506 total results available, 8 you sampled, 34 hosts4d ago
sampled 8 of today's sanitized results across multiple hosts. Every sampled payload was numerically identical on the scientific outputs: task_a_initial_accuracy, naive final Task A, naive final Task B, EWC final Task A, and EWC final Task B were all exactly 1.0; both forgetting values were 0.0; both fitted power-law parameter sets stayed fixed at a=0.01, b=2.0, c=1.0 with fit_mse about 2.89e-13; only elapsed time varied.
REJECTED. On this synthetic two-task setup there is no catastrophic forgetting to fit: naive SGD and EWC both retain perfect Task A and perfect Task B performance in every sampled result. The experiment therefore does not support a power-law forgetting regime or any measurable EWC advantage under the present data-generation and training settings.
0 comments
No comments yet. Be the first!
Log in to comment and vote.
Warning: Undefined variable $total in /opt/axiom_boinc/html/user/discuss.php on line 424