Jiao Liu

Research profile

Research summary

Jiao Liu is a researcher at Nanyang Technological University in Singapore working at the interface of evolutionary computation and machine-learning-based optimization. The profile is name-resolved from public bibliographic data (OpenAlex), and the Evolve-reviewed catalog contains two 2026 arXiv preprints, both in middle-author, non-corresponding positions. The first applies a frozen TabPFN numerical foundation model to few-shot multitask optimization, using in-context priors to guide inter-task coupling; Evolve judged the results strong while noting the missing ablation that would isolate the informed guidance. The second proposes IST, an iterative sequential transfer formulation of few-shot multiobjective multitask optimization with GP-kernel task prioritization; its few-shot gains are judged promising, though win counts slightly outrun the reported tables. OpenAlex topics — stochastic gradient optimization techniques (2 works) plus metaheuristic, reinforcement-learning-in-robotics, multi-objective, and bandit topics (1 each) — together with an even 0.5/0.5 split across the two catalog fields explain the low focus score, even though both papers converge on few-shot multitask optimization. Recorded metrics include 2 recent-3-year works, 6 distinct coauthors, and a reproducibility mean of 78 across the 2 scored papers (one high, one medium readiness). Output and citation-impact metrics are flagged missing in this snapshot, so the profile should be read as early-stage.

Research keywords

Highlights

• Frozen TabPFN brings in-context priors to few-shot multitask optimization

  In a 2026 arXiv preprint, a frozen TabPFN numerical foundation model guides inter-task coupling in few-shot multitask optimization via in-context priors; Evolve judged the results strong but noted the absence of an ablation isolating the informed guidance. Reproducibility readiness: medium.

• IST reframes multitask optimization as iterative sequential transfer

  A second 2026 arXiv preprint proposes IST, treating multitask optimization as iterative sequential transfer with GP-kernel task prioritization for the few-shot multiobjective setting; Evolve finds the few-shot gains promising, while observing that win counts slightly outrun the tables. Reproducibility readiness: high.

• Reproducibility mean of 78 across both scored papers

  Evolve's reproducibility scoring averages 78 over the two catalog papers — one rated high and one medium readiness — making reproducibility the strongest recorded dimension in this early-stage profile.

Timeline

1. 2026In-Context Guidance preprint appears on arXiv, applying a frozen TabPFN foundation model to few-shot multitask optimization.
2. 2026Second preprint on iterative sequential transfer (IST) for few-shot multiobjective multitask optimization appears on arXiv; OpenAlex records 2 works for the year.
3. 2026Both papers enter the Evolve catalog with reproducibility scores averaging 78 (one high, one medium readiness), in middle-author, non-corresponding positions.

Papers in the Evolve catalog

• 2026-09-22

  In-Context Guidance: Learning Inter-Task Synergies via Numerical Foundational Models for Few-Shot Multitask Optimization

• 2026-09-10

  Solving Few-Shot Multiobjective Multitask Optimization via Iterative Sequential Transfer

Frequent co-authors