Self-StrAE at SemEval-2024 Task 1: Making Self-Structuring AutoEncoders Learn More With Less
This work addresses efficient representation learning for NLP tasks, but it is incremental as it builds on an existing method with specific optimizations.
The paper tackled improving Self-Structuring AutoEncoders by adding reconstruction as an auxiliary objective and increasing independent channels, resulting in enhanced embedding quality with fewer parameters, such as reducing non-embedding parameters to seven and effective pre-training with 10M tokens across multiple languages.
This paper presents two simple improvements to the Self-Structuring AutoEncoder (Self-StrAE). Firstly, we show that including reconstruction to the vocabulary as an auxiliary objective improves representation quality. Secondly, we demonstrate that increasing the number of independent channels leads to significant improvements in embedding quality, while simultaneously reducing the number of parameters. Surprisingly, we demonstrate that this trend can be followed to the extreme, even to point of reducing the total number of non-embedding parameters to seven. Our system can be pre-trained from scratch with as little as 10M tokens of input data, and proves effective across English, Spanish and Afrikaans.