Wangyang Ying

I have published papers in top-tier venues, including KDD, NeurIPS, EMNLP, AAAI, CIKM, IJCAI, and TKDD. A complete list of publications is available on Google Scholar.

1. [TKDD] Feature Selection as Deep Sequential Generative Learning.
   • Formulates feature selection as a sequential generative process, enabling structured and controllable selection of informative features for downstream prediction tasks.
2. [KDD] Unsupervised Generative Feature Transformation via Graph Contrastive Pre-training and Multi-objective Fine-tuning.
   • Transforms features into simpler and more discriminative representation spaces, enabling strong downstream performance with lightweight models.
3. [NeurIPS] Sculpting Features from Noise: Reward-Guided Hierarchical Diffusion for Task-Optimal Feature Transformation.
   • Proposed a reward-guided latent diffusion framework that reformulates discrete feature transformation as a generative process, enabling stable global exploration and task-optimal feature construction beyond local continuous search.
4. [AAAI] Efficient Post-Training Refinement of Latent Reasoning in Large Language Models.
   • Proposed a training-free, post-training approach to improve the stability and accuracy of multi-step reasoning in large language models by refining latent reasoning trajectories.
5. [IJCAI] Unsupervised Feature Transformation via In-context Generation, Generator-critic LLM Agents, and Duet-play Teaming.
   • Introduced an agent-based, generative formulation for discrete search and feature transformation, enabling effective learning from unlabeled data through iterative feedback.
6. [CIKM] Revolutionizing Biomarker Discovery: Leveraging Generative AI for Bio-Knowledge-Embedded Continuous Space Exploration.
   • Applied generative feature selection to high-dimensional biological data to identify a compact set of informative features, improving disease prediction performance.
7. [EMNLP] Title2Event: Benchmarking Open Event Extraction with a Large-scale Chinese Title Dataset.
   • Event extraction from short news titles, enabling event-level representation, aggregation, and event-driven news search.