Abstract

1. Introduction

1-1 Autoregressive(AR)

LM의 objective 수식

<aside> 💡 $input\ sequence : x= (x_1, x_2, ..., x_n)$

$forward\ likelihood : p(x) = \sum_{t=1}^T\ p(x_t|x_{<t})$

$training\ objective(forward) : max_θ \ log\ p_θ(x)= max_θ \sum_{t=1}^T log\ p(x_t∣x_{<t})$

</aside>

Likelihood & objective

장단점