Despite the remarkable success of generative models, they are not without their shortcomings, which may pose substantial risks and significant repercussions in critical applications. Therefore, it is essential to comprehend potential risks and to be aware of flaws and misconceptions associated with these models.
Motivated by these considerations, I conduct theoretical and practical investigations of the biases that generative models are subject to. Specifically, my thesis explores different types of inductive bias and dataset bias in generative models and introduces the novel concept of generator bias.