Shijing Tu

ORCID: 0009-0009-6609-762X
Publications
Citations
Views
---
Saved
---
About
Contact & Profiles
Research Areas
  • Reinforcement Learning in Robotics
  • Crystallization and Solubility Studies
  • X-ray Diffraction in Crystallography
  • Synthetic Organic Chemistry Methods
  • Sentiment Analysis and Opinion Mining
  • Chemical Synthesis and Reactions
  • Catalytic C–H Functionalization Methods
  • Pregnancy-related medical research
  • Catalytic Alkyne Reactions
  • Liver Disease Diagnosis and Treatment
  • Diet, Metabolism, and Disease
  • Advanced Text Analysis Techniques
  • Global Maternal and Child Health
  • Text and Document Classification Technologies
  • Maternal and Perinatal Health Interventions
  • Radical Photochemical Reactions
  • Alcohol Consumption and Health Effects
  • Fluorine in Organic Chemistry

Southwest Medical University
2021-2022

ABSTRACT Objective We aimed to evaluate the differences in clinical features and lifestyle between Han ethnic minority populations Guangdong Province, China their impacts on ever‐growing burden of metabolic dysfunction‐associated steatotic liver disease (MASLD). Methods In this cross‐sectional investigation China, one most densely populated areas with imbalanced development, multistage stratified random sampling was used. Demographic, socioeconomic, data participants were collected....

10.1111/1751-2980.13331 article EN Journal of Digestive Diseases 2025-02-16

Offline preference-based reinforcement learning (PbRL) typically operates in two phases: first, use human preferences to learn a reward model and annotate rewards for reward-free offline dataset; second, policy by optimizing the learned via RL. However, accurately modeling step-wise from trajectory-level preference feedback presents inherent challenges. The bias introduced, particularly overestimation of predicted rewards, leads optimistic trajectory stitching, which undermines pessimism...

10.1609/aaai.v39i20.35388 article EN Proceedings of the AAAI Conference on Artificial Intelligence 2025-04-11

To assess the current status of caesarean delivery (CD) in China, propose reference CD rates for China overall, and by regions, investigate main indications CDs identify possible areas safe reduction. A multicentre cross-sectional study. total 94 hospitals across 23 provinces China. 73 977 randomly selected deliveries. We used a modified Robson classification to characterise subgroups World Health Organization (WHO) C-Model calculate rates. In 2015-2016, overall rate was 38.9% (95% CI...

10.1111/1471-0528.16951 article EN BJOG An International Journal of Obstetrics & Gynaecology 2021-09-25

A chemodivergent photocatalytic approach to 1-pyrrolines and 1-tetralones from alkyl bromides vinyl azides has been developed through chemoselectively controllable intermolecular [3 + 2] [4 cyclization. This photoredox-neutral two-component protocol involves radical addition switchable distal C(sp3)-H functionalization enabled by iminyl radical-mediated 1,5-hydrogen atom transfer. Meanwhile, chemoselectivity between C(sp3)-N bond formation C(sp3)-C(sp2) is precisely switched photocatalysts...

10.3389/fchem.2022.1058596 article EN cc-by Frontiers in Chemistry 2022-10-25

Offline preference-based reinforcement learning (PbRL) typically operates in two phases: first, use human preferences to learn a reward model and annotate rewards for reward-free offline dataset; second, policy by optimizing the learned via RL. However, accurately modeling step-wise from trajectory-level preference feedback presents inherent challenges. The bias introduced, particularly overestimation of predicted rewards, leads optimistic trajectory stitching, which undermines pessimism...

10.48550/arxiv.2412.09104 preprint EN arXiv (Cornell University) 2024-12-12

Preference-based reinforcement learning (PbRL) provides a powerful paradigm to avoid meticulous reward engineering by rewards based on human preferences. However, real-time feedback is hard obtain in online tasks. Most work suppose there "scripted teacher" that utilizes privileged predefined provide preference feedback. In this paper, we propose RL Self-augmented Large Language Model Feedback (RL-SaLLM-F) technique does not rely information for PbRL. RL-SaLLM-F leverages the reflective and...

10.48550/arxiv.2412.16878 preprint EN arXiv (Cornell University) 2024-12-22
Coming Soon ...