Unverified50% confidenceSolutionExact time
Using pseudo-labels as reward references for GRPO rather than as SFT training targets can significantly improve ASR models' robustness to label noise.
1
Sources
50%
Confidence
Long-term
Relevance
10/26/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedSupervised fine-tuning (SFT) trains a model using 'prompt–response' example pairs, transforming it from outputting gibberish to being capable of meaningful question answering.63% similarUnverifiedExisting graph-level OOD generalization work usually assumes that class labels are available during training, using supervised signals to determine which substructures are predictive, thereby suppressing spurious correlations in representation learning.61% similarUnverifiedIf verdict accuracy is the priority, label-only training has the advantage; if controllability of the generation process and source support are more important, rejection sampling provides a score-based selection mechanism.60% similarUnverified开源ASR模型识别准确率需通过实际数据测试验证，不应仅凭模型参数量判断，其在特定领域术语、嘈杂环境或非主流语言上未必达到成熟商用API水准60% similarUnverifiedIn the experiment, Serena mainly played an auxiliary role, used for cross-checking ast-grep and supplementing result verification, and encountered problems such as character encoding conversion and missing type information in external libraries.60% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1098809API
curl https://kongchang.com/api/v1/knowledge/claims/1098809MCP
get_claim(id=1098809)