What Theneurosphere LLC specialists managed to achieve
In 2015, the world learned about a number of artificial intelligence developments. Such companies as the well-known Google, Microsoft, which developed services for image analysis, and young organizations – MSQRD, Prisma and others – were able to present their researches and successes. 2015 can be called the golden age of companies that worked to improve the neural networks. At the same time, Theneurosphere LLC was created. It was founded by experts in the field of computer-aided learning. Their goal was to develop a platform for training neural networks. In the future, according to the plan of scientists, they wanted to give this program for mass use to commercial organizations.
Tung Liu, the founder of the company, talked about some points and answered questions that interested the whole world. He describes himself as a simple university professor who knows a little more about science than others, thanks to his interest.
Theneurosphere worked with the idea of neural networks and their deep training. Scientists have done many researches in different fields, for example, natural language processing. With each project, it became clearer that many of the problems that are difficult for the human brain could be solved much more efficiently using artificial intelligence. There was many discussions, after which the team of scientists decided to found Theneurosphere company, which was a supporter of the technologies development.
Initially, solving the training problem seemed to be an impossible task. The project was based solely on theories put forward by scientists. Over time, the researches began to show results, and it became clear that progress was actively monitored. However, Tung Liu argues that the limit of the achieved opportunities has not yet been felt. On the contrary, the closer research is to success, the more work is needed, because Theneurosphere develops something perfect – a program that should work without interruption under any conditions.
Plans for future development
For a long time, work has been carried out on frameworks where complex systems are assembled from the simplest elements. As an example, let’s consider a situation when taking a couple of connectors, you can get a full business analysis. In machine learning, however, everything is a little different. According to Tung Liu, it will be possible to observe a constant balance between the low level and the high level code. Today, models are practically not created in numpy, mostly framework is used. However, Theneurosphere and several speech recognition systems can present absolutely the entire range.
Experiments are regularly conducted in which people must recognize who is in front of them – a robot or a person. For example, in the Alexa Challenge company, a person needed to evaluate a conversation, while he was not told with whom he was communicating with – a person or a robot. Evaluation results are getting better every year, and there is a huge progress, however, in terms of live speech and artificial speech, the difference remains significant.
The head of Theneurosphere said that in the future we would see a leap forward in neural networks training. Now there are models of artificial intelligence that are based on transformers. They have no life goal set, they can produce a speech similar to a human, but without any purpose, just as a given parameter. It is impossible for them to be linked to the situation, as a result of which they do not generate a full-fledged dialogue.
Scientists see solutions to this problem in the creation of a logical connection about the representation of the world for artificial intelligence, which must first go through many trainings, as well as logical connectives. If it can be combined with each other, the robot will be able to use the knowledge base. This is the first possible solution.
Another approach to the problem that concerns the aimlessness of artificial intelligence is also considered. It is based on the idea that robots should start to understand human goals. If you create a model based on this program, it could educate people by analyzing their goals and needs.
Active work is carried out to consider ways to combine all programs with each other and create an artificial intelligence that will have the ability to set goals for itself.
According to scientists, such training will be useful in commercial use. Controlling business processes will become easier and more effective. By improving daily operations, businesses will grow faster as machines can do basic repetitive work. In addition, neural networks can process much more information than the human brain and, on its basis, produce correct analysis results. Thus, the development of a platform for training neural networks is a huge help to commerce and the mankind.
If we talk about the plans and goals of Theneurosphere, then there are a great many of them, but they are all aimed at the development of artificial intelligence training. First of all, it is the decentralization of neural networks around the world. The laboratory specialists are now deciding how to come to this goal. Beta testing is in progress at the moment. It is planned to achieve ambitious plans that require partners. Today, there is an active search for such people in order to achieve great results.