imagining syntax

← schedopt_v4

fixed_1.0

done · peak unseen_mismatch 50.1

α schedule

α 1.0 ×400

direction
constant
α range
1.0–1.0
stages
1

Results

accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 1.0

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

iterationseen_matchseen_mismatchunseen_matchunseen_mismatch
100 50.9 ± 1.949.6 ± 1.649.9 ± 2.050.1 ± 1.8
110 51.8 ± 2.949.5 ± 2.351.1 ± 2.749.6 ± 2.2
120 51.5 ± 3.749.5 ± 2.251.6 ± 3.349.2 ± 1.5
130 52.5 ± 7.249.3 ± 3.153.0 ± 7.449.3 ± 3.0
140 52.8 ± 10.948.5 ± 5.753.5 ± 10.748.6 ± 4.7
150 54.1 ± 11.047.7 ± 5.054.4 ± 10.848.0 ± 4.6
160 54.9 ± 11.046.6 ± 7.254.4 ± 11.047.1 ± 7.1
170 55.0 ± 11.048.2 ± 3.654.2 ± 11.248.0 ± 3.2
180 58.6 ± 14.646.7 ± 7.057.9 ± 14.846.3 ± 6.7
190 62.3 ± 18.344.1 ± 10.761.8 ± 18.644.2 ± 10.7
200 64.9 ± 19.342.8 ± 12.564.3 ± 19.743.3 ± 12.4
210 69.3 ± 20.140.3 ± 13.168.2 ± 20.740.2 ± 13.9
220 75.0 ± 20.236.1 ± 14.874.3 ± 20.736.0 ± 15.0
230 80.6 ± 20.933.2 ± 16.380.1 ± 21.333.1 ± 16.3
240 84.0 ± 20.729.5 ± 17.383.3 ± 21.330.3 ± 17.6
250 84.9 ± 21.028.3 ± 18.384.6 ± 21.127.9 ± 18.1
260 87.9 ± 19.923.6 ± 19.588.0 ± 19.723.7 ± 19.4
270 88.9 ± 19.419.6 ± 20.089.3 ± 18.919.7 ± 20.2
280 90.6 ± 17.823.2 ± 19.390.8 ± 17.622.6 ± 19.4
290 92.9 ± 15.924.6 ± 20.993.0 ± 15.924.6 ± 20.6
300 94.5 ± 15.022.7 ± 19.994.5 ± 14.822.4 ± 19.3
310 95.2 ± 14.326.1 ± 22.095.0 ± 15.126.1 ± 21.9
320 94.9 ± 15.429.2 ± 20.695.1 ± 14.829.0 ± 20.4
330 95.1 ± 14.629.9 ± 21.695.2 ± 14.529.4 ± 21.2
340 95.3 ± 14.030.7 ± 21.595.4 ± 13.930.1 ± 21.5
350 95.5 ± 13.529.9 ± 25.395.7 ± 12.829.1 ± 24.5
360 96.0 ± 12.327.2 ± 21.595.9 ± 12.426.4 ± 20.8
370 97.2 ± 10.326.2 ± 21.197.1 ± 10.425.4 ± 20.9
380 98.2 ± 7.328.9 ± 23.198.2 ± 7.228.2 ± 23.0
390 98.9 ± 4.724.8 ± 23.699.1 ± 3.724.5 ± 24.1
400 99.7 ± 1.225.3 ± 22.699.8 ± 1.024.9 ± 22.8

Downloads