imagining syntax

← ascent_v1

s4_t4

done · peak unseen_mismatch 98.6

Referenced by

α schedule

α 4.0 ×19 α 4.0 ×19 α 4.0 ×19 α 4.0 ×19 α 4.0 ×19 α 4.0 ×19 α 1.0 ×19 α 1.0 ×19 α 1.0 ×19 α 1.0 ×19 α 1.0 ×19 α 1.0 ×19 α 3.0952 ×19 α 3.0952 ×19 α 3.0952 ×19 α 3.0952 ×19 α 3.0952 ×19 α 3.0952 ×19 α 1.0 ×19 α 1.0 ×19 α 1.0 ×19 α 1.0 ×19 α 1.0 ×19 α 1.0 ×19 α 2.1905 ×19 α 2.1905 ×19 α 2.1905 ×19 α 2.1905 ×19 α 2.1905 ×19 α 2.1905 ×19 α 1.0 ×19 α 1.0 ×19 α 1.0 ×19 α 1.0 ×19 α 1.0 ×19 α 1.0 ×19 α 1.2857 ×18 α 1.2857 ×18 α 1.2857 ×18 α 1.2857 ×18 α 1.2857 ×18 α 1.2857 ×18 α 1.0 ×18 α 1.0 ×18 α 1.0 ×18 α 1.0 ×18 α 1.0 ×18 α 1.0 ×18

direction
mixed
α range
1.0–4.0
stages
48

Figures

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

Results

accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 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
25 51.5 ± 1.649.7 ± 1.649.9 ± 2.049.7 ± 2.1
50 51.1 ± 1.749.9 ± 1.350.1 ± 1.550.4 ± 1.5
75 52.9 ± 2.150.9 ± 1.950.6 ± 1.750.0 ± 2.1
100 58.7 ± 5.555.5 ± 4.252.5 ± 4.051.1 ± 2.2
125 69.9 ± 12.163.5 ± 9.763.2 ± 12.457.4 ± 8.3
150 87.8 ± 12.675.4 ± 15.485.3 ± 14.272.1 ± 16.1
175 97.5 ± 9.275.3 ± 21.397.3 ± 9.673.3 ± 21.7
200 98.3 ± 7.374.5 ± 28.098.4 ± 7.173.4 ± 28.3
225 100.0 ± 0.078.9 ± 24.7100.0 ± 0.077.9 ± 24.6
250 99.6 ± 0.889.3 ± 18.498.8 ± 2.086.5 ± 18.8
275 98.5 ± 1.292.9 ± 11.596.5 ± 3.188.6 ± 11.8
300 96.8 ± 2.392.0 ± 10.492.9 ± 5.686.0 ± 11.3
325 95.6 ± 2.893.3 ± 3.990.3 ± 6.385.2 ± 8.2
350 99.0 ± 1.097.8 ± 1.797.2 ± 2.794.6 ± 3.6
375 100.0 ± 0.199.3 ± 1.899.7 ± 0.598.3 ± 2.8
400 99.9 ± 0.299.7 ± 0.499.6 ± 0.698.2 ± 2.3
450 100.0 ± 0.199.6 ± 0.499.5 ± 0.597.5 ± 2.2
500 97.5 ± 1.195.7 ± 1.692.0 ± 3.686.7 ± 4.5
550 97.1 ± 0.795.2 ± 1.589.8 ± 2.684.9 ± 3.7
600 100.0 ± 0.199.8 ± 0.499.9 ± 0.398.6 ± 2.4
650 99.9 ± 0.199.5 ± 0.699.6 ± 0.596.4 ± 3.1
700 99.7 ± 0.298.3 ± 1.696.8 ± 1.689.2 ± 6.0
750 99.8 ± 0.298.6 ± 1.297.7 ± 1.791.2 ± 6.2
800 99.9 ± 0.199.1 ± 0.998.6 ± 1.192.5 ± 4.8
850 100.0 ± 0.199.3 ± 0.799.2 ± 0.792.8 ± 5.5
900 100.0 ± 0.198.8 ± 1.999.0 ± 1.391.0 ± 8.0

Downloads