Built independently by an author, for readers. Read the story and support ChapterPal
ChapterPal Logo
Log inSign up

keyword

masked auto-encoder pre-training

Masked auto-encoder pre-training is a self-supervised learning paradigm in which a neural network learns generalized data representations by reconstructing missing or masked portions of an input from the visible parts. Under this framework, a subset of input elements, such as image patches or textual tokens, is intentionally masked out. An encoder transforms the visible or partially corrupted input into latent semantic embeddings, after which a decoder reconstructs the original, uncorrupted data using those embeddings. By requiring the model to infer hidden content from contextual cues, masked auto-encoder pre-training effectively captures high-level semantics, global dependencies, and underlying structures, producing robust foundational representations that can be fine-tuned for diverse downstream tasks across computer vision and natural language processing.

1 item

RetroMAE: Pre-Training Retrieval-oriented Language Models Via Masked Auto-Encoder

RetroMAE: Pre-Training Retrieval-oriented Language Models Via Masked Auto-Encoder

Shitao Xiao, Zheng Liu, Yingxia Shao, Zhao Cao

OrganizationsBeijing University of Posts and TelecommunicationsHuawei

Why you should read this

Proposes an asymmetric masked auto-encoder pre-training framework that forces language models to generate superior sentence embeddings for dense retrieval, establishing state-of-the-art performance on BEIR and MS MARCO benchmarks.

Despite pre-training’s progress in many important NLP tasks, it remains to explore effective pre-training strategies for dense retrieval. In this paper, we propose RetroMAE, a new retrieval oriented pre-training paradigm based on Masked Auto-Encoder (MAE). RetroMAE is highlighted by three critical designs. 1) A novel MAE workflow, where the input sentence is polluted for encoder and decoder with different masks. The sentence embedding is generated from the encoder’s masked input; then, the original sentence is recovered based on the sentence embedding and the decoder’s masked input via masked language modeling. 2) Asymmetric model structure, with a full-scale BERT like transformer as encoder, and a one-layer transformer as decoder. 3) Asymmetric masking ratios, with a moderate ratio for encoder: 15~30%, and an aggressive ratio for decoder: 50~70%. Our framework is simple to realize and empirically competitive: the pre-trained models dramatically improve the SOTA performances on a wide range of dense retrieval benchmarks, like BEIR and MS MARCO. The source code and pre-trained models are made publicly available at https://github.com/staoxiao/RetroMAE so as to inspire more interesting research.

Added

2026-09-30