imagining syntax

← schedopt_v5_confirm

free-82f9c620

done · peak unseen_mismatch 96.8

α schedule

α 4.7 ×10 α 4.3 ×10 α 4.7 ×10 α 4.8 ×10 α 4.1 ×10 α 4.5 ×10 α 4.3 ×10 α 4.6 ×10 α 4.7 ×10 α 4.3 ×10 α 4.6 ×10 α 4.1 ×10 α 5.0 ×10 α 4.3 ×10 α 4.7 ×10 α 4.9 ×10 α 4.7 ×10 α 1.8 ×10 α 0.2 ×10 α 0.3 ×10 α 0.1 ×10 α 0.4 ×10 α 2.5 ×10 α 4.2 ×10 α 4.1 ×10 α 4.8 ×10 α 1.8 ×10 α 0.1 ×10 α 0.6 ×10 α 0.5 ×110

direction
mixed
α range
0.1–5.0
stages
30

Figures

schedule.png
schedule.png

Results

accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 0.5

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

iterationseen_matchseen_mismatchunseen_matchunseen_mismatch
100 60.3 ± 7.556.7 ± 6.154.2 ± 5.751.7 ± 3.4
110 64.9 ± 8.961.2 ± 7.756.3 ± 7.553.6 ± 4.9
120 70.0 ± 9.566.4 ± 8.459.5 ± 8.756.2 ± 6.3
130 74.5 ± 9.271.3 ± 8.462.4 ± 9.059.0 ± 7.0
140 77.2 ± 8.074.3 ± 7.363.8 ± 8.960.6 ± 7.1
150 79.1 ± 6.876.5 ± 6.164.5 ± 8.461.6 ± 6.9
160 80.0 ± 5.577.5 ± 4.964.3 ± 7.461.7 ± 6.4
170 80.3 ± 4.778.0 ± 4.064.3 ± 6.861.7 ± 5.6
180 83.7 ± 5.181.3 ± 4.668.1 ± 7.765.3 ± 6.7
190 88.5 ± 6.586.2 ± 6.678.3 ± 10.474.9 ± 10.1
200 94.8 ± 6.192.3 ± 7.591.0 ± 8.787.7 ± 9.8
210 97.4 ± 4.693.4 ± 9.396.1 ± 5.891.1 ± 10.4
220 99.1 ± 1.383.8 ± 17.998.5 ± 2.581.4 ± 18.6
230 99.2 ± 2.489.5 ± 9.099.0 ± 2.986.4 ± 9.4
240 98.5 ± 1.694.0 ± 5.396.8 ± 3.689.5 ± 7.3
250 92.6 ± 4.586.2 ± 3.083.2 ± 11.373.6 ± 7.1
260 90.1 ± 4.686.5 ± 4.180.0 ± 10.073.8 ± 8.9
270 95.5 ± 2.590.9 ± 0.889.7 ± 5.382.5 ± 1.6
280 99.8 ± 0.398.4 ± 1.399.3 ± 0.596.8 ± 1.4

Downloads