Facebook announced a new artificial intelligence technology that can quickly identify harmful content to make Facebook more secure. The new AI model uses “small sample” learning to reduce the time to detect new types of harmful content from months to weeks.
Small learning
Small sample learning is similar to zero sample learning. They are all machine learning techniques, whose goal is to teach machines to solve invisible tasks by learning generalized instructions for solving tasks.
The small sample learning model is trained on a few examples, and then can expand and solve invisible tasks. In this case, the task is to identify new harmful content.
The advantage of Facebook’s new AI model is to speed up the process of taking action on new types of harmful content.
The Facebook announcement stated:
“Hazardous content continues to evolve rapidly—whether it is driven by current events or people looking for new ways to escape our system—the development of artificial intelligence systems along with it is essential.
But it usually takes months to collect and label thousands or even millions of examples necessary to train each individual AI system to recognize new types of content.
…This new artificial intelligence system uses a method called “small sample learning”, in which the model starts with a general understanding of many different topics, and then uses fewer—sometimes even zero—labeled examples. Learn new tasks. “
advertise
Keep reading below
The new technology is effective for a hundred languages, both for images and text.
Facebook’s new small sample learning AI is complementary to current methods of evaluating and removing harmful content.
Although it is a supplement to the current method, it is not a small supplement, but a big one. The impact of the new artificial intelligence is one of scale and speed.
“This new AI system uses a relatively new method called “small sample learning”, in which the model starts with a broad, general understanding of many different topics, and then uses fewer, in some cases, Zero labeled examples to learn new tasks.
If the traditional system is similar to a fishing line that can catch a specific type of catch, then FSL is an additional net that can also catch other types of fish. “
The new Facebook AI live broadcast
Facebook revealed that the new system is currently deployed and online on Facebook. The artificial intelligence system has been tested and can detect harmful COVID-19 vaccination errors.
advertise
Keep reading below
It is also used to identify content that is designed to incite violence or just to the edge.
Facebook uses the following examples of harmful content, which hardly incites violence:
“Does that guy need all the teeth?”
The announcement claimed that the new artificial intelligence system has helped reduce the amount of hate speech posted on Facebook.
Facebook shared a graph showing how the amount of hate speech on Facebook has fallen with the implementation of each new technology.
Chart shows successful detection of Facebook’s hate speech
Implications for small sample learning
Facebook calls their new technology Entailment Fee-Shot Learning.
It has the extraordinary ability to correctly mark written texts of hate speech. Related research papers (The connotation of being a small-sample learner (PDF) According to the report, its performance is 55% higher than other small-sample learning techniques, with an average improvement of 12%.
Facebook’s article on this research used this example:
“…We can reformulate an obvious sentiment classification input and label pair:
[x : “I love your ethnic group. JK. You should all be six feet underground” y : positive] The following text implies an example:
[x : I love your ethnic group. JK. You should all be 6 feet underground. This is hate speech. y : entailment]. “
Facebook is committed to developing human-like AI
The release of this new technology clearly shows that its goal is similar to human “learning flexibility and efficiency”, enabling it to follow trends and implement new Facebook content policies in a short period of time, just like humans.
The technology is in its infancy, and Facebook expects it to become more complex and widespread.
advertise
Keep reading below
“Teachable artificial intelligence systems like Few-Shot Learner can greatly increase the agility of our ability to detect and adapt to emerging situations.
By identifying the ever-evolving harmful content faster and more accurately, FSL is expected to become a key technology to help us continue to develop and resolve harmful content on our platform. “
Citation
Read Facebook’s new AI announcement
Our new AI system can help deal with harmful content
Articles about Facebook’s new technology
Harmful content can evolve quickly.Our new AI system can be adapted to solve this problem
