3 ms·
The replay simulator from history for off-policy eval is clever - avoids expensive rollouts. Curious how they prevent the policy from overfitting to already-dis
by ahmedhossamdev 10d ago
The replay simulator from history for off-policy eval is clever - avoids expensive rollouts. Curious how they prevent the policy from overfitting to already-discovered branches and going stale as the search space expands?