Unverified50% confidenceFactExact time
Under trajectory-level sparse rewards, the TD residual propagates backward along the entire trajectory, causing the credit of early actions to be diluted or even distorted by later outcomes.
1
Sources
50%
Confidence
Long-term
Relevance
10/22/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedSparse rewards combined with long time spans cause gradient-based optimization signals to decay significantly during backpropagation.75% similarVerifiedError cascading refers to small judgment deviations in early steps being progressively amplified as the reasoning chain extends, ultimately producing results far from what was intended.72% similarUnverifiedThe signal of goal drift is usually a slow degradation of some downstream metric, which can only be found by looking at the trend line rather than a single action.68% similarUnverified理论分析表明，在时间步长为T的轨迹上，行为克隆的累积误差以O(T²)的速度增长68% similarUnverifiedTemporal-difference-based methods struggle to attribute sparse final rewards to early actions, resulting in extremely weak learning signals.68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1083960API
curl https://kongchang.com/api/v1/knowledge/claims/1083960MCP
get_claim(id=1083960)