Unverified50% confidenceFactExact time
Experiments show that the geometric structure of ray coverage can explain a large portion of correctly classified test inputs, as well as the prediction behavior on adversarial examples, mislabeled samples, and even out-of-distribution (OOD) samples.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/19/2026
First Seen
Valid until: 1/17/2027
Sources
Related Entities
Related Claims
UnverifiedData leakage testing should adopt an adversarial testing mindset, actively constructing various edge invocation scenarios to verify whether sensitive data is inadvertently exposed to untrusted downstream endpoints.74% similarUnverifiedDeletion testing evaluates explanation faithfulness by masking the regions highlighted by the explanation and observing the degree of drop in model confidence.73% similarUnverifiedThe study proposes a claim-level analysis framework that forms four quadrants by crossing correct/incorrect answers with complete/absent evidence, which can reveal changes masked by aggregate metrics.71% similarUnverifiedDifferences in test granularity, validation methods, and task sources across benchmarks are significant, and one must be especially cautious when comparing numbers across benchmarks.71% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1065868API
curl https://kongchang.com/api/v1/knowledge/claims/1065868MCP
get_claim(id=1065868)