Novo logótipo da Solutional (1)

Modelo de difusão

Partilhar isto
" Voltar ao Índice do Glossário

Diffusion models are a key concept in computador[2] vision, image generation, and natural language processing. They are tools used to perform tasks like image denoising, inpainting, super-resolution, and text generation. They can be trained to clean up images blurred by Gaussian noise. A few examples of these models include denoising diffusion probabilistic models and noise conditioned score networks. They also play a vital role in non-equilibrium thermodynamics, where they help sample from complex probability distributions. They are further enhanced by advanced techniques like variational inference[1] and stochastic gradient descent. In the field of natural language processing, they are used for text generation and summarization, learning the latent structure of text data to produce contextually relevant text. Research entities such as OpenAI[3] e Google[4] Imagen have developed various diffusion models for image and text generation tasks.

Definições de termos
1. inference. Inference is a cognitive process that involves drawing conclusions from available evidence and reasoning. It's a fundamental component of critical thinking and problem-solving, playing a significant role in fields as diverse as scientific research, literature interpretation, and artificial intelligence. There are several types of inference, including deductive, inductive, abductive, statistical, and causal, each with its own unique approach and application. For instance, deductive inference is about deriving specific conclusions from general principles, while inductive inference forms general conclusions from specific observations. On the other hand, abductive inference is about making educated guesses based on available evidence, while statistical and causal inferences involve interpreting data to draw conclusions about a population or to determine cause-and-effect relationships. However, biases, preconceptions, and misinterpretations can influence the accuracy of inferences. Despite these challenges, inference remains an essential skill that can be improved through practice, critical thinking exercises, and engaging in diverse reading materials.
2. computador. Um computador é um dispositivo sofisticado que manipula dados ou informações de acordo com um conjunto de instruções, conhecidas como programas. Por conceção, os computadores podem executar uma vasta gama de tarefas, desde simples cálculos aritméticos até ao processamento e análise de dados complexos. Os computadores evoluíram ao longo dos anos, desde as primitivas ferramentas de contagem, como o ábaco, até às modernas máquinas digitais. O coração de um computador é a sua unidade central de processamento (CPU), que inclui uma unidade lógica aritmética (ALU) para efetuar operações matemáticas e registos para armazenar dados. Os computadores também têm unidades de memória, como ROM e RAM, para armazenar informação. Outros componentes incluem dispositivos de entrada/saída (E/S) que permitem a interação com a máquina e circuitos integrados que melhoram a funcionalidade do computador. As principais inovações históricas, como a invenção do primeiro computador programável por Charles Babbage e o desenvolvimento do primeiro computador eletrónico digital automático, o Atanasoff-Berry Computer (ABC), contribuíram grandemente para a sua evolução. Atualmente, os computadores alimentam a Internet, ligando milhares de milhões de utilizadores em todo o mundo, e tornaram-se uma ferramenta essencial em quase todas as indústrias.
Modelo de difusão (Wikipédia)

In machine learning, diffusion models, also known as diffusion probabilistic models ou score-based generative models, are a class of latent variable generative models. A diffusion model consists of three major components: the forward process, the reverse process, and the sampling procedure. The goal of diffusion models is to learn a diffusion process that generates a probability distribution for a given dataset from which we can then sample new images. They learn the latent structure of a dataset by modeling the way in which data points diffuse through their latent space.

In the case of computer vision, diffusion models can be applied to a variety of tasks, including image denoising, inpainting, super-resolutione image generation. This typically involves training a neural network to sequentially denoise images blurred with Gaussian noise. The model is trained to reverse the process of adding noise to an image. After training to convergence, it can be used for image generation by starting with an image composed of random noise for the network to iteratively denoise. Announced on 13 April 2022, OpenAI's text-to-image model DALL-E 2 is an example that uses diffusion models for both the model's prior (which produces an image embedding given a text caption) and the decoder that generates the final image. Diffusion models have recently found applications in natural language processing (NLP), particularly in areas like text generation and summarization.

Diffusion models are typically formulated as markov chains and trained using variational inference. Examples of generic diffusion modeling frameworks used in computer vision are denoising diffusion probabilistic models, noise conditioned score networks, and stochastic differential equations.

" Voltar ao Índice do Glossário
pt_PT_ao90PT
Deslocar para o topo