In machine learning, the fitness function is often visualised as a landscape with the height at each point in the space of all parameters being the fittness. When the aim is to maximise the fitness functions, this means finding the highest points, the mountain peaks, of the landscape ... hence the term hill climbing. For algorithms, such as simulated annealing where the goal is to minimise the fitness function (the 'energy'), the aim is to find the deepest valley bottoms, and so the term energy landscape is often also used. The landscape analogy is based on two Euclidean parameters being the coordinates of a point of the land on a map. In most situations there are many parameters, indeed many billions for a deep neural network, many of which may be categorcal r binary rtaher than continuous numbers. This means that the analogy, while useful, may not offer perfect intuition in every situation.
Used on Chap. 4: page 76; Chap. 9: pages 175, 183, 184, 185, 186, 187, 190, 196