Discriminative Model
A class of models that directly compute posterior probabilities over predefined candidate options, as opposed to generative models. Discriminative models are commonly used in routing, content moderation, and relevance ranking scenarios that require structured output; they offer low latency and naturally comparable outputs.
Core Facts
Timeline (last 90 days)
Discriminative models (such as face detection and closed-eye classifiers) are only responsible for judging "whether the eyes are closed," outputting a label or probability.
Clef models the decision problem as multi-label classification or conditional probability prediction rather than autoregressive text generation, an approach similar to the discriminative models of the BERT era.
Discriminative models learn which class a given input-output belongs to or its probability, with typical representatives including BERT and its fine-tuned variants.
Discriminative models generally have smaller parameter scales, faster inference speeds, and lower VRAM usage, making them suitable for deployment in resource-constrained local environments.
生成式模型建模联合概率 P(x,y) 或数据分布 P(x)，而判别式模型直接建模条件概率 P(y|x)；生成模型在小样本时更具优势，判别模型在大数据量时通常表现更好
生成模型在小样本时更具优势，判别模型在大数据量时通常表现更好
生成式模型建模联合概率 P(x,y) 或数据分布 P(x)，判别式模型直接建模条件概率 P(y|x)，该区分最初由 Andrew Ng 和 Michael Jordan 在 2001 年的论文中系统比较
All Facts (7)
Discriminative models generally have smaller parameter scales, faster inference speeds, and lower VRAM usage, making them suitable for deployment in resource-constrained local environments.
65%UnverifiedDiscriminative models (such as face detection and closed-eye classifiers) are only responsible for judging "whether the eyes are closed," outputting a label or probability.
50%UnverifiedClef models the decision problem as multi-label classification or conditional probability prediction rather than autoregressive text generation, an approach similar to the discriminative models of the BERT era.
50%UnverifiedDiscriminative models learn which class a given input-output belongs to or its probability, with typical representatives including BERT and its fine-tuned variants.
50%Unverified生成式模型建模联合概率 P(x,y) 或数据分布 P(x)，而判别式模型直接建模条件概率 P(y|x)；生成模型在小样本时更具优势，判别模型在大数据量时通常表现更好
50%Unverified生成式模型建模联合概率 P(x,y) 或数据分布 P(x)，判别式模型直接建模条件概率 P(y|x)，该区分最初由 Andrew Ng 和 Michael Jordan 在 2001 年的论文中系统比较
50%Unverified生成模型在小样本时更具优势，判别模型在大数据量时通常表现更好
50%