The field of predictive modeling is moving towards integrating symbolic and neural approaches to improve accuracy and logical consistency. Recent developments focus on incorporating domain knowledge and temporal logic into predictive models, enabling more informed decision-making. Conformal prediction is also being explored as a means to provide uncertainty quantification and statistical guarantees in various applications, including time-series forecasting and function-valued outputs. Notable papers in this area include:

• A Neuro-Symbolic Predictive Process Monitoring approach that integrates Linear Temporal Logic into autoregressive sequence predictors, improving suffix prediction accuracy and compliance with temporal constraints.
• A Conformal Predictive Monitoring method for multi-modal scenarios, leveraging deep generative models to approximate system dynamics and provide mode-specific prediction intervals.
• A Conformal Prediction for Time-series with Change points algorithm, addressing the challenge of handling time series data with sudden shifts in the underlying process.
• A Split Conformal Prediction method in the function space with neural operators, extending conformal prediction to function-valued outputs and providing finite-sample coverage guarantees.