Export
CHONG BIAN
2 previous affiliations
- China Mobile Research Institute2022–2026
- employee · 2024–2026
- Employee · 2022–2024
- Beihang University2017–2022
- PhD student · 2017–2022
Research Overview
CHONG BIAN focuses on applied machine learning and predictive modeling across diverse engineering domains, including high-speed rail communication systems, audio processing, and battery health estimation. Recent work investigates the application of advanced neural architectures, such as energy-based models and large language models, to complex state estimation and self-supervised learning tasks.
Research Trajectory
The research trajectory of CHONG BIAN demonstrates a consistent focus on applying advanced deep learning architectures to complex, real-world engineering and signal processing challenges. Early work addressed the reliability of ground-train communication in high-speed rail networks. As the first author of Holistic Transmission Performance Prediction of Balise System With Gate-Steered Residual Interweave Networks (2023), Bian tackled the difficulties of predicting long-term transmission performance under multisource disturbances. This research introduced a gate-steered residual interweave architecture, utilizing a gated double-dilated temporal convolutional network (GDTCN) to extract multiscale features and filter out irrelevant information caused by ground-train coupling and on-board demodulation.
Subsequent collaborations expanded into audio processing and self-supervised learning (SSL). In 2025, Bian participated as a co-author in Energy-based Model Guided Self-Supervised Learning for Speaker Verification. This work addressed scenarios with limited labeled data by integrating Energy-Based Models (EBMs) into the DINO (Distillation with No Labels) framework. The resulting EBCA-DINO approach utilized energy scores to evaluate data complexity and uncertainty, enabling adaptive temperature scaling that dynamically adjusted the student model's focus during label-free self-distillation.
Most recently, Bian has led research applying foundation models to energy storage systems. As the first author of Joint state-of-charge and state-of-health estimation of lithium-ion batteries across varying operational stages on differing timescales with large language model: a multi-task prompting method (2026), Bian investigated the use of large language models (LLMs) for battery management. This research developed a multi-task prompting method to simultaneously estimate the state-of-charge and state-of-health of lithium-ion batteries, specifically addressing the challenges of varying operational stages and differing timescales.
Current Directions
Current directions in this research portfolio emphasize the adaptation of generalized machine learning frameworks—such as large language models and energy-based models—for specialized diagnostic and estimation tasks. The most recent papers in the evidence address the challenge of extracting reliable state estimates from complex, multi-timescale systems, particularly in the domain of lithium-ion battery management. This involves developing novel multi-task prompting strategies to bridge the gap between natural language architectures and quantitative engineering metrics like state-of-charge and state-of-health. Additionally, ongoing collaborative work explores the refinement of self-supervised learning techniques to improve model confidence and performance in data-constrained environments.
Profile updated
Loading new papers…
Loading research neighbors…
See something wrong? Report a mistake