Publications

LifeAgentBench: A Multi-dimensional Benchmark and Agent for Personal Health Assistants in Digital Health

Ye Tian$^*$, Zihao Wang$^*$, Onat Gungor, Xiaoran Fan, Tajana Rosing

Published in International Joint Conference on Artificial Intelligence (IJCAI) (under review), 2026

We introduce LifeAgentBench, a large-scale QA benchmark (22,573 questions) for long-horizon, cross-dimensional, and multi-user lifestyle health reasoning over structured health records (diet, activity, sleep, and emotion). We provide a standardized evaluation protocol, evaluate 11 leading LLMs, and propose LifeAgent, a training-free tool-calling baseline that improves performance via multi-step evidence retrieval and deterministic aggregation.

Recommended citation: Tian, Y.$^*$, Wang, Z.$^*$, Gungor, O., Fan, X., & Rosing, T. (2026). LifeAgentBench: A Multi-dimensional Benchmark and Agent for Personal Health Assistants in Digital Health. arXiv:2601.13880. Under review at IJCAI. https://arxiv.org/abs/2601.13880

DailyLLM: Context-Aware Activity Log Generation Using Multi-Modal Sensors and LLMs

Ye Tian, Xiaoyuan Ren, Zihao Wang, Onat Gungor, Xiaofan Yu, Tajana Rosing

Published in 22nd IEEE International Conference on Mobile Ad-Hoc and Smart Systems (MASS) (published), 2025

DailyLLM is a lightweight LLM-based framework for context-aware activity log generation and summarization using multimodal smartphone and smartwatch sensors. It achieves 17% higher BERTScore precision and 10× faster inference compared to state-of-the-art methods while using only a 1.5B-parameter model.

Recommended citation: Tian, Y., Ren, X., Wang, Z., Gungor, O., Yu, X., & Rosing, T. (2025). DailyLLM: Context-Aware Activity Log Generation Using Multi-Modal Sensors and LLMs. In 22nd IEEE International Conference on Mobile Ad-Hoc and Smart Systems (MASS). https://arxiv.org/abs/2507.13737

Double-Flow GAN Model for the Reconstruction of Perceived Faces from Brain Activities

Zihao Wang, Jing Zhao, Hui Zhang

Published in arXiv (preprint), 2024

This paper proposes Double-Flow GAN, a dual-stream generative adversarial framework that reconstructs perceived human faces from fMRI brain activity. The model leverages a feature-aligned latent space to enhance perceptual realism and structural accuracy, achieving state-of-the-art performance in neural decoding.

Recommended citation: Wang, Z., Zhao, J., & Zhang, H. (2024). Double-Flow GAN Model for the Reconstruction of Perceived Faces from Brain Activities. arXiv preprint arXiv:2312.07478. https://arxiv.org/abs/2312.07478v2