The Evolved Transformer

The Evolved Transformer (ET) ia an evolved version of the Transformer architecture created by applying Neural Architecture Search (NAS) algorithm over the standard architecture.The Evolved Transformer was proposed by Google Brain in 2019 and published in their paper with the same name: The Evolved Transformer. The official implementation of the Evolved Transformer can be found on the Tensor2Tensor official GitHub repository: tensor2tensor/

We first construct a large search space inspired by the recent advances in feed-forward sequence models and then run evolutionary architecture search with warm starting by seeding our initial population with the Transformer. To directly search on the computationally expensive WMT 2014 EnglishGerman translation task, we develop the Progressive Dynamic Hurdles method, which allows us to dynamically allocate more resources to more promising candidate models.

The goal of this work is to examine the use of neural architecture search methods to design better feed-forward architectures for seq2seq tasks. Specifically, we apply tournament selection architecture search and warm start it with the Transformer, considered to be the state-of-art and widely-used, to evolve a better and more efficient architecture. To achieve this, we construct a search space that reflects the recent advances in feed-forward seq2seq models and develop a method called Progressive Dynamic Hurdles (PDH) that allows us to perform our search directly on the computationally demanding WMT 2014 EnglishGerman (En-De) translation task.

ET out-performed the human-designed transformer on four well-established language tasks: WMT 2014 English-German, WMT 2014 English-French, WMT 2014 English-Czech and LM1B:


NAS stands for “Network Architecture Search” which is an evolution-based architecture search