Publications

PML4SC Publications

8 papers listed

3 accepted papers

2026 latest publication year

Published

2026

conference Accepted

Closing the Capacity-Convergence Gap: Globally Optimal Configuration of Implicit Neural Representations

Sipeng Chen , Yan Zhang , Shibo Li

ECCV 2026 European Conference on Computer Vision 2026

OptiINR closes the capacity-convergence gap in implicit neural representations by formulating activation and initialization configuration as a principled global optimization problem.

BibTeX

@inproceedings{chen2026optinr,
  title = {Closing the Capacity--Convergence Gap: Globally Optimal Configuration of Implicit Neural Representations},
  author = {Chen, Sipeng and Zhang, Yan and Li, Shibo},
  booktitle = {Computer Vision -- ECCV 2026},
  pages = {134--151},
  year = {2026},
  volume = {17002},
  series = {Lecture Notes in Computer Science},
  publisher = {Springer Nature Switzerland},
  doi = {10.1007/978-3-032-36969-7_8}
}

conference Accepted

Regime-Adaptive Bayesian Optimization via Dirichlet Process Mixtures of Gaussian Processes

Yan Zhang , Xuefeng Liu , Sipeng Chen , Sascha Ranftl , Chong Liu , Shibo Li

ICML 2026 Proceedings of the 43rd International Conference on Machine Learning 2026

A regime-adaptive Bayesian optimization framework that models heterogeneous response patterns with Dirichlet process mixtures of Gaussian processes.

BibTeX

@inproceedings{zhang2026rambo,
  title = {Regime-Adaptive {B}ayesian Optimization via {D}irichlet Process Mixtures of {G}aussian Processes},
  author = {Zhang, Yan and Liu, Xuefeng and Chen, Sipeng and Ranftl, Sascha and Liu, Chong and Li, Shibo},
  booktitle = {Proceedings of the 43rd International Conference on Machine Learning},
  pages = {157148--157179},
  year = {2026},
  volume = {306},
  series = {Proceedings of Machine Learning Research},
  publisher = {PMLR}
}

conference Accepted

COMPOL: A Unified Neural Operator Framework for Scalable Multi-Physics Simulations

Junqi Qu , Tao Wang , Yushun Dong , Hewei Tang , Shibo Li

IJCNN/WCCI 2026 International Joint Conference on Neural Networks, IEEE World Congress on Computational Intelligence 2026

A unified neural operator framework for scalable multi-physics simulations, currently available as an arXiv preprint and accepted by IJCNN 2026.

arXiv

PDF

Code

BibTeX

@article{qu2025compol,
  title={{COMPOL}: A Unified Neural Operator Framework for Scalable Multi-Physics Simulations},
  author={Qu, Junqi and Wang, Tao and Dong, Yushun and Tang, Hewei and Li, Shibo},
  journal={arXiv preprint arXiv:2501.17296},
  year={2025}
}

Preprints

2026

preprint

Conformal Cascade: Distribution-Free Accuracy Guarantees for Multi-Tier LLM Inference

Yifan Dou , Shikan Lian , Shibo Li

arXiv 2026 arXiv preprint 2026

Conformal Cascade uses conformal prediction set size as the deferral rule for multi-tier LLM inference, delivering a distribution-free, finite-sample accuracy guarantee with no model training and black-box API access only.

arXiv

PDF

Code

BibTeX

@article{dou2026conformal,
  title={Conformal Cascade: Distribution-Free Accuracy Guarantees for Multi-Tier {LLM} Inference},
  author={Dou, Yifan and Lian, Shikan and Li, Shibo},
  journal={arXiv preprint arXiv:2607.25018},
  year={2026}
}

preprint

Verbalized Particle Posterior: Bayesian Inference over Natural Language Hypotheses

Yan Zhang , Shikan Lian , Shibo Li

arXiv 2026 arXiv preprint 2026

VPP treats verbalized machine learning as Bayesian inference, maintaining a population of natural-language hypotheses as particles updated by Metropolis-Hastings or sequential Monte Carlo and predicting by Bayesian model averaging.

arXiv

PDF

Code

BibTeX

@article{zhang2026verbalized,
  title={Verbalized Particle Posterior: {Bayesian} Inference over Natural Language Hypotheses},
  author={Zhang, Yan and Lian, Shikan and Li, Shibo},
  journal={arXiv preprint arXiv:2607.22961},
  year={2026}
}

preprint

ConsistencyGate: Preventing Memory Contamination in LLM Agents via Self-Consistency Admission Control

Yan Zhang , Shibo Li

arXiv 2026 arXiv preprint 2026

ConsistencyGate is a write-time admission gate that scores a candidate fact for self-consistency before committing it to an LLM agent's memory, curbing the contamination that otherwise compounds across long trajectories.

arXiv

PDF

Code

BibTeX

@article{zhang2026consistencygate,
  title={{ConsistencyGate}: Preventing Memory Contamination in {LLM} Agents via Self-Consistency Admission Control},
  author={Zhang, Yan and Li, Shibo},
  journal={arXiv preprint arXiv:2607.22962},
  year={2026}
}

preprint

Multi-Fidelity Flow Matching: Cascaded Refinement of PDE Solutions

Sipeng Chen , Junliang Liu , Hewei Tang , Shibo Li

arXiv 2026 arXiv preprint 2026

MFFM calibrates the conditional flow-matching source distribution to the empirical low-to-high-fidelity residual scale, yielding a learned analog of multigrid refinement that reaches the finest grid in one network evaluation per cascade level.

arXiv

PDF

Code

BibTeX

@article{chen2026multi,
  title={Multi-Fidelity Flow Matching: Cascaded Refinement of {PDE} Solutions},
  author={Chen, Sipeng and Liu, Junliang and Tang, Hewei and Li, Shibo},
  journal={arXiv preprint arXiv:2605.16118},
  year={2026}
}

2025

preprint

Dynamic Bayesian Optimization Framework for Instruction Tuning in Partial Differential Equation Discovery

Junqi Qu , Yan Zhang , Shangqian Gao , Shibo Li

arXiv 2025 arXiv preprint 2025

NeuroSymBO frames instruction tuning for PDE discovery as a sequential decision problem and uses Bayesian optimization to adaptively select reasoning strategies during multi-step generation.

arXiv

PDF

Code

BibTeX

@article{qu2025neurosymbo,
  title={Dynamic {Bayesian} Optimization Framework for Instruction Tuning in Partial Differential Equation Discovery},
  author={Qu, Junqi and Zhang, Yan and Gao, Shangqian and Li, Shibo},
  journal={arXiv preprint arXiv:2601.00088},
  year={2025}
}