fixed_1.9

done · peak unseen_mismatch 97.1

α schedule

α 1.9 ×400

accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 1.9

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
100	51.7 ± 2.2	50.0 ± 1.0	50.0 ± 1.8	50.1 ± 1.9
110	52.5 ± 2.4	49.7 ± 1.4	51.1 ± 2.0	49.7 ± 2.0
120	52.3 ± 3.0	49.8 ± 1.7	51.4 ± 3.1	49.7 ± 1.5
130	52.8 ± 3.4	49.8 ± 1.5	51.8 ± 3.4	49.3 ± 1.8
140	55.2 ± 8.9	49.8 ± 3.5	54.6 ± 8.4	49.3 ± 2.3
150	59.0 ± 13.0	49.7 ± 6.8	57.6 ± 13.1	48.2 ± 5.2
160	62.3 ± 14.0	50.6 ± 8.1	60.9 ± 14.0	49.5 ± 6.7
170	65.9 ± 16.9	51.9 ± 9.3	63.7 ± 17.1	50.6 ± 8.0
180	69.2 ± 19.1	53.9 ± 13.1	67.8 ± 19.2	52.4 ± 12.2
190	76.5 ± 19.8	59.9 ± 15.5	75.0 ± 20.0	58.1 ± 15.7
200	81.5 ± 18.4	58.9 ± 13.7	80.5 ± 19.2	57.5 ± 13.8
210	86.4 ± 17.6	63.1 ± 11.9	85.0 ± 18.5	60.7 ± 10.7
220	89.5 ± 16.9	69.2 ± 12.7	88.1 ± 17.6	67.1 ± 11.8
230	91.5 ± 16.2	71.4 ± 18.4	90.3 ± 16.9	69.8 ± 17.2
240	91.6 ± 15.3	71.7 ± 16.0	90.8 ± 15.9	70.5 ± 15.4
250	92.1 ± 14.9	75.3 ± 14.9	91.4 ± 15.3	73.5 ± 14.7
260	94.7 ± 12.2	79.0 ± 13.5	94.5 ± 12.2	76.9 ± 13.4
270	95.2 ± 11.4	83.5 ± 12.6	94.5 ± 12.1	81.4 ± 12.6
280	95.8 ± 8.9	85.1 ± 11.1	95.2 ± 9.8	83.6 ± 11.1
290	97.9 ± 3.3	88.3 ± 4.5	97.5 ± 3.8	88.0 ± 3.1
300	98.7 ± 1.8	91.6 ± 8.1	98.1 ± 2.9	89.7 ± 6.8
310	98.8 ± 1.1	87.7 ± 8.0	96.8 ± 2.9	83.7 ± 6.8
320	99.0 ± 0.9	76.6 ± 16.6	97.2 ± 2.5	78.7 ± 13.1
330	100.0 ± 0.0	69.6 ± 0.0	100.0 ± 0.0	71.8 ± 0.0
340	100.0 ± 0.0	68.0 ± 0.0	100.0 ± 0.0	69.8 ± 0.0
350	100.0 ± 0.0	87.6 ± 0.0	100.0 ± 0.0	85.2 ± 0.0
360	100.0 ± 0.0	89.0 ± 0.0	100.0 ± 0.0	88.2 ± 0.0
370	100.0 ± 0.0	99.6 ± 0.0	99.6 ± 0.0	97.1 ± 0.0