Tag Archives: deep learning

AI learns to generate images from text and begins to better understand our world

Reading Time: 2 minutes


OpenAI, co-founded by Elon Musk, has created the world’s most stunning AI model to date. GPT-3 (Generative Pre-trained Transformer 3) without any special prompts, can compose poems, short stories and songs, making one think that these are the work of a real person. But eloquence is just a gimmick, not to be confused with a human understanding of the environment. But what if the same technologies were trained simultaneously on text and images?

Researchers from the Paul Allen Institute for Artificial Intelligence have created a special, visual-linguistic model. It works with text and images and can generate pictures from text. The pictures look disturbing and strange, not at all like the hyperrealistic “deepfakes” created by generative adversarial networks (GANs). However, this capability has long been an important missing piece.

The aim of the study was to reveal whether neural networks can understand the visual world as humans.  For example a child who has learned a word for an object can not only name it, but also draw the object according to the hint, even if the object itself is absent from his point of view. So the AI2 project team suggested the models do the same: generate images from captions.

The final images created by the model are not entirely realistic upon close inspection. But it is not important. They contain the correct high-level visual concepts. AI simply draws the way a person who cannot draw would draw on paper.

This makes sense: converting text to an image is more difficult than doing the opposite.

“A caption doesn’t specify everything contained in an image,” says Ani Kembhavi, AI2’s computer vision team leader.

Creating an image from text is simply a transformation from smaller to larger. And it’s hard enough for the human mind, apart from programs.  If a model is asked to draw a “giraffe walking along a road,” then it needs to conclude that the road will be gray rather than bright pink, and will pass next to a field rather than the sea. Although all this is not obvious to AI.

Sample images generated by the AI2 model from captions. Source: AI2

This stage of the research shows that neural networks are capable of creating abstractions – a fundamental skill for understanding our world.

In the future, this technology will allow robots to see our world as well as humans, which will open up a huge scope of possibilities. The better the robot understands the environment and uses language to communicate, the more complex tasks it will be able to perform. In the current perspective, programmers can better understand the aspects of machine learning

“Image generation has really been a missing puzzle piece, By enabling this, we can make the model learn better representations to represent the world.”





Tagged , , ,

DeepL – a translator which surpassed Google Translate

Reading Time: 4 minutes

A company doesn’t have to be a technological giant to create a product that exceeds the most popular programs of the same type. There is no doubt that in the world of automatic translation Google, Microsoft, and Facebook are the leaders. And yet it turns out that a small company DeepL has created a translator that sometimes exceeds the quality of the most popular programs of this type.

DeepL logo
Source: https://www.deepl.com/home


How DeepL was created?

It turns out that the key to the development of the translation service was to use the first own product, which is Linguee, a translation search engine on the Internet. The data obtained in this way became training material for artificial intelligence behind DeepL.

Interestingly, Linguee’s co-founder, Gereon Frahling, once worked for Google Research but left in 2007 to continue his new venture.

Currently, DeepL supports 42 language combinations between Polish, English, German, French, Spanish, Italian and Dutch. Already now, artificial intelligence is learning more, such as Mandarin, Japanese and Russian. There are plans to introduce an API, by means of which it will be possible to develop new products and implement the mechanism in other services.

The team has been working with machine learning for years, for tasks bordering on basic translation, but finally, they began a fervent work on a completely new system and a company called DeepL.


What is the advantage of DeepL?

Once again, people realized that AI is learning all the time – to the benefit of consumers, of course. The artificial intelligence behind the DeepL not only accurately recognizes words and selects translations, but is also able to understand certain linguistic nuances, perfectly copes with changed sentence patterns, which makes the result of a user’s inquiry extremely natural – as if it was written by a human being.

The company also has its own supercomputer, which is located in Iceland and operates at 5.1 petaflops. According to press releases with such equipment DeepL is ranked 23rd in the Top 500 supercomputers worldwide.


The statistics do not lie

The blind test compared the new product and solutions from Google, Facebook, and Microsoft. Professional translators were supposed to choose the best results of the mechanisms in the comparison without knowing the author of the translations:

DeepL’s blind testing results
Source: https://techcrunch.com/2017/08/29/deepl-schools-other-online-translators-with-clever-machine-learning/


But that’s not all, because in the BLEU results DeepL also gets great scores. BLEU is an algorithm for evaluating the quality of translation.


Why do others recommend DeepL instead of Google Translate?

The main advantage of DeepL in the context of Google Translate is much better knowledge (or rather a detection) of idioms, phrases, and phraseological compounds. Where, for example, Google Translate is weakening and literal meaning is being found, DeepL can surprisingly offer a more nuanced and much more specific language solution. The translation is not a literal translation of the text, but one that best harmonizes with the contexts and connotations characteristic of the words.

The passage from a German news article rendered by DeepL
Source: https://techcrunch.com/2017/08/29/deepl-schools-other-online-translators-with-clever-machine-learning/

The passage from a German news article rendered by Google Translate
Source: https://techcrunch.com/2017/08/29/deepl-schools-other-online-translators-with-clever-machine-learning/


No wonder that DeepL is gaining recognition all over the world. Here are some reviews:

Thanks to more French-sounding phrases DeepL has also surpassed other services.Le Monde, France

In the first test, from English to Italian, it was very accurate. In particular, he understood the meaning of the sentence well, instead of being stunned by the literal translation.La Repubblica, Italy

DeepL from Germany surpasses Google Translate. A short WIRED test shows that the results of DeepL are by no means worse than those of its best competitors, and in many cases even surpass them. Translated texts are often much more fluid; where Google Translate creates completely meaningless word strings, DeepL can at least guess the connection.WIRED.de, Germany

We were impressed with how artificial intelligence selects the translations and how the results of its work look afterward. Personally, I had the impression that on the other side sits a man who on speed translates.Antyweb, Poland


The DeepL tool has been made available to a wider audience – for free in the form of a website.

Now it is only a matter of waiting for DeepL to advertise its tool, because although it does not have a large language base, at first glance the accuracy of the translations definitely exceeds the most popular tools of this type.

It’s worth watching how the product will develop further as the current achievements of DeepL are really promising.

Did any of you choose DeepL instead of Google Translate?



[1] https://techcrunch.com/2017/08/29/deepl-schools-other-online-translators-with-clever-machine-learning/

[2] https://www.deepl.com/blog/20180305.html

[3] https://www.dw.com/en/deepl-cologne-based-startup-outperforms-google-translate/a-46581948

[4] https://www.forbes.com/sites/samanthabaker1/2019/06/27/will-this-german-startup-win-the-translation-game/

[5] https://www.deutsche-startups.de/2018/07/05/deepl-koelner-uebersetzungskoenig-macht-millionengewinn/

[6] https://www.forbesdach.com/artikel/davids-erbe-und-igels-strategie.html

[7] https://www.letemps.ch/societe/deepl-meilleur-traducteur-automatique

Tagged , , , , , , ,

Duolingo – the best machine learning startup to work for in 2020

Reading Time: 3 minutes

Duolingo is starting the year off strong. They have been named one of the top startups to work for, in the growing field of machine learning. These and many other insights are from a Crunchbase Pro analysis completed using Glassdoor data to rank the best machine learning startups to work for in 2020. Why is Duolingo a unique company?

Duolingo logo
Source: https://www.duolingo.com/

Duolingo AI Research

Duolingo AI Research is one of Duolingo’s fastest-growing teams. They are using real-world data to develop new hypotheses about language and learning, test them empirically, and ship products based on their research. Duolingo has revolutionized language learning for more than 300 million people around the world. They keep on bringing creative, interdisciplinary ideas on how to deliver a high-quality education to anyone, anywhere, through AI.


Duolingo AI team logo
Source: https://research.duolingo.com/


Tools and data from Duolingo

Duolingo use AI to adapt longer learning content to learners’ level. The startup is regularly releasing their internal tools to the public so everyone can read more about their research innovations. One of them is CEFR Checker. This tool determines whether texts are appropriate for beginner, intermediate, or advanced learners of English or Spanish. It works by analyzing vocabulary and highlighting words by their reading proficiency level according to the Common European Framework of Reference (CEFR). Duolingo uses interactive tools like this one to help people revise content (e.g., Podcasts and Stories) for particular levels.

The Duolingo CEFR Checker: an AI tool for adapting learning content
Source: https://making.duolingo.com/the-duolingo-cefr-checker-an-ai-tool-for-adapting-learning-content

The Duolingo CEFR Checker: an AI tool for adapting learning content
Source: https://making.duolingo.com/the-duolingo-cefr-checker-an-ai-tool-for-adapting-learning-content


Duolingo is also committed to sharing data and findings with the broader research community. SLAM Shared Task is an example project. It contains data for the 2018 Shared Task on Second Language Acquisition Modeling (SLAM). This corpus contains 7 million words produced by learners of English, Spanish, and French. It includes user demographics, morph-syntactic metadata, response times, and longitudinal errors for 6k+ users over 30 days.


Why people should consider working at Duolingo?

The language-learning app Duolingo is valued at $1.5 billion after a $30 million investment by Alphabet’s CapitalG. Bookings growth has risen from $1 million to $100 million in less than three years for the most downloaded and top-grossing education app worldwide. What is more, Pittsburgh’s first venture capital-funded $1 billion start-up plans to increase staff by 50% with the new funding. Duolingo has been adding user and revenue at an impressive pace, continuing to solidify its position as the No. 1 way to learn a language globally.


Why people should consider working in the machine learning field?

Demand reminds high for technical professionals with machine learning expertise. According to Indeed, Machine Learning Engineer job openings grew 344% between 2015 to 2018 and have an average base salary of $146,085 according to their Best Jobs In The U.S. Study.

It can be safely stated that Duolingo is developing very dynamically. There is also no doubt that the rapid growth of a startup also means the development of its employees.

Would you choose to join Pittsburgh’s unicorn if you had such a chance? What do you think about Duolingo’s contribution to the development of the education sector?



[1] https://www.forbes.com/sites/louiscolumbus/2020/12/29/the-best-machine-learning-startups-to-work-for-in-2020-based-on-glassdoor/#71505e744886

[2] http://blog.indeed.com/2019/03/14/best-jobs-2019/

[3] https://www.cnbc.com/2019/12/03/google-funded-duolingo-first-1-billion-start-up-from-pittsburgh.html

[4] https://making.duolingo.com/the-duolingo-cefr-checker-an-ai-tool-for-adapting-learning-content

[5] https://making.duolingo.com/how-machine-learning-helps-duolingo-prioritize-course-improvements

[6] https://cefr.duolingo.com/

[7] https://research.duolingo.com/

Tagged , , , , ,