AI Paper Atlas

Research map / Architectures and efficiency / RNNs and normalization

PALBERT: Teaching ALBERT to Ponder

Nikita Balagansky, Daniil Gavrilov

NeurIPS 2022 · Architectures and efficiency › RNNs and normalization · approach: Neural scaling laws · 3 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
Replace sampled exit-layer inference with deterministic Q-exit (ablation)PonderNet adapted to ALBERT · SST-2, RTE, CoLAtask score: 91.1 ± 0.6 -> 92.2 ± 0.3; 73.5 ± 1.9 -> 77.3 ± 1.4; 50.8 ± 2.2 -> 55.7 ± 0.9Improved
Use closed-form expectation instead of sampling the exit layer (ablation)PonderNet adapted to ALBERT · SST-2, RTE, CoLAtask score: 91.1 ± 0.6 -> 92.3 ± 0.4; 73.5 ± 1.9 -> 76.8 ± 3.0; 50.8 ± 2.2 -> 55.9 ± 2.2Improved
Fine-tune the Lambda layer with its own learning rate (ablation)PALBERT with Q-exit · SST-2accuracy: 92.2 ± 0.3 -> 92.7 ± 0.4Improved
Fine-tune the Lambda layer with its own learning rate (ablation)PALBERT with Q-exit · RTEaccuracy: 77.3 ± 1.4 -> 77.3 ± 1.4No clear effect
Fine-tune the Lambda layer with its own learning rate (ablation)PALBERT with Q-exit · CoLAMatthews correlation: 55.7 ± 0.9 -> 56.5 ± 1.2Improved
Use a 3-layer rather than 1-layer Lambda MLP (ablation)PALBERT with Lambda LR · SST-2, RTE, CoLAtask score: 92.7 ± 0.4 -> 92.6 ± 0.3; 77.3 ± 1.4 -> 77.0 ± 1.4; 56.5 ± 1.2 -> 56.3 ± 2.4Mixed
Concatenate the current and previous hidden states for Lambda input (ablation)PALBERT with Q-exit and Lambda LR · SST-2, RTE, CoLAtask score: 92.7 ± 0.4 -> 93.0 ± 0.3; 77.3 ± 1.4 -> 76.5 ± 1.6; 56.5 ± 1.2 -> 56.9 ± 1.9Mixed
Use Q-exit and the revised Lambda architecture instead of sampled PonderNetALBERT early-exit models · GLUE testMacro: 77.2 -> 80.6Improved
Use PALBERT instead of ALBERT PABEEALBERT early-exit models · GLUE testMacro: 79.3 -> 80.6Improved
Use PRoBERTa instead of RoBERTa PABEERoBERTa early-exit models · GLUE testMacro: 81.3 -> 82.1Improved
Increase the Q-exit threshold above 0.5 (ablation)PALBERT · GLUE dev taskstask scoreHurt
Use a smaller Q-exit threshold (ablation)PALBERT · GLUE tasksinference speedImproved

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