imagining syntax

← schedopt_v4

free-29959d29

done · peak unseen_mismatch 86.9

α schedule

α 4.814 ×16 α 2.75 ×22 α 2.287 ×12 α 3.633 ×16 α 2.344 ×88 α 1.027 ×11 α 2.037 ×21 α 0.2759 ×28 α 3.877 ×29 α 3.747 ×34 α 4.64 ×23 α 4.64 ×100

direction
mixed
α range
0.2759–4.814
stages
12

Results

accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 4.64

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

iterationseen_matchseen_mismatchunseen_matchunseen_mismatch
100 55.4 ± 5.852.2 ± 3.652.5 ± 3.949.6 ± 1.8
110 58.8 ± 8.854.2 ± 6.454.6 ± 7.050.9 ± 3.9
120 64.4 ± 10.456.9 ± 8.358.5 ± 9.852.8 ± 6.0
130 71.4 ± 12.961.9 ± 10.364.8 ± 12.756.4 ± 8.1
140 77.8 ± 14.568.4 ± 12.771.5 ± 15.162.3 ± 10.7
150 83.3 ± 14.474.0 ± 13.877.1 ± 15.267.7 ± 12.2
160 87.4 ± 14.078.8 ± 13.582.7 ± 14.873.1 ± 12.4
170 92.0 ± 13.283.8 ± 14.689.5 ± 14.179.5 ± 13.8
180 92.3 ± 13.684.6 ± 14.989.5 ± 14.679.5 ± 14.4
190 91.6 ± 14.983.5 ± 15.488.9 ± 14.978.7 ± 13.8
200 93.1 ± 15.381.7 ± 19.792.4 ± 15.879.4 ± 19.6
210 90.1 ± 18.466.4 ± 20.289.6 ± 19.065.0 ± 20.0
220 90.3 ± 17.861.7 ± 18.689.8 ± 18.460.4 ± 18.1
230 92.0 ± 15.769.9 ± 14.491.3 ± 16.469.6 ± 14.6
240 92.7 ± 14.379.4 ± 12.691.6 ± 15.777.5 ± 13.0
250 94.1 ± 11.389.3 ± 10.392.4 ± 13.486.9 ± 11.9
260 91.3 ± 11.185.0 ± 12.087.9 ± 12.780.7 ± 12.5
270 90.0 ± 9.088.5 ± 6.986.3 ± 11.883.2 ± 9.5
280 89.8 ± 8.888.1 ± 7.584.2 ± 10.683.3 ± 10.5
290 87.2 ± 7.486.3 ± 6.580.4 ± 8.778.8 ± 8.9
300 87.0 ± 7.786.2 ± 6.280.3 ± 8.478.6 ± 7.9
310 87.5 ± 6.287.5 ± 4.780.8 ± 7.078.8 ± 6.3
320 86.8 ± 6.686.6 ± 3.877.6 ± 7.175.6 ± 6.9
330 88.0 ± 4.987.7 ± 2.276.5 ± 5.875.4 ± 4.3
340 88.2 ± 2.387.5 ± 0.778.2 ± 3.175.8 ± 2.4
350 87.1 ± 3.285.9 ± 1.177.6 ± 3.675.1 ± 2.2
360 86.6 ± 2.985.8 ± 0.975.8 ± 4.073.1 ± 2.5
370 88.4 ± 0.787.1 ± 1.177.4 ± 3.274.8 ± 3.3
380 86.6 ± 1.786.1 ± 0.275.7 ± 1.773.9 ± 1.2
390 86.8 ± 1.885.6 ± 1.074.7 ± 1.472.7 ± 0.7
400 87.4 ± 1.685.2 ± 0.874.4 ± 2.471.4 ± 0.6

Downloads