Backpropagation Through Time

التحديث التراجعي عبر الزمن

خوارزمية تدريب الشبكات التكرارية عبر فرد الشبكة على كل الخطوات الزمنية ثم تطبيق قاعدة السلسلة (الانتشار العكسي) من الخطوة الأخيرة إلى الأولى. عملية الفرد هذه هي ما يُنتج جُداء مصفوفات يعقوبي.

An algorithm for training RNNs by unrolling the network across all time steps and applying the chain rule (backpropagation) from the last step to the first. This unrolling produces the product of Jacobian matrices.

Also translated asالانتشار العكسي عبر الزمن، BPTT

First appears in this corpus in: Backpropagation Through Time: What It Does and How to Do It (1990)

Appears in these papers

Backpropagation Through Time: What It Does and How to Do It1990 in the sky ✦
Backpropagation Through Time: What It Does and How to Do It1990 in the sky ✦
On the Difficulty of Training Recurrent Neural Networks2013 in the sky ✦
Growing Neural Cellular Automata2020 in the sky ✦
Growing Neural Cellular Automata2020 in the sky ✦
Neural Turing Machines2014 in the sky ✦
Recurrent Neural Networks (RNNs): A Gentle Introduction and Overview2019 in the sky ✦
Recurrent Neural Networks (RNNs): A Gentle Introduction and Overview2019 in the sky ✦
Universal Language Model Fine-Tuning for Text Classification2018 in the sky ✦