Title: A neural network account to Kant's philosophical aesthetics

Abstract: According to Kant's (1724 -- 1804) philosophical aesthetics, laid down in his Critique of the Power of Judgement (1790), beauty is "subjective purposefulness", reflected by the "harmony of the cognitive faculties", which are "understanding" and "imagination". On the one hand, understanding refers to the mental capability to find regularities in sensory manifolds, while imagination refers to intuition, fantasy, and creativity of the mind, on the other hand. Inspired by the reinforcement learning theory of Schmidhuber, I present a neural network analogy for the harmony of the faculties in terms of generative adversarial networks (GAN) - also often employed for artificial music composition -- by identifying the generator module with the faculty of imagination and the discriminator module with the faculty of understanding. According to the GAN algorithm, both modules are engaged in an adversarial game, thereby optimizing a particular objective function. In my reconstruction, the convergence of the GAN algorithm during the reception of art, either music or fine, entails the harmony of the faculties and thereby a neural network analogue of subjective purposefulness, i.e., beauty.