Research map / Label-efficient and robust learning / Recognition and long-tail
What Comes Next and Why: Interpretable Next-Event Prediction with Neuro-Symbolic Rules
Tim Nico Bauerschmidt, Isabel Valera, Jilles Vreeken
Read the paperCheck your idea against 137,217 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes an interpretable next-event prediction approach that learns neuro-symbolic rules to explain why a predicted event occurs, rather than prioritizing accuracy alone.
Problem
Sequential next-event prediction methods prioritize accuracy at the expense of interpretability, and existing neuro-symbolic rule-learning approaches are not readily extended to next-event prediction.
Approach
Extends neuro-symbolic symbolic rule learning to next-event prediction so that predictions come with explicit, human-readable rules; the abstract does not specify details beyond this.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add residual network to Rules (Runes) (ablation) | Rules · Rolling Mill | accuracy (%): 34.13 -> 70.44 | Improved |
| add residual network to Rules (Runes) (ablation) | Rules · Amazon | accuracy (%): 6.53 -> 17.55 | Improved |
| add residual network to Rules (Runes) (ablation) | Rules · Chorales | accuracy (%): 21.00 -> 37.00 | Improved |
| add residual network to Rules (Runes) (ablation) | Rules · Holidays | accuracy (%): 97.57 -> 100.00 | Improved |
| add residual network to Rules (Runes) (ablation) | Rules · Commits | accuracy (%): 19.60 -> 18.31 | Hurt |
| use Rules instead of tree ensembles or PPM for compactness | Rules · real-world datasets | number of rules: Rolling Mill: 77 vs 640723 (XGBoost) | Improved |
| increase the number of training samples | Rules vs other models · synthetic data | accuracy | Mixed |
| lower reliability of consequents following antecedents | Rules · synthetic ground-truth recovery | rule recovery | Hurt |
| increase random noise events in the objective | Rules · synthetic ground-truth recovery | rule recovery | No clear effect |
| use PPM (time-agnostic Markov) on rule-generated data | PPM · synthetic data | accuracy | Hurt |
| anneal predicate-layer temperature first, then antecedent-layer temperature (ablation) | Rules training · training schedule | optimization stability | Improved |
Closest papers by meaning
- Neuro-Symbolic Temporal Point Processes (ICML 2024 · similarity 0.88)
- Explainable Neural Networks with Guarantee: A Sparse Estimation Approach (AAAI 2025 · similarity 0.84)
- Towards Understanding the Universality of Transformers for Next-Token Prediction (ICLR 2025 · similarity 0.83)
- A Neuro-Symbolic Benchmark Suite for Concept Quality and Reasoning Shortcuts (NeurIPS 2024 · similarity 0.83)
- Deciphering Raw Data in Neuro-Symbolic Learning with Provable Guarantees (AAAI 2024 · similarity 0.83)
- Large Language Models are Interpretable Learners (ICLR 2025 · similarity 0.83)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
