Perhaps one of the most everyday uses of artificial intelligence that we can experience in our day-to-day lives is through interaction with artificial vision and object identification systems. From unlocking our smartphone to searching for images on the Internet. All these functionalities are possible thanks to artificial intelligence models in the field of image detection and classification. In this post we compile some of the most important open image repositories, thanks to which we have been able to train current image recognition models.

Introduction

Let's go back for a moment to late 2017, early 2018. The possibility of unlocking our smartphones with some kind of fingerprint reader has become widespread. With more or less success, most manufacturers had managed to include the biometric reader in their terminals. The unlocking time, the ease of use and the extra security provided were exceptional compared to the classic systems of passwords, patterns, etc. As has been the case since 2008, the undisputed leader in digital innovation in mobile terminals - Apple - revolutionised the market once again by incorporating an innovative unlocking system in the iPhone X using an image of our face. The so-called FaceID system scans our face to unlock the terminal in tenths of a second without having to use our hands. The probability of identity theft with this system was 1 in 1,000,000; 20 times more secure than its predecessor TouchID

Let this little story about an everyday functionality be used to introduce an important topic in the field of artificial intelligence, and in particular in the field of computer image processing: AI model training image repositories. We have talked a lot in this space about this field of artificial intelligence. A few months after the launch of FaceID, we published a post on AI, in which we mentioned near-human-level image classification as one of the most important achievements of AI in recent years. This would not be possible without the availability of open banks of annotated images^[1]  to train image recognition and classification models. In this post we list some of the most important (freely available) image repositories for model training.

Of course, recognising the number plate of a vehicle at the entrance to a car park is not the same as identifying a lung disease in an X-ray image. The banks of annotated images are as varied as the potential AI applications they enable.

Probably the 2 best known image repositories are MNIST and ImageNET.

• MNIST, is a set of 70,000 black and white images of handwritten numbers normalised in size, ready to train number recognition algorithms. Professor LeCun's original paper is from 1998.

• ImageNET is a huge database of concepts (words or sets of words). Each concept with its own meaning is called a synset. Each synset is represented by hundreds or thousands of images. ImageNET's own website cites the project as an indispensable tool for the recent advancement of Deep Learning and computer vision.

The project has been instrumental in advancing computer vision and deep learning research. The data is available for free to researchers for non-commercial use

The most widely used subset of ImageNet is ImageNet Large Scale Visual Recognition Challenge ILSVRC, an image classification and localisation dataset. This image subset was used from 2010 to 2017 for the worldwide object detection and image classification competitions. This dataset covers 1000 object classes and contains more than one million training images, 50,000 validation images and 100,000 test images. This subset is available in Kaggle.

In addition to these two classic repositories that are already part of the history of image processing by artificial intelligence, we have some more current and varied thematic repositories. Here are some examples:

• The very annoying CAPTCHAs and reCAPTCHAs that we find on a multitude of websites to verify that we are human trying to access are a good example of artificial intelligence applied to the field of security. Of course, CAPTCHAs also need their own repository to check how effective they are in preventing unwanted access. We recommend reading this interesting article about the history of these web browsing companions.

• As we have seen several times in the past, one of the most promising applications of AI in the field of imaging is to assist physicians in diagnosing diseases from a medical imaging test (X-ray, CT scan, etc.). To make this a reality, there is no shortage of efforts to collect, annotate and make available to the research community repositories of quality, anonymised medical images to train models for detecting objects, shapes and patterns that may reveal a possible disease. Breast cancer are 30% of all cancers in women worldwide. Hence the importance of having image banks that facilitate the training of specific models.

• The diagnosis of blood-based diseases often involves the identification and characterisation of patient blood samples. Automated methods (using medical imaging) to detect and classify blood cell subtypes have important medical applications.

• Three years ago, Covid19 burst into our lives, turning developed societies upside down with this global pandemic with terrible consequences in terms of human and economic loss. The entire scientific community threw itself into finding a solution in record time to tackle the consequences of the new coronavirus. Many efforts were made to improve the diagnosis of the disease. Some techniques relied on AI-assisted image analysis. At the same time, health authorities incorporated a new element in our daily routine - face masks. Even today, in some situations the mask is still mandatory, and during these 3 years we have had to monitor its proper use in almost all kinds of places. So much so that in recent months there has been a proliferation of specific image banks to train AI and computer vision models to detect the use of masks autonomously.

• For more information on open repositories related to health and wellbeing, we leave you with this post we published a few months ago.

In addition to these curious examples cited in this post, we encourage you to explore Kaggle's section of datasets that include images as data. You only have 10,000 sets to browse through ;)

Content prepared by Alejandro Alija, expert in Digital Transformation and Innovation.

The contents and views expressed in this publication are the sole responsibility of the author.

After several months of tests and different types of training, the first massive Artificial Intelligence system in the Spanish language is capable of generating its own texts and summarising existing ones. MarIA is a project that has been promoted by the Secretary of State for Digitalisation and Artificial Intelligence and developed by the National Supercomputing Centre, based on the web archives of the National Library of Spain (BNE).

This is a very important step forward in this field, as it is the first artificial intelligence system expert in understanding and writing in Spanish. As part of the Language Technology Plan, this tool aims to contribute to the development of a digital economy in Spanish, thanks to the potential that developers can find in it.

The challenge of creating the language assistants of the future

MarIA-style language models are the cornerstone of the development of the natural language processing, machine translation and conversational systems that are so necessary to understand and automatically replicate language. MarIA is an artificial intelligence system made up of deep neural networks that have been trained to acquire an understanding of the language, its lexicon and its mechanisms for expressing meaning and writing at an expert level.

Thanks to this groundwork, developers can create language-related tools capable of classifying documents, making corrections or developing translation tools.

The first version of MarIA was developed with RoBERTa, a technology that creates language models of the "encoder" type, capable of generating an interpretation that can be used to categorise documents, find semantic similarities in different texts or detect the sentiments expressed in them.

Thus, the latest version of MarIA has been developed with GPT-2, a more advanced technology that creates generative decoder models and adds features to the system. Thanks to these decoder models, the latest version of MarIA is able to generate new text from a previous example, which is very useful for summarising, simplifying large amounts of information, generating questions and answers and even holding a dialogue.

Advances such as the above make MarIA a tool that, with training adapted to specific tasks, can be of great use to developers, companies and public administrations. Along these lines, similar models that have been developed in English are used to generate text suggestions in writing applications, summarise contracts or search for specific information in large text databases in order to subsequently relate it to other relevant information.

In other words, in addition to writing texts from headlines or words, MarIA can understand not only abstract concepts, but also their context.

More than 135 billion words at the service of artificial intelligence

To be precise, MarIA has been trained with 135,733,450,668 words from millions of web pages collected by the National Library, which occupy a total of 570 Gigabytes of information. The MareNostrum supercomputer at the National Supercomputing Centre in Barcelona was used for the training, and a computing power of 9.7 trillion operations (969 exaflops) was required.

Bearing in mind that one of the first steps in designing a language model is to build a corpus of words and phrases that serves as a database to train the system itself, in the case of MarIA, it was necessary to carry out a screening to eliminate all the fragments of text that were not "well-formed language" (numerical elements, graphics, sentences that do not end, erroneous encodings, etc.) and thus train the AI correctly.

Due to the volume of information it handles, MarIA is already the third largest artificial intelligence system for understanding and writing with the largest number of massive open-access models. Only the language models developed for English and Mandarin are ahead of it. This has been possible mainly for two reasons. On the one hand, due to the high level of digitisation of the National Library's heritage and, on the other hand, thanks to the existence of a National Supercomputing Centre with supercomputers such as the MareNostrum 4.

The role of BNE datasets

Since it launched its own open data portal (datos.bne.es) in 2014, the BNE has been committed to bringing the data available to it and in its custody closer: data on the works it preserves, but also on authors, controlled vocabularies of subjects and geographical terms, among others.

In recent years, the educational platform BNEscolar has also been developed, which seeks to offer digital content from the Hispánica Digital Library's documentary collection that may be of interest to the educational community.

Likewise, and in order to comply with international standards of description and interoperability, the BNE data are identified by means of URIs and linked conceptual models, through semantic technologies and offered in open and reusable formats. In addition, they have a high level of standardisation.

Next steps

Thus, and with the aim of perfecting and expanding the possibilities of use of MarIA, it is intended that the current version will give way to others specialised in more specific areas of knowledge. Given that it is an artificial intelligence system dedicated to understanding and generating text, it is essential for it to be able to cope with lexicons and specialised sets of information.

To this end, the PlanTL will continue to expand MarIA to adapt to new technological developments in natural language processing (more complex models than the GPT-2 now implemented, trained with larger amounts of data) and will seek ways to create workspaces to facilitate the use of MarIA by companies and research groups.

Content prepared by the datos.gob.es team.

Data has become one of the pillars of society's digital transformation process, which also challenges sectors such as justice and law enforcement. Thanks to them, access to information and statistics has been improved, allowing decision-making to be based on objective figures to which new techniques such as automation and artificial intelligence can be applied.

Thus, and with the aim of continuing to delve into the advantages derived from the data ecosystem, on 17 and 18 October, the University of Salamanca is organising, in collaboration with the Ministry of Justice, a symposium on Justice and Data.

What will be the themes to be addressed?

During the two days of the event and through the various presentations, the aim will be to discuss "the role of data for the proper functioning of public services". In other words, how open data can help to improve the efficiency and effectiveness with regard to citizens and the services offered to them.

In line with this idea, the questions that will form part of the symposium will revolve around the following themes:

• Personalised assistants
• Data Analytics
• Designing Data Visualisations
• Governance, Transparency and Open Data
• AI - NLP
• AI - Other
• Robotisation
• Data Sharing Spaces
• Data, AI, RPA Training

Thus, while the first day will be made up of conferences by relevant people from the Justice, Law and academic sectors, the second day will showcase the different initiatives of the international technology sector and the legal-tech sector related to data.

Likewise, in simultaneous rooms, public and private projects in which technology is applied to the services of Justice and Law will be analysed.

In conclusion, it is an event that seeks to become a meeting point for innovation in the field of justice. So that, through the exchange of experiences and success stories, between the administration, institutions and private companies in any field, it will be possible to guide the use of data to provide and give solutions to specific problems.

How can I attend?

Attendance at the conference will be free of charge and will take place at the Hospederia Arzobispo Fonseca in Salamanca. In order to attend, it will be necessary to fill in the following registration form. As with other similar events, this one will also be broadcast live online.

According to the latest analysis conducted by Gartner in September 2021, on Artificial Intelligence trends, Chatbots are one of the technologies that are closest to deliver effective productivity in less than 2 years. Figure 1, extracted from this report, shows that there are 4 technologies that are well past the peak of inflated expectations and are already starting to move out of the valley of disillusionment, towards states of greater maturity and stability, including chatbots, semantic search, machine vision and autonomous vehicles.

Figure 1-Trends in AI for the coming years.

In the specific case of chatbots, there are great expectations for productivity in the coming years thanks to the maturity of the different platforms available, both in Cloud Computing options and in open source projects, especially RASA or Xatkit. Currently it is relatively easy to develop a chatbot or virtual assistant without AI knowledge, using these platforms.

How does a chatbot work?

As an example, Figure 2 shows a diagram of the different components that a chatbot usually includes, in this case focused on the architecture of the RASA project.

Figure 2- RASA project architecture

One of the main components is the agent module, which acts as a controller of the data flow and is normally the system interface with the different input/output channels offered to users, such as chat applications, social networks, web or mobile applications, etc.

The NLU (Natural Languge Understanding) module is responsible for identifying the user's intention (what he/she wants to consult or do), entity extraction (what he/she is talking about) and response generation. It is considered a pipeline because several processes of different complexity are involved, in many cases even through the use of pre-trained Artificial Intelligence models.

Finally, the dialogue policies module defines the next step in a conversation, based on context and message history. This module is integrated with other subsystems such as the conversation store (tracker store) or the server that processes the actions necessary to respond to the user (action server).

Chatbots in open data portals as a mechanism to locate data and access information

There are more and more initiatives to empower citizens to consult open data through the use of chatbots, using natural language interfaces, thus increasing the net value offered by such data. The use of chatbots makes it possible to automate data collection based on interaction with the user and to respond in a simple, natural and fluid way, allowing the democratization of the value of open data.

At SOM Research Lab (Universitat Oberta de Catalunya) they were pioneers in the application of chatbots to improve citizens' access to open data through the Open Data for All and BODI (Bots to interact with open data - Conversational interfaces to facilitate access to public data) projects. You can find more information about the latter project in this article.

It is also worth mentioning the Aragón Open Data chatbot, from the open data portal of the Government of Aragón, which aims to bring the large amount of data available to citizens, so that they can take advantage of its information and value, avoiding any technical or knowledge barrier between the query made and the existing open data. The domains on which it offers information are: 

• General information about Aragon and its territory
• Tourism and travel in Aragon
• Transportation and agriculture
• Technical assistance or frequently asked questions about the information society.

Conclusions

These are just a few examples of the practical use of chatbots in the valorization of open data and their potential in the short term. In the coming years we will see more and more examples of virtual assistants in different scenarios, both in the field of public administrations and in private services, especially focused on improving user service in e-commerce applications and services arising from digital transformation initiatives.

Content prepared by José Barranquero, expert in Data Science and Quantum Computing.

The contents and points of view reflected in this publication are the sole responsibility of the author.

The pandemic situation we have experienced in recent years has led to a large number of events being held online. This was the case of the Iberian Conference on Spatial Data Infrastructures (JIIDE), whose 2020 and 2021 editions had a virtual format. However, the situation has changed and in 2022 we will be able to meet again to discuss the latest trends in geographic information.

Seville will host JIIDE 2022

Seville has been the city chosen to bring together all those professionals from the public administration, private sector and academia interested in geographic information and who use Spatial Data Infrastructures (SDI) in the exercise of their activities.

Specifically, the event will take place from 25 to 27 October at the University of Seville. You can find more information here.

Focus on user experience

This year's slogan is "Experience and technological evolution: bringing the SDI closer to citizens".  The aim is to emphasise new technological trends and their use to provide citizens with solutions that solve specific problems, through the publication and processing of geographic information in a standardised, interoperable and open way.

Over three days, attendees will be able to share experiences and use cases on how to use Big Data, Artificial Intelligence and Cloud Computing techniques to improve the analysis capacity, storage and web publication of large volumes of data from various sources, including real-time sensors.

New specifications and standards that have emerged will also be discussed, as well as the ongoing evaluation of the INSPIRE Directive.

Agenda now available

Although some participations are still to be confirmed, the programme is already available on the conference website. There will be around 80 communications where experiences related to real projects will be presented, 7 technical workshops where specific knowledge will be shared and a round table to promote debate.

Among the presentations there are some focused on open data. This is the case of Valencia City Council, which will talk about how they use open data to obtain environmental equity in the city's neighbourhoods, or the session dedicated to the "Digital aerial photo library of Andalusia: a project for the convergence of SDIs and Open-Data".

How can I attend?

The event is free of charge, but to attend you need to register using this form. You must indicate the day you wish to attend.

For the moment, registration is open to attend in person, but in September, the website of the conference will offer the possibility of participating in the JIIDE virtually.

Organisers

The Jornadas Ibéricas de Infraestructuras de Datos Espaciales (JIIDE) were born from the collaboration of the Directorate General of Territory of Portugal, the National Geographic Institute of Spain and the Government of Andorra. On this occasion, the Institute of Statistics and Cartography of Andalusia and the University of Seville join as organisers.

Artificial intelligence (AI) has been pervasive in our lives for several years now. While there is no exact definition of AI, one description could be "the ability of a machine to display human-like capabilities such as reasoning, learning, planning and creativity". This process is done through the creation and application of algorithms. In other words, AI refers to the ability of computers, software and other machines to think and act as humans would.

Artificial intelligence allows the development of different use cases that facilitate decision-making or provide solutions to complex problems. As a result, artificial intelligence has been able to revolutionise not only the business world but also the social sphere, with applications ranging from the rapid detection of cancer to the fight against deforestation in the Amazon, to name but a few examples.

Given all these advantages, it is not surprising that in recent years the demand for professional profiles related to this field has grown. Therefore, here are some examples of interesting courses and training that could help you to broaden your knowledge of artificial intelligence.

Online courses

Elements of AI

• Taught by: University of Helsinki and Reaktor
• Duration: 50 hours
• Language: English
• Price: Free

This massive open educational project (MOOC), which has already enrolled more than 750,000 students, offers you the opportunity to learn what artificial intelligence is, as well as showing how it can affect your work and personal life and how it will evolve in the coming years.

This course, which offers all citizens the possibility of training and first-hand knowledge of how AI works and the opportunities it offers, has been promoted in our country by the Secretary of State for Digitalisation and Artificial Intelligence together with the UNED.

Building AI

• Taught by: University of Helsinki and Reaktor
• Duration: 50 hours
• Language: English
• Price: Free

The authors of the previous course launched this new course some time later with the aim of closing the gap between courses for beginners such as 'Elements of AI' and the vast majority of training courses in this field, which usually present a more advanced level.

This course, which starts where the previous one ends, will help you to delve deeper into elements such as machine learning, neural networks or some practical applications of AI. It also offers you the option of setting up your first artificial intelligence project and getting started in programming if you wish.

Specialised programme: Introduction to artificial intelligence

• Taught by: Coursera (UNAM)
• Duration: 8 meses
• Language: Spanish
• Price: Free

This course is aimed at people who are interested in learning more about the different developments that have been generated over the last few years in the field of artificial intelligence.

If you opt for this training you will learn how to implement AI technology for a specific purpose, how to compare the solution you have developed with other existing ones or how to report the results obtained in a structured test.

Machine Learning Crash Course

• Taught by: Google
• Duration: 15 hours
• Idioma: English
• Price: Free

Through this course you will learn some key concepts such as the detailed study of machine learning and you will take your first steps with the TensorFlow API, among others.

To attend this course, it is recommended: some programming experience (preferably in Python), basic knowledge of machine learning, statistics, linear algebra and calculus.

Machine Learning

• Taught by: Coursera (Stanford)
• Duration:: 60 hours
• Language: English (English subtitles)
• Price: Free (47€ if you wish to apply for a certificate)

This is a MOOC on machine learning created by Andrew Ng, founder of the Google Brainproject in 2011 and the online course platform Coursera.

Through this course you will cover topics such as supervised and unsupervised learning, statistical pattern recognition or the application of 'best practices' in this field. In addition, you will also learn how to apply learning algorithms to the construction of intelligent robots, among many other aspects.

Masters

The current training offer on artificial intelligence does not only come in the form of courses. An increasing number of universities and study centres are offering their students specialised programmes and university master's degrees related to the field of AI. Here are just a few examples:

• Official Master's Degree in Computer Security and AI, Rovira i Virgili University (60 ECTS credits, one academic year): its programme covers research topics related to information protection, application security, machine learning, modelling and problem solving, among others.
• University Master's Degree in Artificial Intelligence, University of Santiago de Compostela (USC) (90 ECTS credits, 18 months): it offers students the possibility of broadening their knowledge in areas such as natural language processing, robotics, AI fundamentals or data engineering.
• Master's Degree in Intelligent Systems and Numerical Applications in Engineering, University of Las Palmas de Gran Canaria (ULPGC) (60 ECTS credits, 1 academic year): it introduces students to the bases and fundamentals of some leading areas such as computational modelling and numerical simulation of engineering problems, the use and development of intelligent and autonomous systems, or methods of data analysis and interpretation.
•  Master in Artificial Intelligence, Universitat Politécnica de Catalunya (90 ECTS credits, 18 months): it offers a solid foundation and advanced knowledge in artificial intelligence. In this master's degree you will learn key concepts in computational intelligence, computer vision and multi-agent systems.
• Master's Degree in Applied Artificial Intelligence, Universidad Carlos III Madrid (60 ECTS credits, 1 academic year): it allows students to understand the most relevant AI methods and techniques and apply them to develop appropriate solutions to different types of problems.
• Master's Degree in Artificial Intelligence, UNIR (60 ECTS credits, 1 academic year): in this master's degree you will learn techniques of computational perception and artificial vision, automatic reasoning and planning, machine learning and deep learning, Natural Language Processing, as well as the technologies needed to implement them.
• Master's Degree in AI Research, Spanish Association for Artificial Intelligence (AEPIA) (60 ECTS credits, one academic year): it has three specialisations: Learning and Data Science, Web Intelligence and Reasoning and Planning.
• Master in Artificial Intelligence, Universidad Europea (30 ECTS credits, 8 months): aimed at providing students with an integrated vision of Artificial Intelligence and the mastery of advanced Machine Learning and Computational Optimisation techniques.

This has been just a small compilation of training courses related to the field of artificial intelligence that we hope will be of interest to you. If you know of any other course or master's degree that you would like to recommend, do not hesitate to leave us a comment or send us an email to dinamizacion@datos.gob.es.

The AMETIC association represents companies of all sizes linked to the Spanish digital technology industry, a key sector for the national GDP. Among other issues, AMETIC seeks to promote a favorable environment for the growth of companies in the sector, promoting digital talent and the creation and consolidation of new companies.

At datos.gob.es we spoke with Antonio Cimorra, Director of Digital Transformation and Enabling Technologies at AMETIC, to reflect on the role of open data in innovation and as the basis for new products, services and even business models.

Full interview:

1. How does open data help drive digital transformation? What disruptive technologies are the most benefited by the opening of data?

Open data is one of the pillars of the data economy , which is called to be the basis of our present and future development and the digital transformation of our society. All industries, public administrations and citizens themselves have only just begun to discover and use the enormous potential and usefulness that the use of data brings to improving the competitiveness of companies, to the efficiency and improvement of services. of Public Administrations and to social relations and people's quality of life.

2. One of the areas in which they work from AMETIC is Artificial Intelligence and Big Data, among whose objectives is to promote the creation of public platforms for sharing open data. Could you explain to us what actions you are carrying out or have carried out for this?

At AMETIC we have an Artificial Intelligence and Big Data Commission in which the main companies that provide this technology participate . From this area, we work on the definition of initiatives and proposals that contribute to disseminating their knowledge among potential users, with the consequent advantages that their incorporation in the public and private sectors entails. Outstanding examples of actions in this area are the recent presentation of the AMETIC Artificial Intelligence Observatory, as well as the AMETIC Artificial Intelligence Summit. which in 2022 will celebrate its fifth edition that will focus on showing how Artificial Intelligence can contribute to meeting the Sustainable Development Goals and the Transformation Plans to be executed with European Funds

3. Open data can serve as a basis for developing services and solutions that give rise to new companies . Could you tell us an example of a use case carried out by your partners?

Open data, and very particularly the reuse of public sector information, are the basis for the development of countless applications and entrepreneurial initiatives both in consolidated companies in our technology sector and in many other cases of small companies or startups found in this source of information the motor of development of new businesses and approach to the market.

4. What types of data are most in demand by the companies you represent?

At present, all industrial and social activity data are in great demand by companies , due to their great value in the development of projects and solutions that have been demonstrating their interest and extension in all areas and types of organizations and users. usually.

5. It is also essential to have data sharing initiatives such as GAIA-X , built on the values ​​of digital sovereignty and data availability. How have companies received the creation of a national hub ?

The technology sector has received the creation of the GAIA-X national hub very positively, understanding that our contribution from Spain to this European project will be of enormous value to our companies from very different fields of activity. Data sharing spaces in sectors such as tourism, health, mobility, industry, to give a few examples, have Spanish companies and experiences that are an example and a reference at European and global level .

6. Right now there is a great demand for professionals related to data collection, analysis and visualization. However, the supply of professionals, although it is growing, continues to be limited . What should be done to boost training in skills related to data and digitization?

The supply of technology professionals is one of the biggest problems for the development of our local industry and for the digital transformation of society. It is a difficulty that we can describe as historical, and that far from going less, every day there is a greater number of positions and profiles to cover. It is a worldwide problem that shows that there is no single or simple formula to solve it, but we can mention the importance of all social and professional agents developing joint and collaborative actions that allow the digital training of our population from an early age. and cycles and specialized training and degree programs that are characterized by their proximity to what will be the professional careers for which it is necessary to have the participation of the business sector

7. During the last years, you have been part of the jury of the different editions of the Aporta Challenge. How do you think these types of actions contribute to boosting data-driven businesses?

The Aporta Challenge has been an example of support and encouragement for the definition of many projects around open data and for the development of its own industry that in recent years has been growing very significantly with the availability of data of very different groups, in many cases by the Public Administrations, and their subsequent reuse and incorporation into applications and solutions of interest to very different users.

Open data constitutes one of the pillars of the data economy, which is called to be the basis of our present and future development and of the digital transformation of our society.

8. What are the next actions that are going to be carried out in AMETIC linked to the data economy?

Among the most outstanding actions of AMETIC in relation to the data economy, it is worth mentioning our recent incorporation into the national hub of GAIA-X for which we have been elected members of its board of directors, and where we will represent and incorporate the vision and contributions of the digital technology industry in all the data spaces that are constituted , serving as a channel for the participation of the technology companies that carry out their activity in our country and that have to form the basis of the projects and use cases that integrate into the European network GAIA-X in collaboration with other national hubs.

Have you ever wondered how Alexa is able to recognise our voice and understand what we are saying (sometimes better than others)? Have you ever stopped to think about how Google is able to search for images similar to the one we are providing it with? You probably know that these techniques belong to the field of artificial intelligence. But don't be fooled, building these sophisticated models is only within the reach of a few. In this post we tell you why and what we ordinary mortals can do to train artificial intelligence models.

Introduction

In recent years we have witnessed incredible and surprising advances in the field of training Deep Learning models. On previous occasions we have cited the most relevant examples such as GPT-3 or Megatron-Turing NLG. These models, optimised for natural language processing (NLP), are capable of writing complete articles (practically indistinguishable from those written by a human) or making coherent summaries of classic works, hundreds of pages long, synthesising the content in just a few paragraphs. Impressive, isn't it?

However, these achievements are far from cheap. That is, the complexity of these models is such that thousands of gigabytes of pre-processed information - what we call annotated datasets - are needed, which have been previously analysed (labelled) by a human expert in the field. For example, the latest training of the Megatron-Turing NLG model, created in collaboration between Microsoft and NVIDIA, used 270 billion tokens (small pieces of text that can be words or sub-words that form the basis for training these natural language models). In addition to the information needed to be able to train these models, there is the fact of the special computational needs that these trainings require. To execute training tasks for these models, the most advanced machines (computers) in the world are needed, and training times are counted in weeks. Although there is no official data, some sources quote the cost of training the latest models such as GPT-3 or Megatron-Turing in the tens of millions of dollars. So how can we use and train models if we do not have access to the most powerful computing clusters in the world?

The answer: Transfer Learning

When working on a machine learning or deep learning project and we do not have access to large datasets ready for training, we can start from pre-trained models to create a new model adjusted or tuned to our specific use case. In other words, we load a previously trained model with a very large set of training data and re-train its final layers to fit our particular data set. This is known as Transfer Learning.

Original source: Transfer Learning in Deep Learning: Beyond our models. Post by Josu Alonso on Medium.

Simplifying a lot, we could say that traditional Machine Learning applies to isolated learning tasks, where it is not necessary to retain the acquired knowledge, while in a Transfer Learning project the learning is the result of previous tasks, achieving good precision in less time and with less data. This brings many opportunities, but also some challenges, such as the new domain inheriting biases from the previous domain.

Let's look at a concrete example. Suppose we have a new Deep Learning challenge and we want to make an automatic dog breed classifier. In this case, we can apply the transfer learning technique starting from a general image classification model, and then fit it to a specific set of dog breed photographs. Most of the pre-trained models are based on a subset of the ImageNet database, which we have already discussed on several occasions. The neural network (from ImageNet), which is the base type of algorithm used in these image classification models, has been trained on 1.2 million images of over 1000 different object categories such as keyboards, coffee cups, pencils and many animals. By using a pre-trained network to apply Transfer Learning, we get much faster and easier results than if we had to train a network from scratch.

For example, this code snippet shows the process of starting from a pre-trained model with ImageNet and re-training or adding new layers to achieve fine adjustments to the original model.

Conclusions

Training a general-purpose deep learning model is not within everyone's reach. There are several barriers, from the difficulty of accessing quality training data in sufficient volume, to the computational capacity needed to process billions of images or texts. For more limited use cases, where we only require a refinement of generalist models, applying the Transfer Learning technique allows us to achieve fantastic results in terms of accuracy and training time, at a cost that is affordable for most data scientists. Transfer Learning applications are very numerous and specialised websites are full of application examples. In line with this trend, Style Transfer Learning, which consists of reconstructing images based on the style of a previous image, has recently become very popular. We will continue to discuss this topic in future posts.

Example of Style Transfer Learning in Kaggle

Content prepared by Alejandro Alija, expert in Digital Transformation and Innovation.

The contents and views expressed in this publication are the sole responsibility of the author.

Today, Artificial Intelligence (AI) applications are present in many areas of everyday life, from smart TVs and speakers that are able to understand what we ask them to do, to recommendation systems that offer us services and products adapted to our preferences.

These AIs "learn" thanks to various techniques, including supervised, unsupervised and reinforcement learning. In this article we will focus on reinforcement learning, which focuses mainly on trial and error, similar to how humans and animals in general learn.

The key to this type of system is to correctly set long-term goals in order to find an optimal global solution, without focusing too much on immediate rewards, which do not allow for an adequate exploration of the set of possible solutions.

Simulation environments as a complement to open data sets.

Unlike other types of learning, where learning is usually based on historical datasets, this type of technique requires simulation environments that allow training a virtual agent through its interaction with an environment, where it receives rewards or penalties depending on the state and actions it performs. This cycle between agent and environment can be seen in the following diagram:

Figure 1 -  Scheme of learning by reinforcement [Sutton & Barto, 2015]

That is, starting from a simulated environment, with an initial state, the agent performs an action that generates a new state and a possible reward or penalty, which depends on the previous states and the action performed. The agent learns the best strategy in this simulated environment from experience, exploring the set of states, and being able to recommend the best action policy if configured appropriately.

The best-known example worldwide was the success achieved by AlphaGo, beating 18-time world champion Lee Sedol in 2016. Go is an ancient game, considered one of the 4 basic arts in Chinese culture, along with music, painting and calligraphy. Unlike chess, the number of possible game combinations is greater than the number of atoms in the Universe, being a problem impossible to solve by traditional algorithms.

Curiously, the technological breakthrough demonstrated by AlphaGo in solving a problem that was claimed to be beyond the reach of an AI, was eclipsed a year later by its successor AlphaGo Zero. In this version, its creators chose not to use historical data or heuristic rules. AlphaGo Zero only uses the board positions and learns by trial and error by playing against itself.

Following this innovative learning strategy, in 3 days of execution he managed to beat AlphaGo, and after 40 days he became the best Go player, accumulating thousands of years of knowledge in a matter of days, and even discovering previously unknown strategies.

The impact of this technological milestone covers countless areas, and AI solutions that learn to solve complex problems from experience can be counted on. From resource management, strategy planning, or the calibration and optimization of dynamic systems.

The development of solutions in this area is especially limited by the need for appropriate simulation environments, being the most complex component to build. However, there are multiple repositories to obtain open simulation environments that allow us to test this type of solutions.

The best known reference is Open AI Gym, which includes an extensive set of libraries and open simulation environments for the development and validation of reinforcement learning algorithms. Among others, it includes simulators for the basic control of mechanical elements, robotics applications and physics simulators, two-dimensional ATARI video games, and even the landing of a lunar module. In addition, it allows to integrate and publish new open simulators for the development of our own simulators adapted to our needs that can be shared with the community:

Figure 2 - Examples of visual simulation environments offered by  Open AI Gym

Another interesting reference is Unity ML Agents, where we also find multiple libraries and several simulation environments, also offering the possibility of integrating our own simulator:

Figure 3 - Examples of visual simulation environments offered by Unity ML Agents

Potential applications of reinforcement learning in public administrations

This type of learning is used especially in areas such as robotics, resource optimization or control systems, allowing the definition of optimal policies or strategies for action in specific environments.

One of the best-known practical examples is the DeepMind algorithm used by Google to reduce by 40% the energy consumption required to cool its data centers in 2016, achieving a significant reduction in energy consumption during use, as can be seen in the following graph (taken from the previous article):

Figure 4 - Results of the DeepMind algorithm on the energy consumption of Google's data centers.

The algorithm employed uses a combination of deep learning and reinforcement learning techniques, together with a general purpose simulator to understand complex dynamic systems that could be applied in multiple environments such as transformation between energy types, water consumption or resource optimization in general.

Other possible applications in the public domain include the search and recommendation of open datasets through chatbots, or the optimization of public policies, as is the case of the European project Policy Cloud, applied for example in the analysis of future strategies of the different designations of origin of wines from Aragon.

In general, the application of this type of techniques could optimize the use of public resources by planning action policies that result in more sustainable consumption, reducing pollution, waste and public spending.

Content prepared by Jose Barranquero, expert in Data Science and Quantum Computing.

The contents and views expressed in this publication are the sole responsibility of the author.

Over the last decade we have seen how national and international institutions, as well as national governments and business associations themselves have been warning about the shortage of technological profiles and the threat this poses to innovation and growth. This is not an exclusively European problem - and therefore also affects Spain – but, to a greater or lesser extent, it occurs all over the world, and has been further aggravated by the recent pandemic.

Anyone who has been working for some time, and not necessarily in the technology world, has been able to observe how the demand for technology-related roles has been increasing. It's nothing more than the consequence of companies around the world investing heavily in digitization to improve their operations and innovate their products, along with the growing presence of technology in all aspects of our lives.

And within technology professionals, during the last few years there is a group that has become a kind of unicorn due to its particular scarcity, data scientists and the rest of professionals related to data and artificial intelligence: data engineers, machine learning engineers, artificial engineering specialists in all areas, from data governance to the very configuration and deployment of deep learning models, etc.

This scenario is especially problematic for Spain where salaries are less competitive than in other countries in our environment and where, for starters, the proportion of IT workers is below the EU average. Therefore, it is foreseeable that Spanish companies and public administrations, which are also implementing projects of this type, will face increasing difficulties in recruiting and retaining talent related to technology in general, and data and artificial intelligence in particular.

When there is a supply problem, the only sustainable solution in the medium and long term is to increase the production of what is in short supply. In this case, the solution would be to incorporate new professionals into the labour market as the only mechanism to ensure a better balance between supply and demand. And this is recognized in all national and European strategies and plans related to digitalization, artificial intelligence and the reform of education systems, both higher education and vocational training.

Spanish Strategies

The National Artificial Intelligence Strategy dedicates one of its axes to the promotion of the development of digital skills with the aim of putting in place all the means to ensure that workers have an adequate mastery of digital skills and capabilities to understand and develop Artificial Intelligence technologies and applications. The Spanish government has planned a wide range of education and training policies whose basis is the National Digital Skills Plan, published in January 2021 and aligned with the Digital Agenda 2025.

This plan includes data analytics and artificial intelligence as cutting-edge technological areas within specialized digital skills, that is, "necessary to meet the labor demand for specialists in digital technologies: people who work directly in the design, implementation, operation and/or maintenance of digital systems".

In general, the national strategy presents policy actions on education and digital skills for the entire population throughout their lives. Although in many cases these measures are still in the planning phase and will see a major boost with the deployment of NextGenerationEU funds, we already have some pioneering examples such as the training and employment guidance programs for unemployed and young people tendered last year and recently awarded. In the case of training for unemployed people, actions such as the Actualízate program and the training project for the acquisition of skills for the digital economy are already underway. The actions awarded that are aimed at young people are scheduled to start in the first quarter of 2022. In both cases the objective is to provide free training actions aimed at the acquisition and improvement of ICT skills, personal skills and employability, in the field of transformation and the digital economy, as well as orientation and job placement. Among these ICT skills, those related to data and artificial intelligence will undoubtedly have an important weight in the training programs.

The role of universities

On the other hand, universities around the world, and of course Spanish universities, have been adapting curricula and creating new training programs related to data and artificial intelligence for some time now. The first to adapt to the demand was postgraduate training, which, within the higher education system, is the most flexible and quickest to implement. The first batch of professionals with specific training in data and artificial intelligence came from diverse disciplines. As a result, among the veterans of corporate data teams we can find different STEM disciplines, from mathematics and physics to virtually any engineering. In general, what these pioneers had in common was to have taken Masters in Big Data, data science, data analytics, etc. complemented with non-regulated training through MOOCs.

Currently, the first professionals who have completed the first degrees in data science or data engineering, which were reformed by the pioneering universities - but which are now already implemented in many Spanish universities - are beginning to reach the labor market. These professionals have a very high degree of adaptation to the current needs of the labor market, so they are in great demand among companies.

For universities, the main pending challenge is for university curricula in any discipline to include knowledge to work with data and to understand how data supports decision making. This will be vital to support the EU target of 70% of adults having basic digital skills by 2025.

Large technology companies developing talent

An idea of the size of the problem posed by the shortage of these skills for the global economy is the involvement of technology giants such as Google, Amazon or Microsoft in its solution. In recent years we have seen how practically all of them have launched large-scale free materials and programs to certify personnel in different areas of technology, because they see it as a threat to their own growth, even though they are not exactly the ones having the greatest difficulty in recruiting the scarce existing talent. Their vision is that if the rest of the companies are not able to keep up with the pace of digitalization this will cause their own growth to suffer and that is why they invest heavily in certification programs beyond their own technologies, such as Google's IT Support Professional Certificate or AWS's Specialized Program: Modern Application Development with Python.

Other multinational companies are addressing the talent shortage by retraining their employees in analytics and artificial intelligence skills. They are following different strategies to do this, such as incentivizing their employees to take MOOCs or creating tailored training plans with specialized providers in the education sector. In some cases, employees in non-data related roles are also encouraged to participate in data science training, such as data visualization or data analytics.

Although it will take time to see their effects due to the high inertia of all these measures, they are certainly going in the right direction to improve the competitiveness of companies that need to keep up with the high global pace of innovation surrounding artificial intelligence and everything related to data. For their part, professionals who know how to adapt to this demand will experience a sweet moment in the coming years and will be able to choose which projects to commit to without worrying about the difficulties that, unfortunately, affect employment in other areas of knowledge and sectors of activity.

Content prepared by Jose Luis Marín, Senior Consultant in Data, Strategy, Innovation & Digitalization.

The contents and views reflected in this publication are the sole responsibility of the author.