Researchers have developed the Foundation Event Predictor (FEP), a novel in-context architecture designed for next-event prediction across diverse event logs. FEP utilizes causal self-attention and cross-attention, efficiently parallelized with FlashAttention, to aggregate next-event outcomes. To enhance generalization across different systems, the model incorporates permutation-robust embedding learning, random fixed embeddings, empirical quantile transforms, and robust normalization techniques. Pretrained on 696 real-world datasets and evaluated on 50 unseen datasets, FEP demonstrates effective generalization without parameter adaptation, often matching or exceeding the performance of models trained specifically for each target dataset. AI
IMPACT This research could improve the ability of AI systems to understand and predict events in complex, real-world operational systems.
RANK_REASON The cluster describes a research paper detailing a new model architecture for event prediction.
Read on Hugging Face Daily Papers →
- Causal self-attention
- empirical quantile transforms
- Event prediction
- Flashattention
- Foundation Event Predictor
- foundation model
- Heterogeneous Event Logs
- permutation-robust embedding learning
- random fixed embeddings
- Robust normalization protocols for multiplexed fluorescence bioimage analysis
- School of Economics and Management of the University of Porto
- arXiv
- Hugging Face
AI-generated summary · Google Gemini · from 2 sources. How we write summaries →
