Unverified50% confidenceFactExact time
FSPO directly operates in function space on the policy's output distribution (the next-token probability distribution given an input), avoiding parameter symmetry and optimization path complexity, and is more direct and stable than alignment in parameter space.
1
Sources
50%
Confidence
Long-term
Relevance
10/15/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedFunctional Bayesian Optimization (FBO) deals with variables that exist in an infinite-dimensional function space rather than a finite-dimensional vector space.73% similarUnverifiedFSPO (Function-Space Policy Optimization) proposes sparsely storing the next-token distributions of historical policies on specific inputs during training as fixed temporal anchor targets, rather than maintaining a continuously updated teacher model.73% similarUnverifiedFor continuous action spaces, PPO is typically paired with a Gaussian policy output, where the network simultaneously predicts the mean and standard deviation of actions.62% similarUnverifiedFyxer uses fine-tuning to make the model better suited for vertical tasks such as email classification, priority judgment, and drafting replies.61% similarUnverifiedActivated parameters directly determine the computation (FLOPs) and theoretical latency at inference time, while the total parameter count determines GPU memory/RAM usage.61% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1037327API
curl https://kongchang.com/api/v1/knowledge/claims/1037327MCP
get_claim(id=1037327)