REVIEW 16 cited by
Long-tail learning via logit adjustment
Not yet reviewed by Pith; the record is open.
This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.
SPECIMEN: schema-true, not a live event
T0 review · schema-true
One-sentence machine reading of the paper's core claim.
pith:XXXXXXXX · record.json · timestamp
read the original abstract
Real-world classification problems typically exhibit an imbalanced or long-tailed label distribution, wherein many labels are associated with only a few samples. This poses a challenge for generalisation on such labels, and also makes na\"ive learning biased towards dominant labels. In this paper, we present two simple modifications of standard softmax cross-entropy training to cope with these challenges. Our techniques revisit the classic idea of logit adjustment based on the label frequencies, either applied post-hoc to a trained model, or enforced in the loss during training. Such adjustment encourages a large relative margin between logits of rare versus dominant labels. These techniques unify and generalise several recent proposals in the literature, while possessing firmer statistical grounding and empirical performance.
Forward citations
Cited by 16 Pith papers
-
Recurrent Contrastive Learning for Imbalanced Medical Image Classification
A recurrent contrastive learning framework with temporal memory queues and anchors improves balanced accuracy for imbalanced medical image classification.
-
From Perturbation Correction to Geometry-Aware Sampling: Sharpness-Guided Equilibrium Sampling for Balanced Flat Minima in Long-Tailed Learning
Sharpness-Guided Equilibrium Sampling reweights long-tailed training batches using cumulative class counts and SAM perturbation-loss gaps, improving tail accuracy by up to 10.8 points.
-
Loss Landscape Topology Reveals Why Simple Baselines are Competitive at 3D Point Cloud Segmentation Under Class Imbalance
Uniform cross-entropy is competitive with 11 imbalance-aware losses in point-based 3D segmentation; specialized methods give small, architecture-dependent gains.
-
Revisiting Scene Graph Generation from the Perspective of Detector-Conditioned Reachability
A dual-query scene graph generation method unifies detector-based and query-based reasoning in a single decoder, achieving state-of-the-art results on Visual Genome, Open Images v6, and GQA-200.
-
Generative Distribution Distillation
Knowledge distillation is reformulated as conditional diffusion over teacher feature tokens, with class-center contraction replacing the classification loss, yielding state-of-the-art ImageNet distillation numbers.
-
MoMBS: Mixed-order minibatch sampling enhances model training from diverse-quality images
Pairing high-difficulty with low-difficulty images, ranked by loss and uncertainty, makes minibatch training more effective and improves accuracy on four computer vision tasks.
-
Mutually Exclusive Multiclass Lesion Segmentation in Neuroimaging: Binary-Guided Weak Supervision with Inter-Class Orthogonality
Binary-guided mutual exclusivity with inter-class orthogonality yields accurate multiclass weakly supervised neuroimaging lesion segmentation from image-level labels alone.
-
SciLT: Long-tailed Image Classification under Scientific Image Domains
On scientific long-tailed image tasks, foundation-model fine-tuning gains are limited; SciLT fuses penultimate and final ViT features under dual supervision to improve balanced accuracy.
-
Addressing Imbalanced Domain-Incremental Learning through Dual-Balance Collaborative Experts
DCE trains frequency-aware experts with complementary losses plus a Gaussian-sampled dynamic selector, reporting SOTA accuracy on four imbalanced domain-incremental benchmarks.
-
Mixture Experts with Test-Time Self-Supervised Aggregation for Tabular Imbalanced Regression
A mixture-of-experts model with test-time self-supervised weight adjustment improves tabular imbalanced regression under three different test distributions, reporting a 7.1% average MAE gain.
-
Tuning the Right Foundation Models is What you Need for Partial Label Learning
Fine-tuning foundation models such as CLIP gives large, robust gains in partial label learning and makes the choice of PLL algorithm nearly irrelevant.
-
Synthetic Data Augmentation using Pre-trained Diffusion Models for Long-tailed Food Image Classification
A two-stage diffusion-based data augmentation pipeline with confusing-class negative prompts improves long-tailed food image classification accuracy on Food101-LT and VFN-LT.
-
Aligned Contrastive Loss for Long-Tailed Recognition
ACL modifies supervised contrastive loss by removing non-effective positives from the denominator, adding class centers, and re-weighting negatives by inverse class frequency, yielding new state-of-the-art accuracy on...
-
Mixture of Balanced Information Bottlenecks for Long-Tailed Visual Recognition
A balanced information bottleneck loss, extended to a mixture over intermediate layers, improves reported accuracy on three long-tailed visual recognition benchmarks.
-
Subtyping Breast Lesions via Generative Augmentation based Long-tailed Recognition in Ultrasound
A dual-phase framework with a sketch-guided diffusion synthesizer and an RL-based adaptive sampler improves long-tailed breast ultrasound subtype classification over the tested baselines.
-
Learning from Limited and Imperfect Data
A doctoral thesis compiling nine peer-reviewed papers on long-tailed image generation, long-tailed recognition, semi-supervised learning, and domain adaptation.
Discussion (0). Sign in to comment.