Deep continual learning pipelines for out-of-hospital medical emergencies text classification under the presence of dataset shifts.
This repository contains code developed for implementing and evaluating various deep continual learning text classification pipelines. It also facilitates the examination of temporal dataset shifts. The associated code accompanies a manuscript published at Computers in Biology and Medicine.
The study considers three types of temporal dataset shifts:
- Temporal prior probability shifts.
- Temporal covariate shift.
- Temporal concept shift.
For deep continual text classification baselines, the following approaches were evaluated:
- Static modeling.
- Single fine-tuning.
- Joint training.
The deep continual learning text pipelines implemented include:
- Cumulative learning.
- Continual fine-tuning.
- Replay.
- Synaptic intelligence.
This code has been tested on a protected database of out-of-hospital medical emergencies from the Valencian Community (Spain), encompassing a total of 1 982 746 independent medical incidents.
Furthermore, this code can serve as a template for numerous other applications that require a deep continual text classification approach.
The methods and evaluation results are published in the following article, please cite it if you use this code:
Pablo Ferri, Vincenzo Lomonaco, Lucia C. Passaro, Antonio Félix-De Castro, Purificación Sánchez-Cuesta, Carlos Sáez and Juan M García-Gómez. "Deep continual learning for medical call incidents text classification under the presence of dataset shifts". Computers in Biology and Medicine, 108548 (2024). https://doi.org/10.1016/j.compbiomed.2024.108548.
- Main developer: Pablo Ferri, Ph.D.
- Authors: Pablo Ferri (UPV), Vincenzo Lomonaco (UNIPI), Lucia C.Passaro (UNIPI), Antonio Félix-De Castro (GVA), Purificación Sánchez-Cuesta (GVA), Carlos Sáez (UPV) and Juan M García-Gómez (UPV)
Copyright: 2024 - Biomedical Data Science Lab, Universitat Politècnica de València, Spain (UPV)
This project was supported by the Ministry of Science, Innovation, and Universities of Spain through the FPU18/06441 program and partially funded by the PNRR-M4C2-Investment 1.3, Extended Partnership PE00000013-FAIR (Future Artificial Intelligence Research)-Spoke 1 Human-centered AI, financed by the European Commission under the NextGeneration EU program.