Research map / Time series / Causal and generative time series
Learning Temporal Point Processes for Efficient Retrieval of Continuous Time Event Sequences
Vinayak Gupta, Srikanta Bedathur, Abir De
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use query-conditioned cross-attention instead of self-attention | NEUROSEQRET · Audio, Celebrity, Electricity, Health, Sports | MAP and NDCG@10: MAP: 55.8$\pm$1.8 -> 56.2$\pm$2.1; 64.4$\pm$1.9 -> 65.1$\pm$1.9; 30.7$\pm$0.7 -> 32.4$\pm$0.8; 35.9$\pm$0.9 -> 37.4$\pm$ | Improved |
| Apply NEUROSEQRET rather than the strongest baseline | NEUROSEQRET · All five datasets | MAP and NDCG@10: Table 1: CrossAttn-NEUROSEQRET scores exceed the strongest baseline in each dataset and metric. | Improved |
| Add Fisher-kernel similarity to the model-independent score (ablation) | CrossAttn-NEUROSEQRET · Audio, Celebrity, Health | MAP (%): 36.1$\pm$0.0 -> 56.2$\pm$2.1; 43.7$\pm$0.0 -> 65.1$\pm$1.9; 18.9$\pm$0.0 -> 37.4$\pm$0.9 | Improved |
| Add model-independent time and mark similarities to Fisher-kernel scoring (ablation) | CrossAttn-NEUROSEQRET · Audio, Celebrity, Health | MAP (%): 53.9$\pm$1.9 -> 56.2$\pm$2.1; 62.5$\pm$1.3 -> 65.1$\pm$1.9; 33.6$\pm$0.7 -> 37.4$\pm$0.9 | Improved |
| Add model-independent mark similarity (ablation) | CrossAttn-NEUROSEQRET · Audio, Celebrity, Health | MAP (%): 54.6$\pm$1.9 -> 56.2$\pm$2.1; 63.1$\pm$1.4 -> 65.1$\pm$1.9; 33.7$\pm$0.7 -> 37.4$\pm$0.9 | Improved |
| Add model-independent time similarity (ablation) | CrossAttn-NEUROSEQRET · Audio, Celebrity, Health | MAP (%): 55.7$\pm$2.0 -> 56.2$\pm$2.1; 63.7$\pm$1.8 -> 65.1$\pm$1.9; 35.9$\pm$0.8 -> 37.4$\pm$0.9 | Improved |
| Add query unwarping (ablation) | CrossAttn-NEUROSEQRET · Audio, Celebrity, Health | MAP (%): 55.2$\pm$2.2 -> 56.2$\pm$2.1; 62.9$\pm$2.0 -> 65.1$\pm$1.9; 34.3$\pm$0.9 -> 37.4$\pm$0.9 | Improved |
| Use supervised ranking loss instead of vanilla MTPP maximum-likelihood retrieval | RMTPP, SAHP, and THP · All five datasets | MAP and NDCG@10: Table 1 reports higher Rank-RMTPP, Rank-SAHP, and Rank-THP values than their vanilla counterparts across the datasets. | Improved |
| Remove even distribution of positive and negative hash bits (ablation) | NEUROSEQRET trainable hashing · Retrieval efficiency-accuracy tradeoff | NDCG@10 vs. reduction factor | Hurt |
| Remove the hash-bit binarization term (ablation) | NEUROSEQRET trainable hashing · Retrieval efficiency-accuracy tradeoff | NDCG@10 vs. reduction factor | Hurt |
| Remove the hash-bit redundancy penalty (ablation) | NEUROSEQRET trainable hashing · Retrieval efficiency-accuracy tradeoff | NDCG@10 vs. reduction factor | Hurt |
| Learn hash codes from trained embeddings rather than using random hyperplanes | NEUROSEQRET retrieval hashing · Retrieval efficiency-accuracy tradeoff | NDCG@10 vs. reduction factor | Improved |
Closest papers by meaning
- Predictive Querying for Autoregressive Neural Sequence Models (NeurIPS 2022 · similarity 0.87)
- User-Dependent Neural Sequence Models for Continuous-Time Event Data (NeurIPS 2020 · similarity 0.86)
- EasyTPP: Towards Open Benchmarking Temporal Point Processes (ICLR 2024 · similarity 0.84)
- Neuro-Symbolic Temporal Point Processes (ICML 2024 · similarity 0.84)
- Deep Continuous-Time State-Space Models for Marked Event Sequences (NeurIPS 2025 · similarity 0.83)
- Interacting Diffusion Processes for Event Sequence Forecasting (ICML 2024 · 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.
