T600

done · peak unseen_mismatch 99.8

α schedule

α 5.0 ×13 α 5.0 ×13 α 5.0 ×13 α 5.0 ×13 α 5.0 ×13 α 5.0 ×13 α 5.0 ×13 α 5.0 ×13 α 0.7 ×13 α 0.7 ×13 α 0.7 ×13 α 0.7 ×13 α 0.7 ×13 α 0.7 ×13 α 0.7 ×13 α 0.7 ×13 α 3.1429 ×13 α 3.1429 ×13 α 3.1429 ×13 α 3.1429 ×13 α 3.1429 ×13 α 3.1429 ×13 α 3.1429 ×13 α 3.1429 ×13 α 0.7 ×12 α 0.7 ×12 α 0.7 ×12 α 0.7 ×12 α 0.7 ×12 α 0.7 ×12 α 0.7 ×12 α 0.7 ×12 α 1.2857 ×12 α 1.2857 ×12 α 1.2857 ×12 α 1.2857 ×12 α 1.2857 ×12 α 1.2857 ×12 α 1.2857 ×12 α 1.2857 ×12 α 0.7 ×12 α 0.7 ×12 α 0.7 ×12 α 0.7 ×12 α 0.7 ×12 α 0.7 ×12 α 0.7 ×12 α 0.7 ×12

direction: mixed
α range: 0.7–5.0
stages: 48

Figures

Results

accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 0.7

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
25	51.5 ± 1.5	49.9 ± 1.3	49.8 ± 1.9	49.6 ± 2.2
50	50.9 ± 1.8	50.0 ± 1.2	50.2 ± 1.6	50.4 ± 1.5
75	54.0 ± 1.9	51.6 ± 1.9	51.1 ± 1.7	50.1 ± 2.2
100	62.4 ± 6.2	58.5 ± 4.6	54.2 ± 4.3	51.8 ± 2.1
125	73.9 ± 15.2	63.8 ± 10.4	68.7 ± 17.1	58.4 ± 9.7
150	89.5 ± 13.8	68.2 ± 15.1	87.7 ± 15.0	66.1 ± 14.3
175	98.1 ± 7.9	59.2 ± 25.1	98.0 ± 8.5	57.6 ± 24.8
200	100.0 ± 0.1	53.4 ± 31.3	99.9 ± 0.3	52.7 ± 31.2
225	100.0 ± 0.0	71.2 ± 31.1	100.0 ± 0.0	70.4 ± 30.9
250	99.8 ± 0.3	89.4 ± 22.1	99.6 ± 1.1	88.0 ± 21.6
275	99.1 ± 1.2	94.0 ± 13.7	98.1 ± 2.6	91.2 ± 13.4
300	97.6 ± 2.0	93.2 ± 9.3	95.2 ± 3.8	88.0 ± 10.5
325	100.0 ± 0.1	99.7 ± 0.6	100.0 ± 0.1	99.6 ± 0.7
350	100.0 ± 0.0	98.4 ± 6.9	100.0 ± 0.0	98.4 ± 7.0
375	100.0 ± 0.0	99.8 ± 0.6	100.0 ± 0.0	99.8 ± 0.7
400	100.0 ± 0.0	98.1 ± 7.1	100.0 ± 0.0	97.5 ± 6.7
450	99.8 ± 0.3	98.6 ± 1.7	99.2 ± 1.2	95.9 ± 3.5
500	99.7 ± 0.3	98.9 ± 0.9	99.0 ± 0.9	94.7 ± 4.5
550	100.0 ± 0.0	99.8 ± 0.6	100.0 ± 0.1	98.9 ± 1.7
600	100.0 ± 0.0	99.0 ± 3.3	99.9 ± 0.1	97.4 ± 4.3

T600

α schedule

Figures

Results

Downloads