The overarching theme of “Mastering the Mountains” is exploring reinforcement learning (RL) through an interactive lens, using the mountain car problem as a pedagogical tool to showcase how learning agents deal with dynamic environments. The exhibition aims to make complex computational theories more digestible through intuitive, hands-on understanding, allowing participants to engage with the core principles of RL in a gamified format.
Reinforcement learning, a sub-class of machine learning, is rooted in how agents learn to make decisions through interactions with their environment, aiming to maximise cumulative rewards. This exhibit translates these abstract ideas into a tangible experience by guiding participants through three phases of the mountain car problem. As players progress, they face the evolving complexity of the environment and the influence of a reward system, allowing them to understand the basics of agent-environment interactions in RL.
The exhibit introduces audiences to the core mechanics of RL in a hands-on manner. It illustrates how various advanced RL algorithms operate. Players gain an intuitive understanding of key RL concepts by stepping into the role of a learner (agent). Gamifying RL concepts demystify a field that is often regarded as mathematically dense and abstract. The accompanying visualisations of various RL algorithms further illustrate how artificial agents learn and adapt over time.
In an academic context, the exhibit builds on the Mountain Car problem, a well-known benchmark in RL research, and introduces it as an accessible entry point for novices. Initially described in Andrew Moore’s PhD thesis, the problem is lauded for its ability to capture the complexity of RL with a simple, comprehensible scenario.