Deep learning has driven many recent advances in process analytics, especially for predictive and prescriptive monitoring. However, standard objectives like cross-entropy optimize local next-step likelihoods and only implicitly capture control-flow structure. This results in models achieving high token-level accuracy while permitting imprecise global behavior.
We introduce DIFF-ERO, a conformance-aware loss function for deep learning models on process data. DIFF-ERO is a differentiable formulation of entropy-based stochastic conformance that incorporates control-flow information during training. Our approach constructs batch-level stochastic transition matrices with soft edge memberships, allowing structural precision and recall signals to directly inform backpropagation. The loss is model-agnostic and can be applied whenever the final representation parametrizes stochastic transitions.
We instantiate DIFF-ERO in transformer encoder-decoder pipelines for next-activity prediction and use it jointly with cross-entropy to analyze its theoretical components with respect to convergence. Across benchmarks comparing other loss functions and targets, DIFF-ERO shows improved predictive performance where structure matters most while maintaining parity elsewhere. At the same time, the learned stochastic automaton converges towards the structural ground truth, indicating that the network internalizes process model structure.
Blogger's Review: The introduction of the DIFF-ERO loss function provides enhanced structural consistency for deep learning models in process mining, particularly in scenarios where precise control flows are crucial. Its model-agnostic nature also lays the groundwork for broader applications, making it promising for future explorations in various domains.