Unverified50% confidenceFactExact time
When batch_size scales linearly with the number of environments, the variance of the gradient estimate decreases at a rate of about 1/batch_size.
1
Sources
50%
Confidence
Long-term
Relevance
10/13/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedWith a fixed learning_rate, when batch_size scales linearly with the number of environments and num_epoch remains unchanged, the total displacement pushing the policy toward a new distribution in a single iteration may increase, manifesting as violent oscillation or even collapse of rewards in the early stages of training.64% similarUnverifiedThe harmonic mean is sensitive to weak links; as long as performance in any one dimension is poor, the overall score will be noticeably pulled down.61% similarUnverifiedIn batch-1 decoding scenarios, because arithmetic intensity is extremely low, the performance bottleneck must fall on the memory bandwidth side, and the theoretical token ceiling can be calculated from measured bandwidth multiplied by weight bytes.60% similarUnverifiedIn this error bound, the exponent governing the convergence rate is determined by the intrinsic dimension of the data (the Wasserstein-p dimension), rather than the ambient dimension.60% similarUnverifiedMini-batch gradient descent (Mini-batch GD) is a compromise between batch gradient descent and stochastic gradient descent, and is also the most commonly used gradient descent method in industry.59% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1015931API
curl https://kongchang.com/api/v1/knowledge/claims/1015931MCP
get_claim(id=1015931)