Unverified50% confidenceFactExact time
PRISMA introduces an instance-dependent difficulty measure ρ* (taking values between 1 and d), which characterizes the intrinsic difficulty of designing a shared sampling distribution; a smaller ρ* indicates stronger commonality among policies and high efficiency of shared sampling.
1
Sources
50%
Confidence
Long-term
Relevance
10/16/2026
First Seen
Sources
Related Entities
Related Claims
Unverified块划分粒度是多目标优化问题：块越小近似精度越高但元数据开销增大且GPU并行效率降低72% similarUnverifiedPRISMA uses an adaptive doubling procedure to adjust the regularization strength λ online without knowing ρ*, avoiding the requirement of knowing the instance difficulty in advance and incurring only logarithmic additional overhead.71% similarUnverified对齐问题涉及目标规范、内部对齐和分布偏移等多个技术层面67% similarUnverifiedEPD disaggregation introduces new architectural complexity, requiring evaluation of intermediate embedding vector transmission overhead, the ratio of instance counts at each stage, and scheduling strategies under load imbalance—representing a trade-off between performance and complexity.66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1042070API
curl https://kongchang.com/api/v1/knowledge/claims/1042070MCP
get_claim(id=1042070)