Deep reinforcement learning: How to avoid the hype and make it work for you

Description

Deep learning, or more specifically, deep reinforcement learning, has become a hot topic in the general rush to launch AI products. Some practitioners of deep reinforcement learning are prone to “Maslow’s hammer” cognitive bias, trying to solve a problem with deep reinforcement learning because they are fascinated by it or know it is good for marketing. Other practitioners, particularly small startups, are prone to fear of the unknown. Since deep reinforcement learning’s implementation in autonomous robots and self-driving cars is not well understood, they equate it to something that should be applied only when traditional deterministic AI and shallow machine learning models are tried and they fail.

Drawing on Conversica’s experience applying deep reinforcement learning to enable the company’s AI assistants to have autonomous multiturn conversations with millions of humans to fulfill business objectives, Sid Reddy explains how to avoid the hype and decide which use cases are best for its application. Sid shares best practices in deep reinforcement learning and highlights techniques that have been successful in various applications of artificial intelligence. Sid also discusses how to avoid some common pitfalls such as dimensionality curse, covers the basics of the Markov decision process, using conversational AI as an example, and explains how to optimally set up the environment, states, agent actions, transition probabilities, reward functions, and end states. You’ll also learn when to use end-to-end reinforcement learning and when it’s more effective to use deep reinforcement learning as a component.