A finite-k decomposition reveals a bias toward over-prediction in failure rate extrapolation from evaluation data, addressed by a new forecastability loss that improves held-out forecast accuracy in language-model and RL experiments.
Catastrophic-risk-aware reinforcement learning with extreme-value-theory-based policy gradients.The Journal of Finance and Data Science, 11:100165
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Training ML Models with Predictable Failures
A finite-k decomposition reveals a bias toward over-prediction in failure rate extrapolation from evaluation data, addressed by a new forecastability loss that improves held-out forecast accuracy in language-model and RL experiments.