Error Accumulation
A phenomenon in a one-way pipeline where small deviations from an upstream stage are treated as established facts by downstream stages and processed onward, ultimately being amplified into significant geometric and layout errors.
Core Facts
Timeline (last 90 days)
Error accumulation in multi-turn editing is harder to correct than in text generation, because images are continuous high-dimensional signals and lack the discrete-token "resampling" mechanism that interrupts error propagation.
Strategies for addressing error accumulation include Monte Carlo tree search-style lookahead planning, hierarchical planning, and confidence-based adaptive execution.
When editing the same image multiple times in succession, the minor deviations introduced by each generation are inherited and amplified by the next edit, leading to error accumulation.
All Facts (4)
In long-chain tasks, if each step has an error rate of approximately 5%, the overall success rate may drop to below 36% after 20 steps.
70%UnverifiedError accumulation in multi-turn editing is harder to correct than in text generation, because images are continuous high-dimensional signals and lack the discrete-token "resampling" mechanism that interrupts error propagation.
50%UnverifiedStrategies for addressing error accumulation include Monte Carlo tree search-style lookahead planning, hierarchical planning, and confidence-based adaptive execution.
50%UnverifiedWhen editing the same image multiple times in succession, the minor deviations introduced by each generation are inherited and amplified by the next edit, leading to error accumulation.
50%