Unverified50% confidenceBenchmarkExact time
For modern deep models with d=10000, √log(10000)≈3, meaning the path overhead of NGD can reach on the order of 3 times the optimum.
1
Sources
50%
Confidence
Long-term
Relevance
10/7/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedDepth Anything、MiDaS、DPT等基础模型大幅提升了相对深度质量，但输出仍为无尺度的相对值，需要额外锚定手段还原绝对度量61% similarUnverifiedThe compound scaling mathematical formulation of EfficientNet is depth d=α^φ, width w=β^φ, and resolution r=γ^φ, where α, β, and γ are determined by grid search on the baseline model B0.58% similarUnverifiedOllama-hosted DeepSeek models adopt a zero data retention (ZDR) policy.58% similarUnverifiedDeepSeek V3 and R1, leveraging the MoE architecture, compress inference costs to a fraction of those of traditional dense models while maintaining top-tier performance.57% similarUnverified单目深度估计的代表性模型包括Intel的MiDaS、Meta的DPT以及Depth Anything系列57% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/983534API
curl https://kongchang.com/api/v1/knowledge/claims/983534MCP
get_claim(id=983534)