imagining syntax

← schedopt_v4

free-5f5c2004

done · peak unseen_mismatch 83.8

α schedule

α 3.509 ×20 α 1.399 ×17 α 4.376 ×25 α 1.878 ×87 α 3.244 ×19 α 1.198 ×24 α 4.021 ×11 α 1.01 ×16 α 3.794 ×48 α 1.613 ×18 α 4.141 ×15 α 4.141 ×100

direction
mixed
α range
1.01–4.376
stages
12

Results

accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 4.141

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

iterationseen_matchseen_mismatchunseen_matchunseen_mismatch
100 53.5 ± 4.850.6 ± 2.451.7 ± 4.150.1 ± 2.1
110 55.7 ± 7.450.7 ± 3.353.1 ± 6.750.0 ± 3.1
120 58.3 ± 9.651.8 ± 4.755.9 ± 9.250.8 ± 4.6
130 62.6 ± 13.853.7 ± 6.360.0 ± 13.752.0 ± 5.8
140 67.5 ± 16.957.8 ± 10.664.9 ± 17.154.8 ± 9.6
150 73.0 ± 18.261.5 ± 12.870.3 ± 18.458.1 ± 11.6
160 78.4 ± 18.267.0 ± 14.875.4 ± 18.763.5 ± 14.5
170 81.7 ± 17.371.7 ± 15.378.6 ± 17.868.4 ± 14.2
180 85.5 ± 16.774.4 ± 15.683.3 ± 17.771.7 ± 15.1
190 88.1 ± 16.775.2 ± 15.786.8 ± 17.372.7 ± 15.3
200 90.7 ± 15.574.5 ± 14.889.6 ± 16.371.7 ± 13.9
210 91.9 ± 14.778.7 ± 15.790.4 ± 15.876.0 ± 15.1
220 92.6 ± 14.478.1 ± 17.291.6 ± 15.475.4 ± 16.9
230 93.3 ± 13.276.1 ± 16.292.5 ± 14.173.8 ± 15.3
240 94.8 ± 11.080.1 ± 14.493.7 ± 12.877.8 ± 14.4
250 95.6 ± 8.485.0 ± 13.794.0 ± 11.082.2 ± 13.8
260 95.0 ± 7.386.9 ± 9.592.3 ± 11.283.0 ± 11.4
270 94.2 ± 6.885.4 ± 10.491.0 ± 9.580.3 ± 12.0
280 94.6 ± 5.185.0 ± 9.691.9 ± 6.581.1 ± 10.2
290 97.4 ± 3.788.6 ± 8.694.8 ± 6.083.7 ± 7.8
300 95.5 ± 4.489.5 ± 3.989.9 ± 9.182.5 ± 7.0
310 95.8 ± 3.992.4 ± 3.490.3 ± 7.083.8 ± 7.5
320 94.3 ± 4.290.8 ± 1.888.7 ± 6.883.3 ± 3.8
330 94.0 ± 3.290.1 ± 0.887.7 ± 4.981.8 ± 2.0
340 93.8 ± 4.188.8 ± 3.686.7 ± 6.578.7 ± 5.1
350 93.0 ± 3.489.8 ± 1.384.4 ± 5.379.3 ± 3.5
360 92.9 ± 3.991.0 ± 2.284.7 ± 5.781.0 ± 5.0
370 92.1 ± 3.690.2 ± 1.582.8 ± 5.079.2 ± 4.2
380 91.5 ± 2.490.0 ± 0.880.8 ± 3.677.0 ± 3.8
390 91.2 ± 1.489.3 ± 1.880.6 ± 2.678.0 ± 3.2
400 89.9 ± 2.489.0 ± 1.079.4 ± 2.475.8 ± 1.2

Downloads