Artificial intelligence technologies are actively developing all over the world. Machine learning and neural networks are finding new uses. Promising cases and solutions appear every year. Let’s talk about three of them, which became a clear trend in 2020, although they appeared just a little earlier.
Recent advances in artificial intelligence have allowed many companies to develop algorithms and tools to automatically create artificial 3D and 2D images. This generative (“creative”) AI allows computers to create texts, audio files, images and other content. In a review by MIT Technology magazine, generative AI is named one of the most promising advances in the world of artificial intelligence over the past decade. According to Cnews, it can already be used in the next generation of applications for automatic programming, content development, creating samples of fine art and other creative, design and engineering activities. For example, NVIDIA has developed software that can create photorealistic faces with only a few photographs of real people. Generative artificial intelligence can also aid healthcare by creating prostheses, organic molecules, and other items from scratch when activated using 3D printing, CRISPR and other technologies.
This approach assumes the distribution of the work of artificial intelligence. Instead of centralized collection and storage of information in one place, the learning process takes place directly on remote points (user devices or local servers) for subsequent work with new algorithms. This approach, first of all, eliminates the need to move large amounts of data to a central server for machine learning tasks and, accordingly, removes the acute issue of data privacy, notes NIX Solutions. The latter factor is important, for example, for medical organizations. Intel recently teamed up with the School of Medicine of University of Pennsylvania to deploy a federated learning network across 29 international medical and research institutions. The aim of the project is to improve the efficiency of diagnostics in detecting brain tumors. The research team published their findings on federated learning and its use in healthcare at the Supercomputing 2020 conference. Published data show that federated machine learning has achieved 99% accuracy in tumor identification.
The key disadvantage of any neural network is that it requires large computational resources and memory, which makes it difficult to deploy in embedded systems with limited hardware resources. The technology of neural network data compression is designed to solve this problem, for which methods such filters as reduction and sharing of parameters, quantization, low-rank factorization, portable or compact convolutional are used.
For example, NVIDIA recently developed a new type of video compression technology that replaces the traditional video codec with a neural network and can dramatically reduce the bandwidth required for signal transmission. According to the developers, in comparison with the widely used H.264 codec, the technology based on neural network data compression provides a tenfold increase in efficiency.