Everybody knows how successful transformers have been in NLP. Is there known work on other domains (e.g that also have a sequential natural way of occurring, such as stock price prediction or other problems)?
$\begingroup$
$\endgroup$
4
-
$\begingroup$ Image generation with transformers instead of GANs is a thing. $\endgroup$– thepackerJun 25 '20 at 7:51
-
$\begingroup$ do you have a paper or a reference? $\endgroup$– Schach21Jun 25 '20 at 17:02
-
1$\begingroup$ cdn.openai.com/papers/Generative_Pretraining_from_Pixels_V2.pdf see also the completions section at their website. openai.com/blog/image-gpt $\endgroup$– thepackerJun 25 '20 at 17:37
-
$\begingroup$ I've recently start seeing transformer-based models being used for traffic forecasting. Here is an example: arxiv.org/abs/2001.02908 $\endgroup$– razvanc92Aug 3 '20 at 8:27
Add a comment
|
$\begingroup$
$\endgroup$
When it talks to other domains such as image or music, using transformer will always face a problem of sequence length limitation. To the best of my knowledge, the bottleneck of self-attention which uses a $n^2$ matrix quite limits transformer being applied to other domains. For example, a 32x32 pixel image, means a sequence of 1024 tokens.
OpenAI did some related research, as the followings.
Generative Modeling with Sparse Transformers: In the paper, transformers with sparse attention are applied to image and waveform.
ImageGPT: A large transformer model trained on language can generate coherent text, the same exact model trained on pixel sequences can generate coherent image completions and samples. (Abstract from the blog)
