Publications

• Heterogeneous Aligned Fusion for Survival Prediction with Missing Modalities

  Paper

  Zheng, Z.; Guo, Y.; Hu, X.; Miao, Y.; Ma, H.; Gao, J.; Huang, J.

  Medical Imaging with Deep Learning (MIDL), 2026 (To appear)

  We study multimodal survival prediction under naturally missing modalities and propose a heterogeneous aligned fusion framework to improve robustness and cross-modality alignment.

• Multiple Abstraction Level Retrieve Augment Generation

  arXiv

  Zheng, Z.; Ni, X.; Hong, P.

  arXiv preprint, 2025

  We propose MAL-RAG, a retrieval-augmented generation framework that retrieves evidence at multiple abstraction levels (multi-sentence, paragraph, section, document) to mitigate context fragmentation and the 'lost-in-the-middle' effect in scientific QA.

• FiBiNet++: Reducing Model Size by Low-Rank Feature Interaction Layer for CTR Prediction

  DOI

  Zhang, P.; Zheng, Z.; Zhang, J.

  CIKM '23, 2023

  We introduce a low-rank feature interaction layer to compress FiBiNet-style CTR models, substantially reducing non-embedding parameters while maintaining competitive performance.