imagining syntax

← schedopt_v5_confirm

flat_1_6

done · peak unseen_mismatch 71.4

Referenced by

α schedule

α 1.6 ×400

direction
constant
α range
1.6–1.6
stages
1

Figures

schedule.png
schedule.png

Results

accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 1.6

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

iterationseen_matchseen_mismatchunseen_matchunseen_mismatch
100 51.4 ± 3.449.6 ± 1.950.8 ± 3.349.6 ± 2.0
110 52.0 ± 4.349.7 ± 1.851.3 ± 4.349.6 ± 1.9
120 52.8 ± 5.449.8 ± 2.152.1 ± 5.449.5 ± 2.1
130 54.6 ± 7.549.8 ± 3.053.6 ± 7.349.4 ± 2.8
140 57.4 ± 11.449.2 ± 5.456.1 ± 11.248.9 ± 4.9
150 60.9 ± 14.949.4 ± 7.859.7 ± 14.848.9 ± 7.7
160 64.7 ± 17.649.9 ± 8.763.7 ± 17.749.3 ± 8.1
170 68.8 ± 19.950.4 ± 11.767.6 ± 20.549.5 ± 11.3
180 72.9 ± 20.951.4 ± 13.371.7 ± 21.350.8 ± 12.9
190 77.0 ± 21.152.6 ± 14.676.0 ± 21.551.8 ± 14.0
200 80.5 ± 21.053.7 ± 17.479.9 ± 21.352.9 ± 16.9
210 83.4 ± 19.955.4 ± 17.782.7 ± 20.454.7 ± 17.1
220 86.8 ± 18.355.6 ± 18.586.1 ± 19.054.8 ± 18.3
230 89.8 ± 16.757.7 ± 21.188.9 ± 17.557.0 ± 20.4
240 91.8 ± 15.658.5 ± 21.191.1 ± 16.257.8 ± 20.4
250 94.1 ± 13.657.8 ± 22.293.4 ± 14.457.1 ± 21.7
260 95.0 ± 13.356.2 ± 22.294.6 ± 13.655.4 ± 21.4
270 95.9 ± 12.557.6 ± 23.295.6 ± 12.856.9 ± 22.3
280 96.1 ± 11.957.0 ± 24.195.8 ± 12.656.1 ± 23.4
290 96.5 ± 11.457.7 ± 24.996.3 ± 11.956.9 ± 24.1
300 97.0 ± 10.557.0 ± 22.496.8 ± 11.056.4 ± 21.8
310 97.2 ± 10.060.6 ± 22.296.9 ± 11.159.9 ± 21.7
320 97.3 ± 10.060.1 ± 23.897.2 ± 10.659.4 ± 23.1
330 97.6 ± 9.557.8 ± 25.997.4 ± 9.957.0 ± 25.4
340 98.0 ± 8.260.3 ± 27.597.8 ± 8.959.5 ± 26.5
350 98.1 ± 7.557.9 ± 23.997.8 ± 8.857.4 ± 23.1
360 98.8 ± 4.658.6 ± 24.798.6 ± 5.358.0 ± 23.9
370 99.2 ± 2.859.8 ± 21.399.0 ± 3.559.2 ± 21.0
380 99.6 ± 1.763.5 ± 20.899.4 ± 2.362.8 ± 20.3
390 99.8 ± 0.767.7 ± 20.399.7 ± 1.267.3 ± 19.4
400 100.0 ± 0.172.0 ± 22.899.8 ± 0.671.4 ± 22.3

Downloads