File size: 1,813 Bytes
4bd54f8 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 | {
"assessment": "verified",
"claim": 5,
"destructive_control": {
"effect": "the unweighted baseline is 4-6x farther from the independently sampled fresh expectation",
"mutation": "discard SOSMC importance weights and use the unweighted ImpDiff particle mean",
"tracking_error_reduction_factors_from_weighting": [
4.174582065578795,
5.547755280665158
]
},
"literal_claim": "On 2D EBM benchmark datasets, SOSMC achieves higher objective values than ImpDiff for small regularization strengths, with particle-based reward estimates that track true expectations throughout optimization (Section 5.2).",
"released_native_artifact_runs": [
{
"beta_kl": 0.1,
"final_impdiff_fresh_reward": 0.9949566426873208,
"final_sosmc_fresh_reward": 0.9985862391233444,
"fresh_eval_steps": [
0,
1000,
2000,
3000,
4000,
5000
],
"impdiff_particle_to_fresh_mae": 0.05335863728920618,
"outer_iterations": 5001,
"particles": 10000,
"sosmc_fresh_reward_advantage": 0.0036295964360236743,
"sosmc_weighted_particle_to_fresh_mae": 0.012781791434685367,
"tracking_error_reduction_factor": 4.174582065578795
},
{
"beta_kl": 0.25,
"final_impdiff_fresh_reward": 0.9689168484330177,
"final_sosmc_fresh_reward": 0.9957832426309585,
"fresh_eval_steps": [
0,
1000,
2000,
3000,
4000,
5000
],
"impdiff_particle_to_fresh_mae": 0.07853053498864175,
"outer_iterations": 5001,
"particles": 10000,
"sosmc_fresh_reward_advantage": 0.02686639419794079,
"sosmc_weighted_particle_to_fresh_mae": 0.014155371139446907,
"tracking_error_reduction_factor": 5.547755280665158
}
]
}
|