s0_e0

done · peak unseen_mismatch 49.6

α schedule

α 0.0 ×50 α 0.0 ×50 α 0.0 ×50 α 0.0 ×50 α 0.0 ×50 α 0.0 ×50 α 0.0 ×50 α 0.0 ×50 α 0.0 ×50 α 0.0 ×50 α 0.0 ×50 α 0.0 ×50 α 0.0 ×50 α 0.0 ×50 α 0.0 ×50 α 0.0 ×50 α 0.0 ×50 α 0.0 ×50 α 0.0 ×50 α 0.0 ×50 α 0.0 ×50 α 0.0 ×50 α 0.0 ×50 α 0.0 ×50

direction: constant
α range: 0.0–0.0
stages: 24

Figures

Results

accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 0.0

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

iteration	seen_match	seen_mismatch	unseen_match	unseen_mismatch
50	50.4 ± 1.6	49.9 ± 1.1	49.8 ± 1.9	49.6 ± 1.9
100	50.0 ± 1.4	50.0 ± 1.8	49.9 ± 0.6	49.4 ± 0.8
150	50.7 ± 4.0	48.9 ± 3.8	50.9 ± 3.9	48.6 ± 2.8
200	60.8 ± 15.5	44.0 ± 9.8	59.4 ± 16.2	44.1 ± 9.5
250	66.2 ± 22.2	42.0 ± 13.4	65.4 ± 22.8	42.4 ± 13.6
300	72.9 ± 22.9	35.8 ± 16.3	72.3 ± 23.0	36.6 ± 17.4
350	84.3 ± 19.9	31.9 ± 18.4	84.3 ± 19.5	31.5 ± 17.9
400	95.8 ± 12.4	22.1 ± 18.3	95.5 ± 13.3	22.7 ± 18.6
450	100.0 ± 0.0	21.4 ± 19.2	100.0 ± 0.0	22.0 ± 19.6
500	100.0 ± 0.0	16.1 ± 18.0	100.0 ± 0.0	16.9 ± 18.7
550	100.0 ± 0.0	18.1 ± 19.7	100.0 ± 0.0	18.8 ± 20.6
600	100.0 ± 0.0	32.1 ± 27.5	100.0 ± 0.0	32.2 ± 27.9
650	100.0 ± 0.0	35.3 ± 27.6	100.0 ± 0.0	35.6 ± 27.9
700	100.0 ± 0.0	24.9 ± 29.7	100.0 ± 0.0	25.2 ± 29.8
750	100.0 ± 0.0	28.8 ± 33.3	100.0 ± 0.0	29.2 ± 33.9
800	100.0 ± 0.0	26.2 ± 19.8	100.0 ± 0.0	25.8 ± 19.5
850	100.0 ± 0.0	21.7 ± 25.1	100.0 ± 0.0	21.9 ± 24.9
900	100.0 ± 0.0	24.4 ± 21.0	100.0 ± 0.0	25.1 ± 22.1
950	100.0 ± 0.0	24.9 ± 17.2	100.0 ± 0.0	24.1 ± 16.6
1000	100.0 ± 0.0	19.2 ± 16.8	100.0 ± 0.0	19.1 ± 16.2
1050	100.0 ± 0.0	18.9 ± 18.9	100.0 ± 0.0	18.6 ± 19.0
1100	100.0 ± 0.0	29.1 ± 21.6	100.0 ± 0.0	28.9 ± 22.5
1150	100.0 ± 0.0	27.6 ± 32.3	100.0 ± 0.0	27.4 ± 32.2
1200	100.0 ± 0.0	26.5 ± 30.0	100.0 ± 0.0	26.1 ± 30.2

s0_e0

α schedule

Figures

Results

Downloads