Unverified50% confidencePredictionExact time
The work gives information-gain-based regret bounds for kernel CPT bandits for a broad class of kernel functions, and the results match the rates of the corresponding risk-neutral kernel methods up to logarithmic factors.
1
Sources
50%
Confidence
Long-term
Relevance
11/5/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedA single unsafe execution rate metric masks capability differences among models, so evidence acquisition behavior and post-recovery decision quality need to be reported as well.66% similarUnverifiedInformation gain γ_T is defined as the reduction in posterior entropy about the unknown function before and after an observation; in kernel methods it is directly related to the effective dimension of the kernel matrix and determines the size of the exploration term in the regret bound of optimistic algorithms.66% similarUnverifiedBayesian methods can use historical failure rates as a prior and combine them with new observations to update the posterior distribution, conservatively estimating true risk when samples are scarce66% similarUnverifiedCross-validation by two independently implemented Lean kernels is an additional defensive measure; if both accept the same proof, the probability of a common error is extremely low.65% similarUnverifiedSupervised fine-tuning (SFT) tends to cause more severe catastrophic forgetting than reinforcement fine-tuning (RFT).65% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1138279API
curl https://kongchang.com/api/v1/knowledge/claims/1138279MCP
get_claim(id=1138279)