blog posts

If You Are Going To Follow AI News, It Is A Good Idea To Have An English Dictionary With Several Etymological Dictionaries By Your Side So That You Can Better Understand The Meaning Of The Sentences You Are Reading, As Different Paradigms Of Deep AI Learning Become More And More Common.

They use common and specialized words in various ways, some of which are sometimes misleading. In particular, algorithms have learned to think before answering questions. Google has developed an intelligent algorithm called PonderNet, an intelligent neural network that has learned to think lightly before answering questions.

Diamond recently unveiled a new algorithm called PonderNet that allows neural networks to learn to think before answering questions and then answer them. This approach improves the ability of neural networks to answer questions and allows them to answer tough questions with more confidence than ever before.

Developed by computer scientists Andrea Bannino, Ian Balaguer, and Charles Blondel, the program aims to integrate neural networks further and optimize computer-aided work.

This intelligent algorithm focuses entirely on computing efficiency and increasing computational accuracy

. As its design describes, PonderNet is a new algorithm that learns to balance the complexity of problems with the number of calculations it must perform.

In connection with this intelligent algorithm, there is an important issue. The time required to solve problems and answer questions does not depend only on the dimensions of the inputs, But complexity is also important.

In addition, the problem with most neural networks today is that the amount of computation used in standard neural networks is not proportional or complex but proportional to the size of the network. To solve this problem, DePayand, in its latest research and its output published in the PonderNet algorithm, can work based on adaptive computation time and other characteristics of adaptive networks.

PonderNet is highly variable and can be used with low standard deviation gradient estimation instead of stochastic. This is achieved by resetting the stop creep as a possible model. Based on this definition, we see that PonderNet is a new algorithm in this field that can adapt the computational complexity to the operational power it needs. To be more precise, it uses resources in proportion to the calculations it needs.

How does the above algorithm work?

As DePaymand points out, if PonderNet realizes that it does not require heavy processing to sense a problem, it will try to answer the problem with the least amount of computation if it feels that heavy computing needs a tough challenge to answer more accurately. Tries to provide the best answer.

Bonino et al. Have developed this algorithm based on research conducted by researchers in recent years on topics such as conditional calculation. However, in this project, Alex Graves, one of Gole’s colleagues, should not overlook the direct and influential role. Graves has a brilliant track record in strategic research related to the design of neural networks and their better interaction with computers. For example, a few years ago, he and his colleagues designed a neural network based on the “neural Turing machine” theory.

As for the PonderNet algorithm, Banino and his team worked on a project that Graves was working on in 2016, called Compatible Computing Time. The project was based on the fact that in the realm of human reasoning, in most cases, there is no symmetry between problem and problem-solving. For example, it may sometimes take minimal effort to articulate a problem, but it may take a long time to discover it. For example, adding two numbers is much easier than dividing them, although the symbol looks identical.

Graves provided a solution for the computer to calculate how long it would take to think about the problem, given its complexity. Sometimes several layers of a neural network are enough to do the calculation. “A neural network program is a transformative machine because it automatically finds a way to convert input into output,” says Graves. “The number of layers of artificial nerve cells through which information must pass to become an accurate output successfully is one way to measure the effort to perform the calculation.”

last word

In standard neural networks, the number of calculations is related to the dimensions of the inputs, and to be more precise. No attempt is made to discover the complexity of the problems what Google’s PonderNet algorithm combines the amount of computing required with the complexity of the problem. PonderNet performs several computational steps to achieve the most effective solution based on previous training and knowledge gained.