A paper posted on arXiv proposes OrderFlowDiT, a model for probabilistic limit-order supply forecasting. The task is to predict the conditional distribution of limit-order volume submitted over the next five minutes, given the current order book and recent order flow.
The model is built on a diffusion transformer with transport regularization. According to the abstract, the problem requires capturing price placement, total volume, and predictive aspects of order flow, which the architecture is designed to handle.
This work appears to be a methodological contribution to financial machine learning, specifically for high-frequency limit-order book modeling. No experimental results are described in the available abstract, so the practical performance of OrderFlowDiT remains to be seen.