Skip to yearly menu bar Skip to main content


Poster

SQ-VAE: Variational Bayes on Discrete Representation with Self-annealed Stochastic Quantization

Yuhta Takida · Takashi Shibuya · WeiHsiang Liao · Chieh-Hsin Lai · Junki Ohmura · Toshimitsu Uesaka · Naoki Murata · Shusuke Takahashi · Toshiyuki Kumakura · Yuki Mitsufuji

Hall E #402

Keywords: [ PM: Bayesian Models and Methods ] [ T: Probabilistic Methods ] [ OPT: Stochastic ] [ MISC: Representation Learning ] [ DL: Generative Models and Autoencoders ]


Abstract:

One noted issue of vector-quantized variational autoencoder (VQ-VAE) is that the learned discrete representation uses only a fraction of the full capacity of the codebook, also known as codebook collapse. We hypothesize that the training scheme of VQ-VAE, which involves some carefully designed heuristics, underlies this issue. In this paper, we propose a new training scheme that extends the standard VAE via novel stochastic dequantization and quantization, called stochastically quantized variational autoencoder (SQ-VAE). In SQ-VAE, we observe a trend that the quantization is stochastic at the initial stage of the training but gradually converges toward a deterministic quantization, which we call self-annealing. Our experiments show that SQ-VAE improves codebook utilization without using common heuristics. Furthermore, we empirically show that SQ-VAE is superior to VAE and VQ-VAE in vision- and speech-related tasks.

Chat is not available.