AI Paper Atlas

Research map / Architectures and efficiency / Quantization

BiT: Robustly Binarized Multi-distilled Transformer

Zechun Liu, Barlas Oguz, Aasish Pappu, Lin Xiao, Scott Yih, Meng Li, Raghuraman Krishnamoorthi, Yashar Mehdad

NeurIPS 2022 · Architectures and efficiency › Quantization · approach: Binary neural networks · 20 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
remove attention-score distillation (keep logits and block outputs only) (ablation)binary BERT baseline · GLUE avgavg score: 45.7 -> 48.0Improved
add two-set binarization (non-negative activations to {0,1}) (ablation)binary BERT baseline · GLUE avgavg score: 48.0 -> 55.3Improved
add elastic binarization with learned scale and threshold (ablation)BiT strong baseline · GLUE avgavg score: 55.3 -> 71.0Improved
add multi-distillation W32A32 -> W1A2 -> W1A1 (ablation)BiT W1A1 · GLUE avgavg score: 71.0 -> 73.5Improved
learn scale alpha in binarization (ablation)two-set strong baseline · GLUE avgavg score: 55.3 -> 69.2Improved
additionally learn bias/threshold beta (ablation)BiT with learnable scale · GLUE avgavg score: 69.2 -> 71.0Improved
binarize attention after softmax to {0,1} instead of removing softmax (Bi-Attention) (ablation)binary BERT · GLUE avgavg score: 51.3 -> 53.0Improved
binarize ReLU outputs to {0,1} instead of {-1,1} (ablation)binary BERT · GLUE avgavg score: 53.0 -> 55.3Improved
quantized teacher with fixed student init vs progressive distillation with full-precision teacher (ablation)BiT · GLUE avgavg score: 73.4 -> 73.5Improved
three-step schedule W1A4 -> W1A2 -> W1A1 instead of two-step (ablation)BiT · MNLI-maccuracyNo clear effect
add data augmentation for small GLUE datasetsBiT W1A1 · GLUE avgavg score: 73.5 -> 78.0Improved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy