The Monthly Briefing

It is often said that life is a succession of learnings, and, as a general rule, the older you get the more wisdom and experience you accumulate. Could this concept be applied to the building of AI models?

With the emergence in 2018 of self-supervised language models such as BERT (Google) - trained on massive amounts of text - an era is beginning in which Transfer Learning is becoming a reality for Natural Language Processing (NLP), just as it has been for the field of Computer Vision since 2013. To improve the way in which we interact with our clients, at BBVA AI Factory we have experimented with Transfer Learning techniques applied to models in different languages. But, what is Transfer Learning?

We refer to Transfer Learning when we reuse knowledge acquired from performing a task to tackle new similar tasks. However, most Machine Learning algorithms can only solve the task for which they were trained.

Let's imagine a Chef who has learnt to cook ravioli carbonara. The goal of Transfer Learning is for our Chef to be able to apply what s/he has learnt, - in this case, cooking pasta - to make a decent spaghetti Bolognese.

Check out this article on how we applied Transfer Learning in Natural Language Processing, in order to process texts in different languages faster. You can also watch it in this video!

A nice chat with BEDROCK

Highly recommended this podcast (Spanish: episode 2, season 2) from BEDROCK - Data by Design. Our colleague María Hernandez Rubio talks about the work we do at the AI Factory - no question is too many!

Maria tells how the work in data science has evolved after years of implementation. From the early more exploratory stages, to productive stages where data models, ML or AI are part of complete data products; where other aspects come into play.

Thank you so much both Bedrock Team and María for this conversation!

Talking about AI in Playz (RTVE)

Playz, RTVE's digital content channel for young people, has recently launched the programme Whaat!?, a space focused on the future of humanity. Our colleague Clara Higuera has participated in the technology-focused episode, in which she talks about what an analytical model does and urges us to think of Artificial Intelligence as a tool for improving our society. Don't miss it!

Further reading

+Which flavor of data professional are you? (Towards Data Science)
Cassie Kozyrkov also uses in this article the simile of the culinary world to explain the different tasks carried out by different professional profiles in the data universe. "There are not enough hours in the day for one person to do everything alone", as she says. Tasks also depend on the project phases.

+Goldman Cleared of Bias in New York Review of Apple Card (Bloomberg)
In the end it seems that there was no bias in the model. May have been an issue of explainability: "deficiencies in customer service and a perceived lack of transparency undermined consumer trust". Read more.

+Data Scientists in Software Teams: State of the Art and Challenges (Microsoft | Miryung Kim, Thomas Zimmermann, Robert DeLine and Andrew Begel)
The authors present a large-scale survey with 793 professional data scientists at Microsoft to understand their educational background, problem topics that they work on, tool usages, and activities. In addition, they cluster these data scientists based on the time spent for various activities and identify 9 distinct clusters of data scientists, and their corresponding characteristics. In the following figure, each row corresponds to a cluster and each column to an activity. The top number in each cell corresponds to the percentage of time a person spends on an activity, and the bottom number show to how many hours this time corresponds.

+Fireside Chat with Andrew Ng (Royal Statistical Society)

+Machine Learning into production (Idealista)
The Data Team from Idealista shares in this article some lessons learned in Data Science projects to improve the percentage of projects that go into production (in spanish).

+SEPLN (NLP Spanish Society) publishes the Natural Language Processing Strategy. (SEPLN)

+How to approach a text classification problem (David Morcuende)
A series of articles on NLP with accessible code, to learn how AI is able to process and classify text (1) (2).

+Your AI Model is ‘Wrong’ Yet It Can Transform Your Business (Towards Data Science | Ganes Kesari)
A recipe for data flywheels: (1) Benchmark the model accuracies against human performance / (2) Improve outcomes by augmenting models with human intelligence / (3) Check if the model has scope for continuous improvement / (4) Compute the business value of your model’s outcomes.

Quote

of the

month

"By itself, petabytes of data is of no consequence to decision making. To inform decisions, data needs to be processed. It needs to be transformed to make it valuable."

Fred Senekal, Head Of Research And Development at Learning Machines

This makes our days 😊

We are happy to help our customers in their day-to-day!

Keep readings folks!

< Read previous issue