Unverified50% confidenceFactExact time
Controlled order-statistics baselines (covering bounded, light-tailed, and heavy-tailed distributions) can largely reproduce the observed log-linear geometry, but cannot explain the significant shift in the slope α after training.
1
Sources
50%
Confidence
Long-term
Relevance
10/14/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedStandard backpropagation is essentially a discrete iteration of gradient descent, and when the learning rate is regarded as an infinitesimally small time step, this discrete process converges to the gradient flow ordinary differential equation dθ/dt = −∇L(θ).72% similarUnverifiedThis approximation error is independent of model capacity: even with a perfect marginal distribution predictor trained on infinite data, the product approximation will inevitably deviate from the joint distribution when writing to multiple positions with dependencies in parallel.70% similarUnverifieddegree参数只对poly核有效，gamma对linear核是无效参数，对这些组合遍历会产生无意义的重复训练69% similarUnverifiedThe Wasserstein distance can provide meaningful gradient signals even when two distributions have no overlap at all, alleviating the mode collapse and vanishing gradient problems in original GAN training.69% similarUnverified低精度引入的数值扰动会在RL的多轮迭代中通过策略梯度的乘积形式被放大，可能导致训练崩溃或模型性能退化68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1023595API
curl https://kongchang.com/api/v1/knowledge/claims/1023595MCP
get_claim(id=1023595)