29)E-2 | (2.58 0.11)E-2 | |
| Potential congestion | LEADS (few-shot) | (2.55 0.48)E-2 | (2.22 0.15)E-2 | (1.04 0.24)E-2 | (1.56 0.10)E-2 | (3.17 0.60)E-2 | (3.99 0.17)E-2 |
| Method | ||||||
|---|---|---|---|---|---|---|
| One-step ( ) | Rollout RMSE ( ) | Rollout@ ( ) | One-step ( ) | Rollout RMSE ( ) | Rollout@ ( ) | |
| SINDy | (9.53 0.00)E-1 | (9.67 0.00)E-1 | (9.70 0.00)E-1 | (9.91 0.00)E-1 | (1.01 0.00)E+0 | (1.03 0.00)E+0 |
| SIAR | (3.91 0.00)E-1 | (8.06 0.00)E-1 | (8.30 0.00)E-1 | (2.92 0.00)E-1 | (6.69 0.00)E-1 | (6.77 0.00)E-1 |
| Neural ODE | (1.13 0.05)E-1 | (3.41 0.26)E-1 | (4.03 0.38)E-1 | (9.94 0.29)E-2 | (3.31 0.28)E-1 | (3.89 0.45)E-1 |
| LEADS | (1.81 0.09)E-1 | (6.55 0.26)E-1 | (7.18 0.40)E-1 | (1.57 0.12)E-1 | (6.08 0.63)E-1 | (6.67 0.77)E-1 |
| CoDA- | (1.13 0.01)E-1 | (3.26 0.12)E-1 | (3.74 0.17)E-1 | (1.05 0.04)E-1 | (3.25 0.10)E-1 | (3.77 0.14)E-1 |
| Game | Method | |||||
|---|---|---|---|---|---|---|
| SINDy | (2.88 0.00)E-1 | (5.76 0.00)E-1 | (6.49 0.00)E-1 | (7.47 0.00)E-1 | (7.37 0.00)E-1 | |
| SIAR | (2.11 0.00)E-1 | (2.41 0.00)E-1 | (2.77 0.00)E-1 | (3.29 0.00)E-1 | (4.65 0.00)E-1 | |
| Neural ODE | (8.72 0.62)E-2 | (2.15 0.18)E-1 | (3.67 0.40)E-1 | (5.82 0.23)E-1 | (7.27 0.14)E-1 | |
| LEADS | (9.71 1.56)E-2 | (1.95 0.10)E-1 | (3.12 0.47)E-1 | (5.10 0.72)E-1 | (7.20 0.35)E-1 | |
| CoDA- | (9.38 0.94)E-2 | (1.89 0.15)E-1 | (3.21 0.21)E-1 | (5.45 0.35)E-1 | (7.35 0.27)E-1 | |
| CoDA- | (9.34 0.99)E-2 | (1.79 0.13)E-1 | (3.29 0.15)E-1 | (5.44 0.27)E-1 | (7.44 0.24)E-1 |
Evolutionary Game Theory
Evolutionary game theory studies how strategic behaviors spread or disappear through selection, imitation, learning, and interaction.
See all 32 papers