AI Paper Atlas

Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning

On the Importance of Firth Bias Reduction in Few-Shot Classification

Saba Ghaffari, Ehsan Saleh, David Forsyth, Yu-Xiong Wang

ICLR 2022 · Label-efficient and robust learning › Zero-/few-shot and meta-learning · approach: Embedding adaptation for few-shot classification · 4 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
Add Firth bias reductionDC · Cross-dataset few-shot classification on CUB, mini-ImageNet and tiered-ImageNet featuresaccuracy: 0.26 ± 0.03; 1.00 ± 0.04; 0.15 ± 0.03; 1.12 ± 0.03; 0.15 ± 0.04; 0.43 ± 0.02; 0.01 ± 0.01; 0.90 ± 0.02Improved
Add Firth bias reductionDC on tiered-ImageNet · 10-way and 15-way, 1-shot and 5-shot, no artificial samplesaccuracy: 0.63 ± 0.04; 0.33 ± 0.03; 0.67 ± 0.03; 0.46 ± 0.03Improved
Replace uniform Firth prior with the imbalanced class-frequency prior (ablation)Firth-penalized logistic classifier · Imbalanced mini-ImageNet few-shot classificationaccuracyHurt
Use L2 regularization instead of Firth bias reduction (ablation)Few-shot logistic classifier · mini-ImageNet, across reported shots and feature backbonesaccuracyNo clear effect
Use confidence-penalty label smoothing instead of Firth bias reduction (ablation)Logistic classifier with ResNet-10 · Few-shot classification, 5-, 10-, and 15-shotaccuracy improvement: 5-shot: 0.13 ± 0.13 % vs 0.23 ± 0.06 %; 10-shot: 0.52 ± 0.14 % vs 0.73 ± 0.07 %; 15-shot: 0.57 ± 0.18 % vs 1.00 ± 0.07 %Hurt
Use unigram label smoothing instead of Firth bias reduction (ablation)Few-shot logistic classifier · Few-shot image classificationaccuracyHurt
Add Firth bias reductionCosine classifier · mini-ImageNet, tiered-ImageNet, and CIFAR-FS; 5- and 10-shotaccuracy: more than 2.5%Improved

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