imagining syntax

← schedopt_v4

free-32cd833d

done · peak unseen_mismatch 79.7

α schedule

α 3.245 ×14 α 1.414 ×14 α 3.9 ×104 α 1.7 ×30 α 3.283 ×16 α 0.9412 ×64 α 3.503 ×13 α 1.551 ×27 α 3.749 ×18 α 3.749 ×100

direction
mixed
α range
0.9412–3.9
stages
10

Results

accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 3.749

seen_* probed at reference α = 1.4 · unseen_* = held-out pairings, uniform (α-independent) · 1000 pairs per condition

iterationseen_matchseen_mismatchunseen_matchunseen_mismatch
100 55.5 ± 7.452.0 ± 4.652.6 ± 5.850.4 ± 2.9
110 58.0 ± 8.554.2 ± 6.153.8 ± 7.351.2 ± 4.5
120 61.7 ± 10.657.7 ± 8.556.8 ± 9.253.0 ± 6.1
130 65.9 ± 11.861.7 ± 10.459.3 ± 10.555.4 ± 7.6
140 69.9 ± 13.365.1 ± 11.963.1 ± 12.758.1 ± 9.5
150 75.4 ± 14.969.1 ± 13.369.1 ± 15.362.0 ± 11.9
160 80.1 ± 16.073.6 ± 14.874.4 ± 17.167.4 ± 13.8
170 83.8 ± 15.178.3 ± 14.177.7 ± 16.771.5 ± 14.3
180 84.8 ± 13.479.9 ± 12.677.0 ± 14.771.1 ± 12.5
190 88.7 ± 13.783.8 ± 13.384.0 ± 15.377.7 ± 13.8
200 91.0 ± 14.483.9 ± 15.388.4 ± 15.679.6 ± 15.9
210 90.7 ± 16.877.8 ± 19.089.2 ± 17.274.7 ± 18.7
220 89.8 ± 18.373.4 ± 17.989.1 ± 18.971.1 ± 18.2
230 89.4 ± 18.273.1 ± 18.288.3 ± 19.670.8 ± 18.4
240 87.7 ± 18.771.1 ± 16.987.0 ± 19.468.3 ± 16.5
250 88.8 ± 15.672.1 ± 12.987.1 ± 17.667.9 ± 13.5
260 91.9 ± 13.781.6 ± 13.790.3 ± 15.277.8 ± 14.7
270 90.5 ± 14.974.5 ± 14.488.9 ± 15.671.1 ± 14.7
280 90.1 ± 16.474.2 ± 13.688.9 ± 16.673.0 ± 13.5
290 91.5 ± 14.682.1 ± 11.389.1 ± 18.578.7 ± 14.0
300 90.3 ± 15.478.3 ± 18.287.5 ± 19.675.8 ± 18.9
310 82.6 ± 17.466.5 ± 0.278.5 ± 21.262.2 ± 3.9
320 84.2 ± 15.870.7 ± 3.077.8 ± 22.268.4 ± 10.3
330 85.8 ± 14.281.0 ± 7.280.1 ± 19.976.0 ± 13.6
340 86.6 ± 13.285.5 ± 11.279.5 ± 20.479.7 ± 16.4
350 76.7 ± 0.076.5 ± 0.062.3 ± 0.064.5 ± 0.0
360 81.6 ± 0.080.5 ± 0.066.1 ± 0.067.5 ± 0.0
370 81.0 ± 0.082.0 ± 0.065.2 ± 0.068.4 ± 0.0
380 82.3 ± 0.083.6 ± 0.066.4 ± 0.068.2 ± 0.0
390 84.3 ± 0.084.8 ± 0.070.2 ± 0.069.9 ± 0.0
400 84.2 ± 0.084.1 ± 0.068.6 ± 0.070.7 ± 0.0

Downloads