AI Paper Atlas

Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes

Pretraining task diversity and the emergence of non-Bayesian in-context learning for regression

Allan Raventos, Mansheej Paul, Feng Chen, Surya Ganguli

NeurIPS 2023 · Probabilistic and scientific modeling › Uncertainty and Gaussian processes · approach: Bayesian view of meta-learning · 11 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
Increase pretraining task diversity beyond about 2^6 tasksPretrained transformer · Linear regression, new tasksMSE relative to dMMSE and RidgeImproved
Keep diversity low and increase the number of sequences per taskPretrained transformer · Linear regression, new tasksalignment with RidgeHurt
Increase sequences per task while keeping diversity below the thresholdPretrained transformer · Linear regression, tasks drawn from the pretraining distributionprediction similarity to dMMSEImproved
Increase sequences per task while keeping diversity above the thresholdPretrained transformer · Linear regression, new tasksprediction similarity to RidgeImproved
Increase training steps from 500K to 1M at fixed batch sizePretrained transformer · Linear regression, new taskstask diversity thresholdNo clear effect
Increase task dimension at fixed SNRPretrained transformer · Linear regression, new taskstask diversity thresholdHurt
Increase task dimension at M=2^20Pretrained transformer · Linear regression, new tasksprediction difference from RidgeNo clear effect
Increase task dimension at M=2^20dMMSE · Linear regression, new tasksprediction difference from RidgeHurt
Add weight decay (ablation)Pretrained transformer · Linear regression, new taskstask diversity thresholdImproved
Add weight decay (ablation)Pretrained transformer · Linear regression, new tasksperformanceHurt
Increase embedding dimension in the small transformer (ablation)Small transformer · Linear regression, new taskstask diversity thresholdHurt
Increase depth in the small transformer (ablation)Small transformer · Linear regression, new taskstask diversity thresholdHurt

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