Unverified85% confidenceFactExact time
反向KL散度D(q||p)倾向于让q集中在p的单个峰值上（mode-seeking），常用于生成模型训练
1
Sources
85%
Confidence
Long-term
Relevance
8/2/2026
First Seen
Sources
From Entropy to Cross-Entropy: An Information-Theoretic Derivation of the Logistic Regression Loss Function
Related Entities
Related Claims
Unverified前向KL散度D(p||q)倾向于让q覆盖p的所有可能性（mode-covering），是变分推断和监督学习中最常用的形式86% similarUnverifiedQuantization-Aware Training (QAT) lets the model retain a full-precision copy of weights, practicing rounding operations in each forward pass to learn weight distributions under extreme compression.70% similarUnverifiedDQN的理论基础是Q-learning，一种无模型强化学习算法，核心是学习Q函数Q(s,a)表示状态s下执行动作a的期望累积折扣奖励68% similarUnverifiedQ-learning、DQN 等算法本质上都是在近似求解贝尔曼最优方程67% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/679898API
curl https://kongchang.com/api/v1/knowledge/claims/679898MCP
get_claim(id=679898)