imagining syntax

← deform_v1

v_0.4

done · peak unseen_mismatch 99.1

Referenced by

α schedule

α 3.0 ×19 α 3.0 ×19 α 3.0 ×19 α 3.0 ×19 α 3.0 ×19 α 3.0 ×19 α 3.0 ×19 α 3.0 ×19 α 0.4 ×19 α 0.4 ×19 α 0.4 ×19 α 0.4 ×19 α 0.4 ×19 α 0.4 ×19 α 0.4 ×19 α 0.4 ×19 α 2.1429 ×19 α 2.1429 ×19 α 2.1429 ×19 α 2.1429 ×19 α 2.1429 ×19 α 2.1429 ×19 α 2.1429 ×19 α 2.1429 ×19 α 0.4 ×19 α 0.4 ×19 α 0.4 ×19 α 0.4 ×19 α 0.4 ×19 α 0.4 ×19 α 0.4 ×19 α 0.4 ×19 α 1.2857 ×19 α 1.2857 ×19 α 1.2857 ×19 α 1.2857 ×19 α 1.2857 ×18 α 1.2857 ×18 α 1.2857 ×18 α 1.2857 ×18 α 0.4 ×18 α 0.4 ×18 α 0.4 ×18 α 0.4 ×18 α 0.4 ×18 α 0.4 ×18 α 0.4 ×18 α 0.4 ×18

direction
mixed
α range
0.4–3.0
stages
48

Figures

schedule.png
schedule.png
trajectory.png
trajectory.png

Results

accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 0.4

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

iterationseen_matchseen_mismatchunseen_matchunseen_mismatch
25 51.4 ± 1.549.5 ± 1.649.8 ± 1.949.7 ± 2.1
50 50.7 ± 1.949.6 ± 1.450.1 ± 1.550.8 ± 1.1
75 51.9 ± 1.850.1 ± 1.950.5 ± 1.750.0 ± 1.9
100 53.8 ± 2.952.1 ± 2.951.5 ± 1.850.0 ± 1.5
125 65.8 ± 10.560.5 ± 8.159.9 ± 9.254.9 ± 5.3
150 79.3 ± 12.473.6 ± 12.872.2 ± 14.166.3 ± 12.7
175 92.1 ± 13.278.1 ± 20.190.5 ± 14.875.3 ± 20.4
200 97.2 ± 11.271.8 ± 25.997.3 ± 10.770.3 ± 25.3
225 97.4 ± 11.463.7 ± 28.097.5 ± 11.062.8 ± 27.7
250 97.5 ± 10.962.4 ± 29.397.7 ± 10.261.8 ± 29.2
275 97.5 ± 10.761.3 ± 33.497.5 ± 10.960.7 ± 33.7
300 98.0 ± 8.555.9 ± 33.198.0 ± 8.755.5 ± 33.3
325 100.0 ± 0.070.2 ± 30.5100.0 ± 0.069.6 ± 30.6
350 100.0 ± 0.183.9 ± 22.999.8 ± 0.682.4 ± 22.9
375 99.6 ± 0.793.0 ± 12.598.8 ± 1.990.9 ± 12.7
400 99.1 ± 1.194.7 ± 9.397.7 ± 3.391.0 ± 11.3
450 98.5 ± 1.397.4 ± 2.094.9 ± 4.892.1 ± 5.7
500 100.0 ± 0.099.4 ± 2.2100.0 ± 0.099.1 ± 2.6
550 100.0 ± 0.091.2 ± 22.0100.0 ± 0.090.4 ± 22.1
600 100.0 ± 0.086.4 ± 24.8100.0 ± 0.084.5 ± 24.9
650 99.9 ± 0.295.4 ± 11.198.4 ± 1.990.6 ± 12.1
700 99.9 ± 0.299.0 ± 0.999.0 ± 0.994.9 ± 4.9
750 99.9 ± 0.198.6 ± 1.498.3 ± 1.693.0 ± 6.5
800 100.0 ± 0.092.8 ± 10.7100.0 ± 0.090.6 ± 14.2
850 100.0 ± 0.082.9 ± 18.6100.0 ± 0.079.6 ± 21.6
900 100.0 ± 0.074.9 ± 22.6100.0 ± 0.072.3 ± 23.7

Downloads