Data Science by ODS.ai 🦜

2021-05-10 16:11:38 ​​Are Pre-trained Convolutions Better than Pre-trained Transformers?

In this paper, the authors from Google Research wanted to investigate whether CNN architectures can be competitive compared to transformers on NLP problems. It turns out that pre-trained CNN models outperform pre-trained Transformers on some tasks; they also train faster and scale better to longer sequences.

Overall, the findings outlined in this paper suggest that conflating pre-training and architectural advances is misguided and that both advances should be considered independently. The authors believe their research paves the way for a healthy amount of optimism in alternative architectures.

Paper: https://arxiv.org/abs/2105.03322

A detailed unofficial overview of the paper: https://andlukyane.com/blog/paper-review-cnnbettertransformers

#nlp #deeplearning #cnn #transformer #pretraining Open / Comment

2021-05-04 11:33:20 ​​MDETR: Modulated Detection for End-to-End Multi-Modal Understanding

Multi-modal reasoning systems rely on a pre-trained object detector to extract regions of interest from the image. However, this crucial module is typically used as a black box, trained independently of the downstream task and on a fixed vocabulary of objects and attributes.
The authors present an end-to-end approach to multi-modal reasoning systems, which works much better than using a separate pre-trained decoder.
They pre-train the network on 1.3M text-image pairs, mined from pre-existing multi-modal datasets having explicit alignment between phrases in text and objects in the image.
Fine-tuning this model achieves new SOTA results on phrase grounding, referring expression comprehension, and segmentation tasks. The approach could be extended to visual question answering.
Furthermore, the model is capable of handling the long tail of object categories.

Paper: https://arxiv.org/abs/2104.12763
Code: https://github.com/ashkamath/mdetr

A detailed unofficial overview of the paper: https://andlukyane.com/blog/paper-review-mdetr

#deeplearning #multimodalreasoning #transformer Open / Comment

2021-04-30 06:40:45 Open / Comment

2021-04-27 17:35:35 The Annotated Transformer

3 years ago Alexander Rush created an incredible notebook supported the "Attention is All You Need" paper giving a possibility to dive in the implementation details and obtain your own transformer :)

We, SkoltechNLP group, within our Neual NLP 2021 course revisited this notebook for adapting it as a seminar. Of course, the original code was created 3 years ago and in some places is incompatible with new versions of required libraries. As a result, we created "runnable with 'Run all Cells' for April 2021" version of this notebook:
https://github.com/skoltech-nlp/annotated-transformer

So if you want to learn the Transformer and run an example in your computer or Colab, you can save your time and use current version of this great notebook. Also, we add some links to the amazing resources about Transformers that emerged during these years:
* Seq2Seq and Attention by Lena Voita;
* The Illustrated Transformer.

Enjoy your Transformer! And be free to ask any questions and leave comments. Open / Comment

2021-04-24 14:52:19 Channel photo updated Open / Comment

2021-04-23 22:46:01 For almost 5 years channel picture beared arbitrary picture found in google and now we updated it with a proper new channel logo generated by neural network. Do you like it? Open / Comment

2021-04-23 22:42:46 Channel photo updated Open / Comment

2021-04-23 15:32:52 Researchers from Berkeley rolled out VideoGPT - a transformer that generates videos.

The results are not super "WOW", but the architecture is quite simple and now it can be a starting point for all future work in this direction. As you know, GPT-3 for text generation was also not built right away. So let's will wait for method acceleration and quality improvement.

Paper
Code
Project page
Demo Open / Comment