New Neural Prior Estimator Learns Class Priors from Latent Representations

Researchers have developed a novel method called the Neural Prior Estimator (NPE) to learn class priors directly from a network's latent representations, rather than relying on explicit class counts. This approach involves attaching lightweight Prior Estimation Modules (PEMs) to the latent space, trained with a logistic objective. Experiments on datasets like CIFAR-10 and CIFAR-100 demonstrated that NPE-LA, the resulting learned logit correction, is competitive with standard methods and improves performance for minority classes. The technique was also shown to be effective as a recalibration mechanism in dense prediction tasks on STARE and ADE20K datasets. AI

IMPACT Introduces a novel technique for handling class imbalance in machine learning models by learning priors from latent representations.

RANK_REASON The cluster contains an academic paper detailing a new method for machine learning. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• ADE20K
• CIFAR-10
• CIFAR-100
• Masoud Yavari
• Neural Prior Estimation
• NPE-LA
• Prior Estimation Modules
• STARE

• paper
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →