Reproduce “TPP-LLM: Modeling Temporal Point Processes by Efficiently Fine-Tuning Large Language Models” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Zefang Liu, Yinzhu Quan
The TPP-LLM framework efficiently models temporal point processes by integrating large language models (LLMs) with dedicated temporal mechanisms, leveraging textual event descriptions and parameter-efficient fine-tuning (PEFT). This approach leads to superior or competitive performance in both next event type prediction accuracy and next event time RMSE across diverse datasets, showcasing the benefit of combining semantic understanding with temporal dynamics.
Key insights
- Leveraging LLMs to process textual event descriptions substantially enhances event type prediction accuracy, particularly in datasets with semantically rich event types.
- Temporal embeddings combined with LLMs effectively capture continuous-time dynamics, leading to accurate next event time predictions.
- Parameter-efficient fine-tuning (LoRA) is crucial for practical adaptation of LLMs to TPP tasks, offering a balance between computational cost and performance retention.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.