✳flâneur — a map of the web's best reading

Jupyter Notebook Viewer

nbviewer.org · 3,355 words · saved by 1 readers

Variational Auto Encoders (VAEs) can be thought of as what all but the last layer of a neural network is doing, namely feature extraction or seperating out the data. Thus given some data we can think of using a neural network for representation generation. Recall that the goal of a generative model is to estimate the probability distribution of high dimensional data such as images, videos, audio or even text by learning the underlying structure in the data as well as the dependencies between the different elements of the data. This is very useful since we can then use this representation to generate new data with similar properties. This way we can also learn useful features from the data in an unsupervised fashion. The VQ-VAE uses a discrete latent representation mostly because many important real-world objects are discrete. For example in images we might have categories like "Cat", "Car", etc. and it might not make sense to interpolate between these categories. Discrete representatio

Jupyter Notebook Viewer pytorch-vq-vae vq-vae.ipynb Notebook VQ-VAE by Aäron van den Oord et al. in PyTorch ¶ Introduction ¶ Variational Auto Encoders (VAEs) can be thought of as what all but the last layer of a neural network is doing, namely feature extraction or seperating out the data. Thus given some data we can think of using a neural network for representation generation. Recall that the goal of a generative model is to estimate the probability distribution of high dimensional data such as images, videos, audio or even text by learning the underlying structure in the data as well as the

Explore this link on the map →

related reading