Unverified50% confidenceFactExact time
GraftOOD adopts a shared velocity-regression objective function, combining two types of supervision signals: group-relative reward correction and sequence-level example supervision.
1
Sources
50%
Confidence
Long-term
Relevance
10/12/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedRecursive reward modeling builds progressively finer reward signals layer by layer, allowing safety evaluation to be upgraded in sync with model capabilities.72% similarUnverified'Whether the extraction is faithful to the source text' and 'whether the signal has predictive power' should be separated into two independent evaluation dimensions: the former belongs to a short feedback loop, while the latter requires long-cycle attribution using delayed labels.70% similarUnverifiedVariance-based contrastive optimization identifies problems in planning and tool invocation by contrasting samples with different performance levels.70% similarUnverifiedIn k-fold cross-validation, each fold produces a set of performance metric values, providing paired observation samples for the Wilcoxon signed-rank test.69% similarUnverifiedDuring model selection, it is necessary to confirm that the similarity objective function used during model training is consistent with the metric used during retrieval; otherwise, systematic bias will occur.68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1001539API
curl https://kongchang.com/api/v1/knowledge/claims/1001539MCP
get_claim(id=1001539)