Centrala begrepp
Developers face common challenges in Deep Reinforcement Learning applications, with comprehension, API usage, and design problems being prominent.
Sammanfattning
The article discusses the challenges faced by developers in Deep Reinforcement Learning (DRL) applications. It presents a taxonomy of challenges based on a large-scale empirical study of Stack Overflow posts. The study categorizes challenges into DRL issues, DL issues, DRL libraries/frameworks, parallel processing & multi-threading, and general programming issues. Key insights include the prevalence of challenges like comprehension, API usage, and design problems. The survey validation confirms that practitioners encounter these challenges and perceive them as critical with medium to high effort required for resolution.
Structure:
- Introduction to Machine Learning adoption.
- Overview of Deep Reinforcement Learning (DRL).
- Challenges faced by developers in DRL applications.
- Taxonomy creation based on Stack Overflow posts.
- Survey validation of identified challenges.
- Comparison of SO post analysis and survey results.
Statistik
Results show that at least 45% of developers experienced 18 of the 21 challenges identified in the taxonomy.
The majority (exceeding 52%) of survey respondents indicated that the most frequent challenges are critical.
Citat
"We provide the first large-scale empirical study of the challenges in the development of DRL applications."
"Results show that all challenges presented in our taxonomy were encountered by the survey respondents."