Traditional computational drug discovery relies almost exclusively on highly task-specific computational models for hit identification and lead optimization. Adapting these specialized models to new tasks requires substantial time, computational power, and expertise—challenges that grow when researchers simultaneously work across multiple targets or properties.
]]>Text-to-image diffusion models have been established as a powerful method for high-fidelity image generation based on given text. Nevertheless, diffusion models do not always grant the desired alignment between the given input text and the generated image, especially for complicated idiosyncratic prompts that are not encountered in real life. Hence, there is growing interest in efficiently fine…
]]>This is part of a series on how researchers at NVIDIA have developed methods to improve and accelerate sampling from diffusion models, a novel and powerful class of generative models. Part 1 introduced diffusion models as a powerful class for deep generative models and examined their trade-offs in addressing the generative learning trilemma. While diffusion models satisfy both the first and…
]]>This is part of a series on how NVIDIA researchers have developed methods to improve and accelerate sampling from diffusion models, a novel and powerful class of generative models. Part 2 covers three new techniques for overcoming the slow sampling challenge in diffusion models. Generative models are a class of machine learning methods that learn a representation of the data they are trained…
]]>After the first successes of deep learning, designing neural network architectures with desirable performance criteria for a given task (for example, high accuracy or low latency) has been a challenging problem. Some call it alchemy and some intuition, but the task of discovering a novel architecture often involves a tedious and costly trial-and-error process of searching in an exponentially large…
]]>