TasksSotADatasetsPapersMethodsSubmitAbout
Papers With Code 2

A community resource for machine learning research: papers, code, benchmarks, and state-of-the-art results.

Explore

Notable BenchmarksAll SotADatasetsPapersMethods

Community

Submit ResultsAbout

Data sourced from the PWC Archive (CC-BY-SA 4.0). Built by the community, for the community.

Papers/Difficulty-Net: Learning to Predict Difficulty for Long-Ta...

Difficulty-Net: Learning to Predict Difficulty for Long-Tailed Recognition

Saptarshi Sinha, Hiroki Ohashi

2022-09-07Meta-LearningLong-tail Learning
PaperPDFCode(official)

Abstract

Long-tailed datasets, where head classes comprise much more training samples than tail classes, cause recognition models to get biased towards the head classes. Weighted loss is one of the most popular ways of mitigating this issue, and a recent work has suggested that class-difficulty might be a better clue than conventionally used class-frequency to decide the distribution of weights. A heuristic formulation was used in the previous work for quantifying the difficulty, but we empirically find that the optimal formulation varies depending on the characteristics of datasets. Therefore, we propose Difficulty-Net, which learns to predict the difficulty of classes using the model's performance in a meta-learning framework. To make it learn reasonable difficulty of a class within the context of other classes, we newly introduce two key concepts, namely the relative difficulty and the driver loss. The former helps Difficulty-Net take other classes into account when calculating difficulty of a class, while the latter is indispensable for guiding the learning to a meaningful direction. Extensive experiments on popular long-tailed datasets demonstrated the effectiveness of the proposed method, and it achieved state-of-the-art performance on multiple long-tailed datasets.

Results

TaskDatasetMetricValueModel
Image ClassificationPlaces-LTTop-1 Accuracy41.7Difficulty-Net (ResNet-152)
Image ClassificationCIFAR-100-LT (ρ=50)Error Rate43.1Difficulty-Net
Image ClassificationCIFAR-100-LT (ρ=10)Error Rate34.78Difficulty-Net
Image ClassificationImageNet-LTTop-1 Accuracy57.4Difficulty-Net (ResNet-50 using RandAugment, single model)
Image ClassificationImageNet-LTTop-1 Accuracy54Difficulty-Net (ResNet-50 w/o using RandAugment, single model)
Image ClassificationImageNet-LTTop-1 Accuracy44.6Difficulty-Net (ResNet-10 w/o using RandAugment, single model
Image ClassificationCIFAR-100-LT (ρ=100)Error Rate47.04Difficulty-Net
Few-Shot Image ClassificationPlaces-LTTop-1 Accuracy41.7Difficulty-Net (ResNet-152)
Few-Shot Image ClassificationCIFAR-100-LT (ρ=50)Error Rate43.1Difficulty-Net
Few-Shot Image ClassificationCIFAR-100-LT (ρ=10)Error Rate34.78Difficulty-Net
Few-Shot Image ClassificationImageNet-LTTop-1 Accuracy57.4Difficulty-Net (ResNet-50 using RandAugment, single model)
Few-Shot Image ClassificationImageNet-LTTop-1 Accuracy54Difficulty-Net (ResNet-50 w/o using RandAugment, single model)
Few-Shot Image ClassificationImageNet-LTTop-1 Accuracy44.6Difficulty-Net (ResNet-10 w/o using RandAugment, single model
Few-Shot Image ClassificationCIFAR-100-LT (ρ=100)Error Rate47.04Difficulty-Net
Generalized Few-Shot ClassificationPlaces-LTTop-1 Accuracy41.7Difficulty-Net (ResNet-152)
Generalized Few-Shot ClassificationCIFAR-100-LT (ρ=50)Error Rate43.1Difficulty-Net
Generalized Few-Shot ClassificationCIFAR-100-LT (ρ=10)Error Rate34.78Difficulty-Net
Generalized Few-Shot ClassificationImageNet-LTTop-1 Accuracy57.4Difficulty-Net (ResNet-50 using RandAugment, single model)
Generalized Few-Shot ClassificationImageNet-LTTop-1 Accuracy54Difficulty-Net (ResNet-50 w/o using RandAugment, single model)
Generalized Few-Shot ClassificationImageNet-LTTop-1 Accuracy44.6Difficulty-Net (ResNet-10 w/o using RandAugment, single model
Generalized Few-Shot ClassificationCIFAR-100-LT (ρ=100)Error Rate47.04Difficulty-Net
Long-tail LearningPlaces-LTTop-1 Accuracy41.7Difficulty-Net (ResNet-152)
Long-tail LearningCIFAR-100-LT (ρ=50)Error Rate43.1Difficulty-Net
Long-tail LearningCIFAR-100-LT (ρ=10)Error Rate34.78Difficulty-Net
Long-tail LearningImageNet-LTTop-1 Accuracy57.4Difficulty-Net (ResNet-50 using RandAugment, single model)
Long-tail LearningImageNet-LTTop-1 Accuracy54Difficulty-Net (ResNet-50 w/o using RandAugment, single model)
Long-tail LearningImageNet-LTTop-1 Accuracy44.6Difficulty-Net (ResNet-10 w/o using RandAugment, single model
Long-tail LearningCIFAR-100-LT (ρ=100)Error Rate47.04Difficulty-Net
Generalized Few-Shot LearningPlaces-LTTop-1 Accuracy41.7Difficulty-Net (ResNet-152)
Generalized Few-Shot LearningCIFAR-100-LT (ρ=50)Error Rate43.1Difficulty-Net
Generalized Few-Shot LearningCIFAR-100-LT (ρ=10)Error Rate34.78Difficulty-Net
Generalized Few-Shot LearningImageNet-LTTop-1 Accuracy57.4Difficulty-Net (ResNet-50 using RandAugment, single model)
Generalized Few-Shot LearningImageNet-LTTop-1 Accuracy54Difficulty-Net (ResNet-50 w/o using RandAugment, single model)
Generalized Few-Shot LearningImageNet-LTTop-1 Accuracy44.6Difficulty-Net (ResNet-10 w/o using RandAugment, single model
Generalized Few-Shot LearningCIFAR-100-LT (ρ=100)Error Rate47.04Difficulty-Net

Related Papers

Are encoders able to learn landmarkers for warm-starting of Hyperparameter Optimization?2025-07-16Imbalanced Regression Pipeline Recommendation2025-07-16CLID-MU: Cross-Layer Information Divergence Based Meta Update Strategy for Learning with Noisy Labels2025-07-16Mixture of Experts in Large Language Models2025-07-15Iceberg: Enhancing HLS Modeling with Synthetic Data2025-07-14Meta-Reinforcement Learning for Fast and Data-Efficient Spectrum Allocation in Dynamic Wireless Networks2025-07-13Geo-ORBIT: A Federated Digital Twin Framework for Scene-Adaptive Lane Geometry Detection2025-07-11The Bayesian Approach to Continual Learning: An Overview2025-07-11