2021-03-17 MAR
Journal Club 9:30-10:30
Daniel
https://research.google/pubs/pub46201/
The dominant sequence transduction models are based on complex recurrent or convolutional neural networks in an encoder-decoder configuration. The best performing models also connect the encoder and decoder through an attention mechanism. We propose a new simple network architecture, the Transformer, based solely on attention mechanisms, dispensing with recurrence and convolutions entirely. Experiments on two machine translation tasks show these models to be superior in quality while being more parallelizable and requiring significantly less time to train. Our model achieves 28.4 BLEU on the WMT 2014 English-to-German translation task, improving over the existing best results, including ensembles by over 2 BLEU. On the WMT 2014 English-to-French translation task, our model establishes a new single-model state-of-the-art BLEU score of 41.8 after training for 3.5 days on eight GPUs, a small fraction of the training costs of the best models from the literature. We show that the Transformer generalizes well to other tasks by applying it successfully to English constituency parsing both with large and limited training data.
A classic AI paper. Seminal paper on Transformers. Even though the paper is on an NLP task (language translation English -> German; English -> French), it has deep implications across data science.
1. Very influential in NLP (gave rise to transformers like GPT and BERT that replaced traditional CNNs and RNNs).
2. Relevant to natural language encoding (e.g., SOCCer bot).
3. Relevant to time-series analysis.
4. Will be relevant to any sequence analysis (e.g., genomics).
5. Starting to revolutionize image analytics.
6. Was critical to solve the protein folding problem (AlphaFold 2).
Jeya's suggestions of additional material:
https://jalammar.github.io/illustrated-transformer
https://arxiv.org/pdf/2010.11929.pdf
https://huggingface.co
https://arxiv.org/pdf/2010.11929.pdf
Hackathon 10:30-11:30
Firestore > BigQuery > R <still most pressing topic>
Nicole, Lorena - how can we tell if the process is faithful
Nicole et. al discussed noSQL pains in BQ tabular formats. These news that GCP, like Microsoft, is heading the noSQL route as well may be interesting: https://www.crn.com/news/applications-os/google-cloud-mongodb-take-their-alliance-to-the-next-level
Secondary data tables - BigQuery
Polygenic Risk Scores
[Jonas] The PGS story - we have an API !
Plotly composite plots
Abhinav show and tell
BigQuery integration
https://episphere.github.io/qaqc, Lorena
MISC
https://www.economist.com/graphic-detail/2021/03/11/how-we-built-our-covid-19-risk-estimator