Unverified50% confidenceOpinionExact time
The 70.80% for ViT on CIFAR-100 and the results on ImageNet come from different settings, so cross-task comparisons should be made with caution.
1
Sources
50%
Confidence
Long-term
Relevance
11/9/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedFixMatch 在 CIFAR-10 仅使用 40 个标注样本(每类 4 个)的场景下达到 94.93% 的准确率68% similarUnverifiedSALU achieves 70.80% Top-1 on CIFAR-100 (ViT).65% similarUnverifiedThe measurement of the precise figure of 90% is subjective and depends on how one defines that a benchmark 'covers' a certain type of task; if the standard is relaxed to count 'AI can assist' as coverage, the proportion will rise.63% similarUnverifiedA paper from Elloe AI Lab points out that nearly 90% of the tested Agents exhibited measurable goal drift after about 30 steps.62% similarUnverifiedOn CIFAR-100 DINOv2 feature flows, guided time sampling reduced the uniformly evaluated velocity loss by 6.78%.62% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1151079API
curl https://kongchang.com/api/v1/knowledge/claims/1151079MCP
get_claim(id=1151079)