Jialiang Tong

Computer-assisted research for Chinese patent medicines and natural products. Services cover ingredient data, potential targets, preliminary safety screening, quality standards, and clinical study design, with a focus on investigating new uses for existing medicines.

A method for prodromal staging and subtype identification in neurodegenerative diseases. It combines variable decorrelation and an improved EM algorithm with progression-sequence constraints (UDC), MMD-DTW, and genetic optimisation to support staging, interpretability, and computational efficiency. The research aims to support earlier risk assessment; clinical benefits require further validation.

Outputs: patent application CN118116582A; global Top 10 in the 2024 BAPP Algorithm Competition.

Research using electricity load data from 10,068 enterprises in Zhejiang. The model combines convolutional networks, LSTM, and self-attention to capture local and long-range patterns, with updated observations and iterative forecasting. Experiments compare performance in real-time and short-term prediction settings.

Thesis: Jialiang Tong. Research on Electricity Data Processing and Load Forecasting. Zhejiang University, 2023.

An agent coordination approach for elderly care, combining data, knowledge, inference, and collaboration. It supports care teams through baseline assessment, changes in condition, hospital visits, and return to care.

Award: second prize, 2026 MinerU Data Intelligence and Frontier Corpus Challenge.

AAT-Former addresses domain shift across market regimes and trading venues. It processes limit order books, trades, and quotes using time bucketing and DTW-K-means clustering, followed by causal convolutions and Transformer-based temporal encoding.

Cross-attention, LSTM, and MLP modules produce return, direction, and volatility signals. A gradient reversal layer and maximum mean discrepancy alignment encourage representations that transfer across regimes and venues.