Google AI has announced WIT: a data series linking text images and context from Wikipedia

The system that includes 11 million images and their contexts in 108 languages ​​is intended for artificial intelligence training. The data series is available to the public and Google will also hold a WIT-based application competition together with the Wikimedia Foundation and the KEGGLE website

דוגמה לניתוח תמונה והקשר מויקיפדיה עבור פרויקט WIT של גוגל AI. צילום יחצ
An example of image analysis and context from Wikipedia for the Google AI WIT project. Press photo

Google is today celebrating its 23rd anniversary. Google AI, one of the company’s junior divisions, announced WIT: a data series linking text images and the Wikipedia context open to the general public for artificial intelligence training.

Research, Google Research has published the details of Google AI’s announcement of WIT – a huge series of images from Wikipedia and their adaptation to text in many languages ​​- for artificial intelligence training.

In their blog on the Google AI site

” Traditionally, these data sets were created by manually adding captions to images, or scanning the web and extracting the alternative text as captions for images. While the previous approach allows Higher quality data, the intensive manual interpretation process limits the amount of data that can be generated. Kim. Another shortcoming of existing data sets is the lack of coverage in non-English languages. The speaker naturally led us to ask: Is it possible to overcome these limitations and create a high-quality, large and multilingual data set with a variety of contents? “

” Today we present the data set Of Wikipedia-based texts and images (WIT), created by extracting multiple texts in image descriptions from Wikipedia articles and image links in Wikipedia. We conducted a rigorous screening that ensured that only high-quality text-image kits would be scanned. As outlined in “WIT: Wikipedia-Based Image Text Data Kit for Multilingual Multilingual Multilingual Machine Learning” presented at SIGIR ’21, the result was a repository of 37.5 million rich text and image examples including 11.5 million Unique images and their descriptions in 108 languages. The WIT dataset is available for download and use under a Creative Commons license. “

The unique advantages of the WIT dataset are:

  • Size: WIT is the largest multi-modal data set of text-examples Image available to the public. Multilingual: 108 languages WIT has 10 languages ​​more than any other data set. Contextual information: Unlike typical multimodal data systems, which have only one caption per image, WIT includes information that includes page-level and section-level relationships. Entities in the real world: Wikipedia, being a broad knowledge base, is rich in real-world entities represented in WIT.

  • Challenging test set: In our recent work received at EMNLP, all the latest models demonstrated Significantly lower performance in WIT compared to traditional evaluation sets (e.g., a decrease of about 30 points in memory). A quality training kit and Challenging Evaluation Index The extensive coverage of diverse concepts in Wikipedia means that WIT’s evaluation systems serve as a challenging criterion, Even for modern models. We found that The average recovery scores

  • Note: This article has been indexed to our site. We do not claim legitimacy, ownership or copyright of any of the content above. To see the article at original source Click Here

    Related Posts
    How to Search for Life as We Don't Know It thumbnail

    How to Search for Life as We Don’t Know It

    In my freshman seminar at Harvard last semester, I mentioned that the nearest star to the sun, Proxima Centauri, emits mostly infrared radiation and has a planet, Proxima b, in the habitable zone around it. As a challenge to the students, I asked: “Suppose there are creatures crawling on the surface of Proxima b? What…
    Read More
    The Elegant Math of Machine Learning thumbnail

    The Elegant Math of Machine Learning

    1 Machines Can Learn!A few years ago, I decided I needed to learn how to code simple machine learning algorithms. I had been writing about machine learning as a journalist, and I wanted to understand the nuts and bolts. (My background as a software engineer came in handy.) One of my first projects was to
    Read More
    Index Of News
    Total
    0
    Share