Unverified50% confidenceFactExact time
With 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.
1
Sources
50%
Confidence
Long-term
Relevance
10/13/2026
First Seen
Sources
Related Entities
Related Claims
Unverified域随机化的代价是早期训练信号更加嘈杂、收敛更慢，但换来策略在整个参数空间内均匀分布的泛化能力78% similarVerifiedOn convergent tasks, there is a clear performance saturation phenomenon after models reach a certain scale, where continued increases in parameter count or training data yield rapidly diminishing marginal contributions to accuracy.77% similarUnverifiedWhen batch size increases, it is generally necessary to increase the learning rate accordingly following the linear scaling rule.76% similarVerifiedGradient accumulation accumulates gradients over multiple small batches before performing a single parameter update, with an effect equivalent to training with a large batch.75% similarUnverifiedExperiments show that repeating training on the same data for more than 4 epochs yields diminishing performance gains and can sometimes result in slight performance degradation.74% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1015932API
curl https://kongchang.com/api/v1/knowledge/claims/1015932MCP
get_claim(id=1015932)