Robot Xiaoice makes 40 lines of poems in 4 seconds

Rock-like love engraved on the island of dreamland

 

Watering time seeds

 

Delight in the tender universe

 

A distant star shines with hope

 

I will thank the drift of life

 

When the time rushes back, the laughter is sweet

 

You are the life of my life

 

—Microsoft AI Robot Xiaobing "To Ten Years Later"

 

Xiao Xiaobing, female, artificial intelligence chatbot, born in May 2014. Although not yet 5 years old, it has already passed 5 iterations and has been extended to Japan, the United States and other places.

 

Xiao Xiaobing's team trained Xiaobing to write poems and went through two stages. First: imitate humans through technology, create appearances, and make the results close to humans, and conduct millions of self-learning and training. Second: Learn the connotation and creation of poetry, that is, the poetic language used, so that readers can form a more vivid, vivid or profound mood in the reading process.

 

What I can be sure of is that it took less than 4 seconds for Xiao Bing to formally form the long line of the 40-line "To Ten Years Later" poem. And, through continuous adjustments and tests, this "draft" speed has stabilized. There is probably no poet in the world, which can be compared with Xiaobing in this regard. It can be seen that the development of artificial intelligence is in full swing.

 

 

John McCarty, the father of artificial intelligence ▲

 

As early as 1956, McCarthy convened researchers from Harvard University, MIT, IBM, and Bell Labs to convene a Dartmouth conference and formally proposed "artificial intelligence".

 

China, “Made in China 2025” was proposed in May 2015, and “Three-year Action Plan for“ Internet + ”Artificial Intelligence” was proposed in May 2016.

 

Currently, the wave of technological innovation represented by Internet +, AR (augmented reality) / VR (virtual reality), Big Data (big data), and AI (artificial intelligence) / deep learning (deep learning) has emerged.

 

In traditional non-destructive testing, the detection accuracy is highly dependent on the judgment level of the operator, which adversely affects the objectivity and consistency of the test results, and also increases the burden on the operator.

 

 

For this reason, Shengtuo Detection Research has developed an auxiliary judgment method based on AI (machine learning) to improve the detection accuracy and reduce the difficulty of operation.

 

 

For AI-assisted detection and judgment, the upgrade detection uses Bayesian networks, a set of secondary learning models that integrate Bayesian, neuron networks, and aggregated neuron networks. The prediction results of each model are weighted according to the accuracy of the training model. The comprehensive result is used as the result of the AI ​​prediction.

 

 

Bayesian Network Model ▲

 

 

Neural Network Model ▲

 

At present, Shengtuo testing has been carried out in the field of sleeve grouting of prefabricated structures, grouting of prestressed bridge tunnels, etc. After a large number of model tests and field verifications, the accuracy of comprehensive recognition accuracy can reach 80% under the conditions covered by training data -85%.

 

The test will continue to gather the amount of testing and monitoring data, and conduct training and identification, to further improve the test accuracy, reduce errors caused by human factors, and provide technical support to ensure the quality of national infrastructure construction.