New technology improves the accuracy of AI models

< IMG SRC = "/Uploads/Blogs/42/C7/IB-FQT3M1V1V1ver_A2A6BBB1C.jpg" Alt = "New Technology improves the accuracy of artificial intelligence models"/> ~ ~ ~ < P > North Carolina's University Researchers have developed an innovative method of detecting and eliminating false correlations in artificial intelligence models without prior knowledge of problematic features that opens a new direction in increasing accuracy and reliability < P > False correlations are one of the main problems of modern artificial intelligence. This phenomenon occurs when the AI ​​systems are defined and relied on & ldquo; signs that are not provoked; rsquo; listed with the task for which they are intended & rdquo;. & Amp; nbsp; < strong > ~ > > > > solutions. & nbsp; a problem is exacerbated by & ldquo; prejudice of simplicity & rdquo;, when the model prefers simplified characteristics instead of more complex.

~ ~ ~ ~ > < p >A classic example of false correlation is a model learning to recognize dogs. If there are many images of dogs with collars in the training set, the AI ​​can identify the collar as the main sign of the dog. This model can mistakenly classify cats with collars as dogs. These inaccuracies undermine trust in artificial intelligence systems.

< H2 > Restriction of traditional approaches

< P > Traditional methods of combating false correlations require expert intervention. Specialists should identify the problematic features and then make changes to the training data. For example, more dogs without collars can be added to solve the problem with collars. Thus AIs will learn to recognize animals regardless of the presence of this accessory.

< p >However, the key restriction of this approach is the need for preliminary knowledge of false features. In practice, it is often impossible to determine such signs in advance. This is especially true of complex data sets with numerous variables. This problem prompted scientists to seek alternative solutions.

< p > Mulchandani Varun and his colleagues from North Carolina University proposed an elegant solution. & Amp; nbsp; < strng > their methodology involves the removal of a small part of the most complex specimens from the study. found that these samples are often the most ambiguous and noisy, forcing the model to use the wrong features.

< h2 > Innovative data trimming method

< P > The essence of the new method is strategic trimming of educational data. Having identified the most difficult samples for classification, researchers remove them from a data set. This process reduces the likelihood that AIs will absorb false correlation. What is important is the & ldquo; pruning & Rdquo; does not have a significant negative impact on the overall efficiency of the model.

< p > The results of the study showed the impressive efficiency of a new approach. & Amp; nbsp; < Em > & ldquo; cutting of false correlations by cutting data & pdquo; < 62; The name of the reviewed work, which will be presented at the International Conference on Educational Representations (ICLR). The proposed method has reached the latest results by surpassing previous approaches to improving models.

< P > It is important to note that this method does not require deep knowledge of specific problematic features. Instead, it focuses on the general characteristics of complex samples. & Amp; nbsp; < strng > This makes the approach universal and applied to different domains. < h2 > Practical value and prospects

< p >The consequences of this study go far beyond academic discussions. It has a direct impact on the development of reliable AI systems for critical industries. As you integrate the technologies of the Health & Rsquo; I, Finance and other areas, the need for methods of increasing the reliability of models becomes more urgent.

< P > The proposed technique provides AII practitioners with a powerful tool for improving models without the need for deep analysis of data features. This is especially valuable in situations with limited resources or time.

& ldquo; focused on removing confusing and complex data, researchers can pave the way to transformational changes in the research community of the SI & RDQUO ;, & ndash; The industry experts say. < p >The development of methods for combating false correlations becomes key for ethical implementation of AI. Modifications based on incorrect associations can enhance prejudice and discriminate. Increasing the accuracy of models, the new method contributes to the development of fair and reliable systems of AI.

< h2 > The future development of the technologies Shi

< p > This study reflects the desire of the scientific community for the continuous improvement of the technologies of the AI. & Amp; nbsp; < strong > Effective methods of detecting and eliminating false correlations is an important step to creating artificial intelligence; AIs become unintended & Rsquo; a capacious part of everyday life, their reliability becomes critical.

< p > North Carolina's study of the University of North Carolina offers a practical solution for a difficult problem. It demonstrates that even without a detailed understanding of specific false features, the quality of AI models can be significantly improved. This approach opens a new direction in the development and training of artificial intelligence models.

< P > In the future, similar methods can be standard practice in the development of AI systems. Combining strategic data trimming with other techniques, researchers and developers will be able to create models that truly understand the world, not rely on surface associations. & Amp; nbsp; < Em > & ldquo; the ability to recognize and pitching; A powerful tool for increasing the reliability of models & rdquo; , & ndash; Emphasize the authors of the study.