Unverified50% confidenceBenchmarkExact time
On ImageNet-512, SpecSlot improves FID by 2.84 relative to the control group while reducing FLOPs by 3.9%.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/30/2026
First Seen
Valid until: 1/28/2027
Sources
Related Entities
Related Claims
Unverified探索性能上界实验发现现有验证粒度并非最优，理想情况下计算量（FLOPs）可减少75倍，同时准确率提升1.1%62% similarUnverifiedGLM-5.2 采用 IndexShare 技术，每四个稀疏注意力层复用同一个索引器，在 1M 上下文长度下将每 token 的 FLOPs 降低了 2.9 倍60% similarUnverifiedDuring training, the held-out FID dropped from 33.7 to 27.0, FD-DINOv2 dropped from 570 to 218, and detector-based object accuracy improved from 65% to 90%.59% similarUnverifiedBlurPool can improve ResNet-50's shift consistency from approximately 89% to approximately 95%, while also yielding a small improvement in classification accuracy of about 0.5%.58% similarUnverifiedOn three multi-hop question answering benchmark datasets, R²Adapter can reduce graph RAG usage by up to 59% while maintaining answer accuracy comparable to full-pipeline graph RAG.58% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1115721API
curl https://kongchang.com/api/v1/knowledge/claims/1115721MCP
get_claim(id=1115721)