GridWorld
Benchmarks
Task NameDataset NameSOTA ResultTrendResults
GridWorld (full)
95Accuracy
22
GridWorld
100Reconstruction Rate (Pre-unlearn)
18
Gridworld (test)
88.25Mechanism Mean AUPR
13
Gridworld newobj
98.22AUROC
13
Gridworld mechanism
84.59AUROC
13
GridWorld 3 x 3
0Validation Error
10
GridWorld (test)
96Pre-unlearn Score
9
GridWorld
98Unlearn Efficacy
9
GridWorld
99Pre-unlearn Acc
9
Gridworld
24.5Speedup
8
8x8 two-room gridworld (test)
0.89Validity (%)
8
GridWorld
0.142Average Update Time (s)
7
GridWorld
0.005Final Magnitude |h|
6
Gridworld
42.9Avg Episode Return
6
GridWorld (test)
1.33Flow Matching Loss
5
GridWorld (val)
1.25Flow Matching Loss
5
Gridworld
78Epsilon Matching
5
Gridworld 40% stochastic-policy states
0.37Epsilon Matching Error
5
Gridworld
33Sample Complexity
5
Gridworld (test)
74.2Usefulness
4
Gridworld o.o.d 20x20 (out-of-domain)
91.5Pass@1
4
Gridworld 10x10 (in-domain)
100Pass@1
4
GridWorld 5x5
6Median Episodes to Converge
3
GridWorld 3x3
6Median Episodes to Converge
3
5x5 Gridworld
0.027CMSE
3