From: Neutron: an attention-based neural decompiler
Hacker’s Delight
Karel
CSmith
Math
NE
Math+NE
Neutron
100%
95.45%
98%
97.4%
97.00%
LSTM+attention
8.69%
81.05%
73%
80%
71.79%
56.67%
LSTM
0
21.7%
37%
31.66%
49.18%