Long-Tailed Distribution
A distribution phenomenon in machine learning in which class sample counts are highly imbalanced: a few head classes account for most samples while numerous tail classes have very few, causing degraded model performance on tail classes. Common countermeasures include resampling and loss weighting.
Timeline (last 90 days)
Common strategies for dealing with long-tail distributions include resampling, loss function weighting (such as focal loss), classifier rebalancing (such as logit adjustment), and merging tail categories into superclasses.
1,900 categories may lead to a highly imbalanced category distribution, with some rare categories having very few samples that the model struggles to learn well.
The benefit of a cascade architecture comes from the long-tail distribution of task difficulty: the vast majority of requests fall into simple categories, so reserving expensive models for the few hard samples lowers average cost with almost no change in quality.
In systematic bias, data leakage and duplication, extreme long-tail, and safety-sensitive scenarios, data cleaning remains irreplaceable, because the network cannot correct structural errors on its own.
Academia generally holds that the long-tail problem in autonomous driving cannot be completely eliminated, but can only have its probability of occurrence continually reduced.
The academic community generally believes that the long-tail problem cannot be completely eliminated, only that its probability of occurrence can be continuously reduced.
The long-tail distribution problem is a common challenge faced by all large-scale AI models: models perform excellently on head scenarios with abundant data but are prone to hallucination on tail scenarios with scarce data.
All Facts (7)
Common strategies for dealing with long-tail distributions include resampling, loss function weighting (such as focal loss), classifier rebalancing (such as logit adjustment), and merging tail categories into superclasses.
50%Unverified1,900 categories may lead to a highly imbalanced category distribution, with some rare categories having very few samples that the model struggles to learn well.
50%UnverifiedThe benefit of a cascade architecture comes from the long-tail distribution of task difficulty: the vast majority of requests fall into simple categories, so reserving expensive models for the few hard samples lowers average cost with almost no change in quality.
50%UnverifiedIn systematic bias, data leakage and duplication, extreme long-tail, and safety-sensitive scenarios, data cleaning remains irreplaceable, because the network cannot correct structural errors on its own.
50%UnverifiedAcademia generally holds that the long-tail problem in autonomous driving cannot be completely eliminated, but can only have its probability of occurrence continually reduced.
50%UnverifiedThe academic community generally believes that the long-tail problem cannot be completely eliminated, only that its probability of occurrence can be continuously reduced.
50%UnverifiedThe long-tail distribution problem is a common challenge faced by all large-scale AI models: models perform excellently on head scenarios with abundant data but are prone to hallucination on tail scenarios with scarce data.
50%