Deep learning approaches for real-world settings: formulation, architecture, optimization

Ruxandra Stoean

Deep learning recently finds itself at the core of most successful tools. We make use of these applications in our everyday lives, as they are incorporated into software or machines from different domains: medicine, industry, finance, education, culture, and entertainment. The increase of computational resources and big data have made this amazing evolution possible. However, at the development end of this impressive image lies a proportionally big effort. One must find the appropriate way to translate the task into a data mining formulation, to decide on the appropriate deep architecture and further optimize the learning. The current presentation aims to share some lessons learned from personal experience in tailoring deep learning to real-world scenarios in medicine, engineering, and cultural heritage.