Unverified50% confidenceFactExact time
掩码自编码器（MAE）由 He et al.（2021）提出，其核心创新在于采用 75% 极高遮盖率随机遮盖图像块后让模型重建
1
Sources
50%
Confidence
Long-term
Relevance
7/16/2026
First Seen
Sources
Related Claims
Unverified去噪塔借鉴图像扩散DiT架构的adaLN时间步机制，用于编码掩码比例，仅增加约150万参数59% similarUnverified通过提供更多具体案例（few-shot prompting）约束模型输出空间，是纠正AI过度设计的有效手段56% similarUnverifiedThe design goal of Claude 4.5 Haiku is to compress inference latency to sub-second levels while reducing the token cost per call to a fraction of that of large models.56% similarUnverifiedModern quantization schemes such as GPTQ, AWQ, and the grouped quantization of the GGUF format can reduce model size by more than 75% while retaining the vast majority of reasoning capability through layer-by-layer calibration and outlier protection.56% similarUnverified2023-2024年间多项研究表明简单的改写攻击可将水印检测率从接近100%降至接近随机猜测水平55% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/531563API
curl https://kongchang.com/api/v1/knowledge/claims/531563MCP
get_claim(id=531563)