Deep Learning applications for COVID-19.

Area of Science:

• Artificial Intelligence
• Computational Biology
• Epidemiology

Background:

• The COVID-19 pandemic spurred rapid advancements in artificial intelligence applications.
• Deep Learning (DL) emerged as a powerful tool across multiple scientific disciplines to address pandemic challenges.

Purpose of the Study:

• To survey the diverse applications of Deep Learning in combating the COVID-19 pandemic.
• To identify key limitations and suggest future research directions for DL in pandemic response.

Main Methods:

• Literature review of Deep Learning applications in Natural Language Processing, Computer Vision, Life Sciences, and Epidemiology.
• Analysis of how data availability and learning task construction influence DL performance.
• Evaluation of current DL capabilities and limitations in the context of COVID-19.

Main Results:

• DL has been applied to Natural Language Processing (information retrieval, question answering, misinformation detection, sentiment analysis), Computer Vision (medical image analysis, ambient intelligence, robotics), Life Sciences (diagnostics, protein structure prediction, drug repurposing), and Epidemiology (spread forecasting).
• Effectiveness of DL applications is contingent on big data availability and task-specific learning.
• Identified limitations include interpretability, generalization, learning from limited data, and data privacy.

Conclusions:

• Numerous Deep Learning systems have been developed to combat COVID-19.
• Addressing DL limitations is crucial for accelerating its adoption and effectiveness in future public health crises.
• Further research is needed to enhance DL's robustness, interpretability, and privacy-preserving capabilities for pandemic preparedness.