In this talk I will first provide some context on the state of the field of quantum machine learning and recent successes of quantum deep learning. Quantum deep learning consists of learning representations of data using composite networks of parameterized quantum transformations, also known as parameterized quantum circuits or quantum neural networks. Following this introduction, I will introduce a new class of generative quantum-neural-network-based models called Quantum Hamiltonian-Based Models (QHBMs). These models provide a novel paradigmatic approach for quantum-probabilistic hybrid variational learning, where one efficiently decomposes the tasks of learning classical and quantum correlations in a way which maximizes the utility of both classical and quantum processors, thereby combining the capabilities of classical probabilistic deep learning and quantum deep learning. Following this will be the introduction of the Variational Quantum Thermalizer (VQT) algorithm for generating the thermal state of a given Hamiltonian and target temperature, and an explanation of how QHBM's are naturally suited for this task. The VQT can be seen as a generalization of the Variational Quantum Eigensolver (VQE) to thermal states. As another application of QHBM's, we will define the tasks of Quantum Modular Hamiltonian learning, where one learns to generatively model mixed quantum states as the thermal state of a learned Hamiltonian. I will provide numerical results demonstrating the efficacy of these techniques in illustrative examples. Namely, using QHBMs for modelling Heisenberg spin systems, to learn entanglement Hamiltonians and compression codes in simulated free Bosonic systems, and to create thermal states for quantum simulation of Fermionic systems.