sigmoid-ReLU test-accuracy gap +0.02800 +/- 0.00000 (d=inf, 100.0% sign-consistent); tanh-ReLU test-accuracy gap +0.02200 +/- 0.00000 (d=inf, 100.0% sign-consistent); softplus-ReLU test-accuracy gap +0.01400 +/- 0.00000 (d=inf, 100.0% sign-consistent)
CONFIRMED On this fixed 30-128-64-10 classification task, activation choice changes end accuracy deterministically rather than marginally. Sigmoid is best in every returned seed, with tanh next and softplus also above ReLU, so this landscape is stable enough to treat as a confirmed ranking result for this setup.
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