A control-flow-aware trace segmentation algorithm enables segment-level SHAP explanations for deep-learning outcome prediction in predictive process monitoring.
Operator thermalization vs eigenstate thermalization
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
abstract
We study the characteristics of thermalizing and non-thermalizing operators in integrable theories as we turn on a non-integrable deformation. Specifically, we show that $\sigma^z$, an operator that thermalizes in the integrable transverse field Ising model, has mean matrix elements that resemble ETH, but with fluctuations around the mean that are sharply suppressed. This suppression rapidly dwindles as the Ising model becomes non-integrable by the turning on of a longitudinal field. We also construct a non-thermalizing operator in the integrable regime, which slowly approaches the ETH form as the theory becomes non-integrable. At intermediate values of the non-integrable deformation, one distinguishes a perturbatively long relaxation time for this operator.
fields
cs.LG 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Feature Attribution-Based Explainability Analysis of Deep Learning Models in Predictive Process Monitoring
A control-flow-aware trace segmentation algorithm enables segment-level SHAP explanations for deep-learning outcome prediction in predictive process monitoring.