Based on a large amount of real data, Baidu’s PaddlePaddle division, a deep learning framework, teamed up with Kesai to launch a series of questions based on cutting-edge real issues. The first one was an INTELLIGENT video clip AI competition focusing on variety shows.

The event ended last month. The competition uses Corsai’s own online data analysis tool k-Lab, where PaddlePaddle deep learning framework can be directly invoked and BROAD datasets can be mounted online. K-lab will be mounted on Baidu Cloud CPU/GPU, and contestants can directly submit the model results to the evaluation system to get model scores. By the end of the competition, a total of 281 teams with a total of 496 people registered, a total of 909 submissions.

This on-line NLP intelligent question and answer contest is to challenge the difficult problem in the field of artificial intelligence — reading comprehension.

At present, there are many classic questions about NLP in the world, such as The SQuAD Challenge sponsored by Stanford University, MS Marco Challenge sponsored by Microsoft, Google DeepMind Open Data Test set for Reading Comprehension, Facebook Open Data test set for Reading comprehension. These are mostly English data sets, and are partly based on SPANof Words or word label classification. However, it is much more difficult to achieve true reading comprehension through machine learning than accurate word meaning recognition and search result invocation. On the one hand, it needs to analyze and understand the meaning in all aspects through algorithms, and on the other hand, it also needs to introduce high-quality data sets to cooperate with training.

BROAD contains the largest Chinese open domain reading comprehension dataset to date, DuReader. The data set is based on real application requirements. All questions are from real questions of Baidu search users. The documents are from real web documents sampled from the whole network and UGC documents known by Baidu.

PaddlePaddle AI series – Intelligent Quiz competition will provide contestants with CPU and GPU computing resources on Baidu Cloud. Contestants need to build models based on text and questions and output correct answers based on data, testing their ability to summarize and revise models. By providing model training for ai to answer practical questions, there may be applications in the future that can provide users with a full set of solutions without opening web pages or manually screening answers, which can replace most AI assistants on the market and save a lot of time.

The tournament is based on PaddlePaddle, a Deep learning platform on Baidu. PaddlePaddle, a massively parallel distributed deep learning framework, is currently the most popular open source deep learning platform in the world (measured by Github Pull Requests). It has integrated a variety of neural networks and deep learning algorithms such as CNN and RNN, and supports many kinds of hardware such as CPU, GPU and FPGA. It has cooperated with Kubernetes in PaddlePaddle EDL elastic deep learning, becoming the world’s first open source AI cloud solution supporting elastic job scheduling. PaddlePaddle is easy to learn, efficient and flexible, supporting business needs in multiple areas such as massive image recognition and classification, machine translation and autonomous driving.

At last year’s Baidu World Conference, PaddlePaddle announced three new features that further enhance ease of use and lower the bar for developers:

  1. Paddlefluid provides control flow structures such as while and IF in high-level languages to improve users’ development efficiency and ensure computational performance by using compiler optimization technology.

  2. PaddlePaddleCloud allows users to create AI applications in a browser and debug them in the cloud, eliminating the need for developers to switch between PCS and computers, increasing productivity.

  3. PaddlePaddleEDL is the world’s first open source AI cloud solution that supports flexible job scheduling through collaboration with Kubernetes.

K-lab, an online data analysis platform, provided full support for this event. K-lab not only encapsulates nearly 100 common AI development tools including PaddlePaddle, but also can directly call Baidu cloud computing power.

NLP Intelligent Question and Answer contest is launched today. We welcome and look forward to your participation in the field of reading comprehension for researchers and university students. Click to read the original text to register and join us to challenge NLP- Artificial general intelligence!

See here may wish to pay attention to the small division look forward to more challenges

Kesci.com is an online community for data talent and industry issues. The K-Lab online data analysis and collaboration platform created by KOSai brings new experience to data workers’ study and work.