Unverified50% confidenceFactExact time
Mutual information measures the degree of dependence between the prediction and the perturbed classifiers; if the prediction changes sharply under perturbation, the MI is small.
1
Sources
50%
Confidence
Long-term
Relevance
11/3/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedMIRA's perturbation scale selection depends on mutual information estimation and may face estimation variance issues when batches are small or the number of classes is very large.80% similarUnverifiedThe value of context-based prediction depends on its hit rate; inaccurate recommendations will increase friction.73% similarUnverifiedCount-based prediction can partially identify which inputs are most sensitive to corpus changes, but substantially underestimates the rate at which divergence decreases in neural models.73% similarUnverifiedMIRA uses an entropy loss weighted by mutual-information confidence, reducing the weight of samples that are perturbation-sensitive or low-confidence.70% similarUnverifiedQuantization errors are usually distributed uniformly across all parameters, whereas the destructiveness of pruning is more concentrated on the removed connections, so the two have different patterns of impact on knowledge distribution.70% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1129524API
curl https://kongchang.com/api/v1/knowledge/claims/1129524MCP
get_claim(id=1129524)