原书出处
Natural Language Processing in Action, Second Edition
Tip In some cases, it can help to seed your neural networks with an initial hidden state other than all zeros. Jonathan Frankle and Michael Carbin found that being intentional about reuse of good initialization values can be key to helping a network find the global minimum loss achievable for a particular dataset.25 Their approach is to initialize all weights and biases, using a random
原书定位详情
来源:a19c7933-4eda-4176-aebd-30fe7e1ad292
正文位置:extracted/Natural Language Processing in Action, Second Edition/_epub-src/OEBPS/Text/chapter-8.html
EPUB SHA-256:0e9a35e49ca446e322e5524e53f39fb6bee5b2b23d9cda10ef052aca770b0b36