AI Paper Atlas

Research map / Architectures and efficiency / Pruning and efficient tuning

LIFT: Language-Interfaced Fine-Tuning for Non-language Machine Learning Tasks

Tuan Dinh, Yuchen Zeng, Ruisu Zhang, Ziqian Lin, Michael Gira, Shashank Rajput, Jy-yong Sohn, Dimitris Papailiopoulos, Kangwook Lee

NeurIPS 2022 · Architectures and efficiency › Pruning and efficient tuning · approach: Adapter-based parameter-efficient tuning · 35 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
fine-tune pretrained GPT-J on non-language tasks via sentence conversion (LIFT) vs logistic regressionLIFT/GPT-3 · circles classificationaccuracy: 48.58 -> 81.17Improved
increase number of classes to 100LIFT/GPT-J · Marginaccuracy: 50.23 vs 81.98 (RBF-SVM)Hurt
fine-tune on full dataset instead of in-context learning with few prompts (ablation)LIFT/GPT-J · Vehicleaccuracy: 25.49 -> 64.31Improved
fine-tune rather than in-context learning at same sample count (ablation)LIFT/GPT-3 · Vehicleaccuracy: 28.82 -> 23.73No clear effect
use pretrained GPT-J instead of random initialization (ablation)LIFT/GPT-J · Wineaccuracy: 22.22 -> 93.52Improved
use LM pretrained on natural language instead of code (CodeGen) (ablation)LIFT · Two Circlesaccuracy: 53.02 -> 75.92Improved
fine-tune GPT-J on gibberish data first (ablation)LIFT/GPT-J · Two Circlesaccuracy: 75.92 -> 68.67Hurt
add correct feature names to prompt (ablation)LIFT/GPT-3 · TAEaccuracy: 65.59 -> 69.89Improved
add correct feature names to prompt (ablation)LIFT/GPT-3 · CMCaccuracy: 57.74 -> 57.40No clear effect
use shuffled feature names instead of correct ones (ablation)LIFT/GPT-3 · TAEaccuracy: 69.89 -> 60.22Hurt
add Gaussian-noise data augmentation (eps=0.1) to training (ablation)LIFT/GPT-J · MNIST Gaussian noise eps=0.1accuracy: 56.14 -> 94.98Improved
add Gaussian-noise data augmentation (eps=0.1) to training (ablation)LIFT/GPT-J · MNIST cleanaccuracy: 96.88 -> 93.78Hurt

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