imagining syntax

← schedopt_v5_confirm

free-00c20121

done · peak unseen_mismatch 90.8

Referenced by

α schedule

α 4.4 ×10 α 4.5 ×10 α 4.7 ×10 α 4.9 ×10 α 4.7 ×10 α 5.0 ×10 α 4.8 ×10 α 4.9 ×10 α 4.8 ×10 α 4.7 ×20 α 4.9 ×10 α 4.7 ×10 α 4.9 ×10 α 4.7 ×10 α 4.9 ×10 α 0.8 ×10 α 0.9 ×10 α 0.8 ×10 α 1.0 ×10 α 0.5 ×10 α 1.0 ×10 α 0.7 ×30 α 0.9 ×20 α 0.8 ×10 α 0.9 ×120

direction
mixed
α range
0.5–5.0
stages
25

Figures

schedule.png
schedule.png

Results

accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 0.9

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

iterationseen_matchseen_mismatchunseen_matchunseen_mismatch
100 60.8 ± 7.757.1 ± 6.254.4 ± 5.851.7 ± 3.5
110 65.7 ± 9.162.0 ± 7.856.7 ± 7.553.6 ± 5.3
120 70.3 ± 9.867.0 ± 8.759.6 ± 8.756.1 ± 6.6
130 74.4 ± 8.471.1 ± 7.561.7 ± 8.658.1 ± 6.7
140 77.2 ± 7.374.3 ± 6.763.1 ± 8.459.9 ± 6.6
150 78.7 ± 5.676.1 ± 5.063.5 ± 7.560.7 ± 6.1
160 79.4 ± 5.077.2 ± 4.363.8 ± 7.061.0 ± 5.8
170 83.0 ± 6.480.5 ± 5.969.3 ± 9.266.3 ± 8.0
180 90.2 ± 7.887.7 ± 7.782.5 ± 11.178.8 ± 10.9
190 95.1 ± 7.292.7 ± 7.691.3 ± 9.788.0 ± 10.1
200 96.8 ± 6.394.4 ± 7.394.3 ± 8.590.5 ± 9.2
210 97.4 ± 5.793.8 ± 8.295.8 ± 7.590.8 ± 9.3
220 97.4 ± 3.383.0 ± 12.296.6 ± 4.680.5 ± 12.1
230 98.7 ± 2.575.6 ± 19.198.2 ± 3.873.0 ± 19.3
240 99.7 ± 0.564.4 ± 20.199.8 ± 0.462.2 ± 19.0
250 100.0 ± 0.066.5 ± 21.1100.0 ± 0.163.5 ± 21.4
260 100.0 ± 0.072.5 ± 13.8100.0 ± 0.070.6 ± 13.4
270 100.0 ± 0.085.4 ± 6.8100.0 ± 0.083.0 ± 6.2
280 100.0 ± 0.087.5 ± 5.9100.0 ± 0.085.1 ± 4.9
290 100.0 ± 0.088.4 ± 8.2100.0 ± 0.086.3 ± 6.6
300 100.0 ± 0.068.2 ± 9.999.9 ± 0.266.8 ± 7.4
310 100.0 ± 0.076.8 ± 15.4100.0 ± 0.075.5 ± 15.4
320 100.0 ± 0.085.1 ± 18.2100.0 ± 0.084.2 ± 19.3
330 100.0 ± 0.084.9 ± 0.0100.0 ± 0.085.1 ± 0.0
340 100.0 ± 0.074.9 ± 0.0100.0 ± 0.073.4 ± 0.0
350 100.0 ± 0.060.8 ± 0.0100.0 ± 0.059.1 ± 0.0
360 100.0 ± 0.059.4 ± 0.0100.0 ± 0.055.9 ± 0.0
370 100.0 ± 0.076.1 ± 0.0100.0 ± 0.076.4 ± 0.0
380 100.0 ± 0.077.1 ± 0.0100.0 ± 0.075.9 ± 0.0
390 100.0 ± 0.055.2 ± 0.0100.0 ± 0.050.8 ± 0.0
400 100.0 ± 0.087.7 ± 0.0100.0 ± 0.088.0 ± 0.0

Downloads