Unverified50% confidenceFactExact time
序列到序列（Seq2Seq）架构由Sutskever等人于2014年提出，已成为神经机器翻译的基础范式
1
Sources
50%
Confidence
Long-term
Relevance
8/21/2026
First Seen
Sources
Related Claims
Unverified贝叶斯定理在朴素贝叶斯分类器、贝叶斯神经网络、变分自编码器（VAE）和扩散模型中都有直接应用56% similarUnverified对抗样本概念由Szegedy等人于2013年在论文《Intriguing properties of neural networks》中正式提出53% similarUnverified稀疏自编码器(SAE)将模型激活分解为人类可解释的特征方向，Anthropic的Claude系列模型可解释性研究大量使用了这一技术53% similarVerified推测解码（Speculative Decoding）是目前最主流的大模型推理加速路径之一，使用小型草稿模型快速生成候选token再由主模型并行验证52% similarUnverifiedNexent unifies the Function Calling interface specifications of different large models through an adaptation layer, supporting seamless switching of the same set of tool definitions across different model backends.52% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/784040API
curl https://kongchang.com/api/v1/knowledge/claims/784040MCP
get_claim(id=784040)