history and links

Cross-entropy descent

Gradient descent that nudges a model's logits by −lr·(q − p) drives the cross-entropy H(P,Q) = H(P) + D(P‖Q) down toward the floor H(P) and never below it, because the gap above the floor is the KL divergence, which is zero only when Q = P.

Versions and lineage

Version
1 · source and history on GitHub
Forked from
nothing — an original
Forks
none yet
Challenges
open challenges on GitHub

Links

Every connection this widget has, with where it came from. A reader can add one or dispute one; an operator decides.

Requires widgets a reader should meet first

    Required by widgets that build on this one

      Related worth seeing next

        Topics where it is filed

          Evidence of use

          A score appears after ten sessions. It weighs checks passed, time engaged, sites embedding it, forks that took, and reader votes.

          Score
          new

          Every change to the links