[논문 리뷰] Taking Notes on the Fly Helps Language Pre-Training (TNF)

Information

Task: Language Modeling
Publisher: ICLR
Year: 2021

Abstract

1. Creating a note dict for rare words in training corpus makes the pre-training faster and stable
2. Introducing note embedding to update and maintain the note dict

Method

1. Contruction of note dictionary
   
   The left box shows the forward pass with the help of the note dictionary. In the input word sequence, $w_2$ is a rare word. Then for tokens 4 and 5 originated from $w_2$, we query the value of $w_2$ in the note dictionary and weighted average it with token/position embeddings. The right box demonstrates how we maintain the note dictionary. After the forward pass of the model, we can get the contextual representations of the word near $w_2$ and use mean pooling over those representations as the note of $w_2$ in the current sentence. Then, we update $w_2$’s value in the note dictionary by a weighted average of the current note and its previous value.

2. Maintaining of note dictionary
   For a rare word $w$ that appears both in the input token sequence $x = {x_1, …, x_i, …, x_n}$ and note dict, denoting the span boundary of $w$ in $x$ as $(s,t)$, where $s$ and $t$ are the starting and ending position. The note of $w$ for $x$ as $$Note(w,x) = \frac{1}{2k + t - s} \sum_{j=s-k}^{t+k}\textbf{c}_j$$ where $\textbf{c}_j \in \mathbb{R}^d$ si the output of the encoder on position $j$ and served as the contextual representation of $x_j$, $k$ is half of the window size that controls how many surrounding tokens we want to take as notes and save their semantics.