AI Paper Atlas

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

Learning To Hallucinate Examples From Extrinsic and Intrinsic Supervision

Liangke Gui, Adrien Bardes, Ruslan Salakhutdinov, Alexander Hauptmann, Martial Hebert, Yu-Xiong Wang

ICCV 2021 · Label-efficient and robust learning › Zero-/few-shot and meta-learning · approach: Fine-tuning strategies for few-shot · 5 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 intrinsic contrastive supervision (DMAS w/ in) to a cosine classifier with plain features (ablation)Cosine Classifier · ImageNet1K 311-way, 1-shottop-5 accuracy: 37.8 -> 43.4Improved
add extrinsic mentor guidance (DMAS w/ ex) to the cosine classifier (ablation)Cosine Classifier · ImageNet1K 311-way, 1-shottop-5 accuracy: 37.8 -> 44.5Improved
add mentor guidance to the plain hallucinator in place of classification loss only (ablation)Cosine Classifier · ImageNet1K 311-way, 1-shottop-5 accuracy: 42.6 -> 44.5Improved
add progressive guidance by weakening the student on top of mentor guidance (ablation)Cosine Classifier DMAS w/ ex · ImageNet1K 311-way, 1-shottop-5 accuracy: 44.5 -> 45.4Improved
add progressive guidance by strengthening the mentor on top of mentor guidance (ablation)Cosine Classifier DMAS w/ ex · ImageNet1K 311-way, 1-shottop-5 accuracy: 44.5 -> 44.3No clear effect
weaken the student progressively instead of strengthening the mentor (ablation)Cosine Classifier DMAS · ImageNet1K 311-way, 1-shottop-5 accuracy: 44.3 -> 45.4Improved
add intrinsic contrastive loss to mentor-guided progressive training (full DMAS) (ablation)Cosine Classifier DMAS w/ ex↓ · ImageNet1K 311-way, 1-shottop-5 accuracy: 45.4 -> 46.5Improved
use DMAS hallucinator instead of plain hallucinator with ProtoNet (ablation)ProtoNet · ImageNet1K 311-way, 1-shottop-5 accuracy: 45.0 -> 45.9Improved
pretrain feature backbone with cosine classifier instead of standard cross-entropy, then apply DMAS (ablation)Cosine Classifier DMAS (full) · ImageNet1K 311-way, 1-shottop-5 accuracy: 46.5 -> 47.9Improved
learned hallucination instead of hand-crafted data augmentation (ablation)ProtoNet · ImageNet1K 311-way, 1-shottop-5 accuracy: 40.2 -> 45.0Improved
add plain hallucinator Gen to DeepEMDDeepEMD · miniImageNet 5-way 1-shotaccuracy: 65.91 -> 64.73Hurt
add DMAS hallucinator to MetaOptNet instead of plain GenMetaOptNet · miniImageNet 5-way 1-shotaccuracy: 62.64 -> 64.55Improved

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