Unverified50% confidenceFactExact time
The traditional deep learning paradigm divides the model into two stages: a training stage for learning parameters, and an inference stage where parameters are frozen and outputs are produced directly.
1
Sources
50%
Confidence
Long-term
Relevance
9/11/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedA frozen model refers to a model whose weight parameters all remain unchanged during the inference phase, with no fine-tuning, adapter training, or online learning performed.72% similarUnverifiedIn the field of machine learning, model generalization capability is divided into three gradients: few-shot learning, one-shot learning, and zero-shot learning.72% similarUnverified深度思考模式下，模型会先在内部生成中间推理步骤，再基于这些步骤得出最终结论，在处理多步逻辑推导任务时表现显著优于直接生成71% similarUnverified少样本学习通过在Prompt中提供2-5个示例帮助模型快速理解输出格式和风格要求71% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/907372API
curl https://kongchang.com/api/v1/knowledge/claims/907372MCP
get_claim(id=907372)