Publications

@article{Bai2026BobFlow, title = {Demonstration of BobFlow: Human-Agent Collaboration Using Dataflows on Data Science Tasks}, author = {Jiadong Bai and Yicong Huang and Chen Li}, year = {2026}, doi = {10.14778/3827998.3828105}, journal = {Proceedings of the VLDB Endowment}, volume = {19}, number = {12}, pages = {4718--4721} }

@inproceedings{Asad2026Texera, title = {Click, Share, Learn: Teaching Data Science Using Apache Texera}, author = {Sarah Asad and Xiaozhen Liu and Kun Woo Park and Xinyuan Lin and Jiadong Bai and Shengquan Ni and Yicong Huang and Chen Li}, year = {2026}, doi = {10.48786/EDBT.2026.61}, booktitle = {Proceedings of the 29th International Conference on Extending Database Technology}, pages = {708--711} }

PRISM uses multi-versioned memory to avoid staleness while enabling parallel training of temporal graph neural networks.

@article{Islam2026PRISM, title = {PRISM: A Training System to Unlock the Potential of Temporal Graph Learning Through Staleness Avoidance}, author = {Md Ashraful Islam and Hojae Son and Suhaas Kiran Doddagaddavalli Gangadharaiah and Marco Serafini}, year = {2026}, doi = {10.14778/3819518.3819544}, journal = {Proceedings of the VLDB Endowment}, volume = {19}, number = {9}, pages = {2196--2209} }

Kalypso makes LLM serving aware of semantic query plans, reusing KV-cache state across operators through memory-aware scheduling.

@article{Son2026Kalypso, title = {Kalypso: Relational LLM Serving}, author = {Hojae Son and Md Ashraful Islam and Huy Gia Cao and Hui Guan and Marco Serafini}, year = {2026}, doi = {10.48550/arXiv.2607.23815}, journal = {arXiv preprint arXiv:2607.23815} }

Metamodels for discrete-event simulations approximate the behavior of simulation models without running expensive simulations. This work extends modular dynamic Bayesian network metamodels to non-Markovian queues by approximating non-exponential distributions using phase-type distributions. Experiments on a G/M/1 queue demonstrate accurate answers to probabilistic and causal queries with orders-of-magnitude faster inference than direct simulation.

@inproceedings{Amaranath2026, title = {Extending Causal Metamodeling to a non-Markovian Queue}, author = {Pracheta Amaranath and Anant Bhide and David D. Jensen and Peter J. Haas}, url = {https://arxiv.org/abs/2606.00795}, doi = {10.48550/arXiv.2606.00795}, year = {2026}, booktitle = {Proceedings of the 2026 Winter Simulation Conference}, note = {To appear} }

@article{Vittis2026, title = {Dominance-Based Data Reduction for Package Queries}, author = {Vasileios Vittis and Azza Abouzied and Peter J. Haas and Alexandra Meliou}, year = {2026}, journal = {Proceedings of the VLDB Endowment}, volume = {19}, number = {12}, pages = {4586--4589}, doi = {10.14778/3827998.3828072} }

Selecting a bundle of items that collectively satisfies constraints is a fundamental task across databases, recommender systems, and text summarization. The Ex2Bundle framework lets users specify intent through example bundles, synthesizes package queries from those examples, and relaxes infeasible constraints while preserving alignment with user intent.

@article{Cho2026, title = {Example-Driven Intent Synthesis for Constrained Data Bundle Retrieval: Focused Text Snippet Extraction and Beyond}, author = {Whanhee Cho and Kuangfei Long and Mahmood Jasim and Matteo Brucato and Alexandra Meliou and Peter J. Haas and Anna Fariha}, url = {https://arxiv.org/abs/2605.19246}, doi = {10.48550/arXiv.2605.19246}, year = {2026}, journal = {arXiv preprint arXiv:2605.19246} }