Model Architecture & Training MethodsShazeer et al. (Google Brain)Google

Sparsely-Gated Mixture-of-Experts Layer

Sparsely-Gated Mixture-of-Experts Layer

Why it matters

The first demonstration, through conditional computation, of an architecture whose total parameters far exceed its active parameters; the ancestor of every MoE

Source — Primary sources first; secondary reports are labelled “Press”

Paperarxiv.orghttps://arxiv.org/abs/1701.06538

Added in Week 35, 2026 · See that week

Same week

Arjovsky, Chintala and Bottou publish Wasserstein GAN on arXivPaper

Nearby on the same axis

OpenAI publishes the original GPT (GPT-1) paperOfficialGoogle posts the BERT paper on arXivPaperFast Transformer DecodingPaperGShardPaper