imagining syntax

← schedopt_v4

free-65657eb1

done · peak unseen_mismatch 83.6

α schedule

α 3.8 ×43 α 1.123 ×16 α 4.139 ×21 α 1.329 ×83 α 4.207 ×98 α 1.047 ×39 α 1.047 ×100

direction
mixed
α range
1.047–4.207
stages
7

Results

accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 1.047

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

iterationseen_matchseen_mismatchunseen_matchunseen_mismatch
100 54.2 ± 6.151.1 ± 2.352.6 ± 5.950.3 ± 1.8
110 56.6 ± 8.851.2 ± 3.654.3 ± 8.650.3 ± 2.9
120 59.9 ± 12.651.9 ± 6.957.6 ± 12.651.0 ± 5.7
130 64.3 ± 16.153.1 ± 9.762.6 ± 16.751.6 ± 8.7
140 69.1 ± 18.554.5 ± 10.667.4 ± 18.653.1 ± 10.0
150 74.5 ± 20.256.0 ± 12.073.0 ± 20.754.5 ± 11.7
160 79.1 ± 19.958.5 ± 13.477.5 ± 20.756.3 ± 12.9
170 84.2 ± 19.064.2 ± 16.382.8 ± 19.862.8 ± 15.7
180 87.2 ± 17.771.2 ± 16.785.7 ± 18.469.8 ± 15.7
190 88.8 ± 16.575.1 ± 16.687.1 ± 17.373.1 ± 16.1
200 89.3 ± 16.477.9 ± 15.287.3 ± 17.274.8 ± 15.1
210 89.5 ± 15.381.8 ± 15.986.9 ± 16.878.4 ± 15.7
220 88.5 ± 15.283.0 ± 15.284.9 ± 16.578.3 ± 14.4
230 89.1 ± 12.884.6 ± 12.884.2 ± 14.578.8 ± 12.7
240 88.7 ± 11.884.5 ± 11.883.0 ± 13.177.6 ± 12.3
250 88.6 ± 10.885.0 ± 11.282.7 ± 12.677.8 ± 11.2
260 88.8 ± 9.485.3 ± 10.881.7 ± 11.576.8 ± 10.8
270 91.9 ± 9.888.8 ± 11.387.2 ± 12.283.3 ± 12.1
280 93.2 ± 9.987.5 ± 15.789.8 ± 12.083.6 ± 15.8
290 94.5 ± 7.486.6 ± 13.791.5 ± 9.183.1 ± 14.1
300 97.8 ± 2.877.3 ± 18.296.8 ± 3.175.9 ± 17.4
310 98.8 ± 1.365.7 ± 14.998.3 ± 1.763.9 ± 13.3
320 99.8 ± 0.265.5 ± 10.399.6 ± 0.566.7 ± 12.0
330 100.0 ± 0.067.5 ± 14.599.9 ± 0.266.9 ± 14.4
340 100.0 ± 0.063.5 ± 19.7100.0 ± 0.063.2 ± 18.3
350 100.0 ± 0.067.7 ± 16.8100.0 ± 0.066.8 ± 18.4
360 100.0 ± 0.066.0 ± 3.1100.0 ± 0.065.2 ± 6.8
370 100.0 ± 0.064.6 ± 6.9100.0 ± 0.065.4 ± 4.2
380 100.0 ± 0.074.4 ± 9.0100.0 ± 0.073.6 ± 5.3
390 100.0 ± 0.069.5 ± 17.5100.0 ± 0.069.6 ± 12.3
400 100.0 ± 0.053.7 ± 11.2100.0 ± 0.055.2 ± 7.4

Downloads