Safe Reinforcement Learning through Meta-learned Instincts

Keyphrases

• 2D Navigation 50%
• Action Space 50%
• Deep Reinforcement Learning (Deep RL) 50%
• Hazardous State 50%
• Inherent Risk 50%
• Instinct 100%
• Meta-learning 50%
• Navigation Task 50%
• New Targets 100%
• Noisy Output 50%
• Reinforcement Learning 100%
• Safe Reinforcement Learning 100%
• Safety-critical Domain 50%
• Training Phase 50%

Computer Science

• Deep Reinforcement Learning 33%
• Navigation Task 33%
• Reinforcement Learning 100%
• Trained Network 66%
• Training Phase 33%

Neuroscience

• Reinforcement Learning 100%