imagining syntax

← schedopt_v5_confirm

free-0c31fbbc

done · peak unseen_mismatch 92.5

Referenced by

α schedule

α 5.0 ×10 α 4.7 ×10 α 4.9 ×10 α 4.6 ×10 α 4.8 ×10 α 4.4 ×10 α 4.9 ×10 α 4.8 ×10 α 4.5 ×10 α 4.6 ×10 α 4.7 ×20 α 4.3 ×10 α 4.9 ×10 α 5.0 ×10 α 4.9 ×10 α 0.8 ×10 α 0.9 ×10 α 0.7 ×10 α 0.8 ×10 α 0.7 ×10 α 1.1 ×10 α 0.8 ×10 α 1.5 ×10 α 0.9 ×10 α 0.3 ×10 α 0.5 ×10 α 1.1 ×10 α 0.6 ×10 α 1.0 ×110

direction
mixed
α range
0.3–5.0
stages
29

Figures

schedule.png
schedule.png

Results

accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 1.0

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

iterationseen_matchseen_mismatchunseen_matchunseen_mismatch
100 61.0 ± 7.757.4 ± 6.454.4 ± 5.851.8 ± 3.7
110 65.6 ± 9.261.9 ± 8.256.5 ± 7.753.8 ± 5.4
120 70.5 ± 9.566.9 ± 8.559.6 ± 8.456.3 ± 6.2
130 74.9 ± 8.771.5 ± 7.862.1 ± 8.958.7 ± 6.8
140 77.5 ± 7.374.7 ± 6.663.4 ± 8.460.2 ± 6.5
150 78.7 ± 6.176.1 ± 5.463.6 ± 7.560.7 ± 5.8
160 79.7 ± 5.177.3 ± 4.563.8 ± 6.860.9 ± 5.6
170 83.2 ± 6.580.8 ± 6.269.8 ± 9.166.5 ± 8.1
180 90.7 ± 7.088.0 ± 7.583.0 ± 10.779.1 ± 10.8
190 95.4 ± 6.393.0 ± 7.191.9 ± 8.888.3 ± 9.3
200 97.3 ± 5.194.6 ± 6.495.1 ± 6.990.7 ± 8.0
210 98.0 ± 4.190.8 ± 10.696.7 ± 5.488.1 ± 10.6
220 99.1 ± 1.287.3 ± 11.598.3 ± 2.084.6 ± 12.3
230 100.0 ± 0.082.8 ± 15.399.9 ± 0.280.4 ± 14.8
240 100.0 ± 0.081.3 ± 13.4100.0 ± 0.180.0 ± 13.4
250 100.0 ± 0.090.3 ± 11.5100.0 ± 0.088.8 ± 12.2
260 100.0 ± 0.072.0 ± 4.4100.0 ± 0.070.5 ± 1.2
270 100.0 ± 0.063.9 ± 4.3100.0 ± 0.063.8 ± 1.6
280 100.0 ± 0.059.0 ± 4.0100.0 ± 0.059.9 ± 2.0
290 100.0 ± 0.071.7 ± 4.0100.0 ± 0.073.0 ± 2.6
300 100.0 ± 0.081.2 ± 13.3100.0 ± 0.082.3 ± 12.8
310 100.0 ± 0.073.0 ± 0.0100.0 ± 0.072.4 ± 0.0
320 100.0 ± 0.093.0 ± 0.0100.0 ± 0.092.5 ± 0.0

Downloads