imagining syntax

← schedopt_v4

fixed_0.9

done · peak unseen_mismatch 50.1

α schedule

α 0.9 ×400

direction
constant
α range
0.9–0.9
stages
1

Results

accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 0.9

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

iterationseen_matchseen_mismatchunseen_matchunseen_mismatch
100 50.7 ± 2.049.8 ± 1.450.1 ± 1.950.1 ± 1.9
110 51.5 ± 1.449.4 ± 1.650.8 ± 1.749.6 ± 1.7
120 51.1 ± 1.949.6 ± 2.151.3 ± 2.249.2 ± 1.4
130 51.0 ± 2.049.6 ± 1.751.4 ± 1.849.6 ± 1.6
140 52.2 ± 5.248.6 ± 3.052.7 ± 4.148.4 ± 3.2
150 52.7 ± 7.947.4 ± 6.253.4 ± 7.647.7 ± 6.3
160 54.7 ± 10.745.8 ± 8.754.7 ± 10.846.6 ± 8.5
170 56.0 ± 12.645.8 ± 10.755.2 ± 12.745.8 ± 10.5
180 58.0 ± 15.344.4 ± 12.457.5 ± 15.444.4 ± 12.4
190 61.4 ± 17.243.5 ± 12.660.7 ± 17.143.8 ± 11.9
200 63.8 ± 19.941.4 ± 15.963.9 ± 20.141.5 ± 15.9
210 66.9 ± 21.840.6 ± 16.766.7 ± 21.940.7 ± 16.8
220 69.7 ± 21.340.9 ± 15.369.2 ± 21.840.5 ± 15.2
230 71.3 ± 21.838.0 ± 17.171.0 ± 22.238.0 ± 17.0
240 75.2 ± 22.033.0 ± 19.074.7 ± 22.133.4 ± 19.5
250 78.2 ± 22.137.8 ± 16.777.9 ± 22.337.3 ± 17.3
260 80.5 ± 21.532.1 ± 19.779.8 ± 22.032.0 ± 20.2
270 83.8 ± 21.333.1 ± 20.583.9 ± 21.133.1 ± 20.6
280 87.1 ± 19.129.1 ± 21.187.2 ± 18.929.4 ± 21.7
290 88.1 ± 18.930.9 ± 18.788.3 ± 19.030.4 ± 18.2
300 91.4 ± 16.031.1 ± 21.391.3 ± 16.130.2 ± 20.5
310 94.1 ± 13.830.4 ± 22.593.7 ± 14.330.1 ± 22.4
320 95.1 ± 14.130.9 ± 22.895.1 ± 14.031.0 ± 22.7
330 95.7 ± 13.130.6 ± 21.295.9 ± 12.530.5 ± 21.5
340 97.3 ± 10.230.4 ± 21.297.3 ± 10.230.6 ± 20.9
350 97.7 ± 9.230.2 ± 20.697.8 ± 9.029.1 ± 20.2
360 98.5 ± 6.338.2 ± 14.898.4 ± 6.837.2 ± 14.1
370 99.3 ± 3.131.1 ± 20.299.3 ± 3.130.7 ± 19.7
380 100.0 ± 0.030.0 ± 23.6100.0 ± 0.129.8 ± 23.7
390 100.0 ± 0.032.1 ± 20.1100.0 ± 0.032.2 ± 20.4
400 100.0 ± 0.034.4 ± 22.3100.0 ± 0.034.6 ± 22.4

Downloads