China
Africa
Explore chapters and articles related to this topic
Implementation
Published in Seyedeh Leili Mirtaheri, Reza Shahbazian, Machine Learning Theory to Applications, 2022
Seyedeh Leili Mirtaheri, Reza Shahbazian
DL takes advantage of using a form of specialized hardware extant in accelerated computing environments. The current conventional regnant solution (NVIDIA) uses Graphics Processing Units (GPUs) as general-purpose processors. GPUs supply immense parallelism for large-scale DL problems, making it possible to scale algorithms vertically to huge amounts of data that are not traditionally computable [101]. GPUs are advantageous solutions for real-world and real-time systems demanding expeditious decision-making and learning (especially in image processing). Field Programmable Gate Array (FPGA) [104] and the lately made public Google Tensor Processing Unit 3.0 (TPU) are alternatives to GPUs. Other IT companies are also proceeding to offer dedicated hardware for DL acceleration, e.g., Kalray with their second generation of DL acceleration device named MPAA2–256 Bostan, focused on mobile devices like autonomous cars.
Machine Learning Applications In Medical Image Processing
Published in Sanjay Saxena, Sudip Paul, High-Performance Medical Image Processing, 2022
Tanmay Nath, Martin A. Lindquist
In subsequent years, three major branches of ML emerged: symbolic learning; statistical methods; and NNs [10]. Symbolic learning algorithms learn new concepts by constructing a symbolic representation of objects in a class. For example, a symbolic learning system will construct a logic to distinguish between two objects based on attributes like size and color. Examples of symbolic learning systems include decision trees (DTs) [11] and inductive logic program [12]. Statistical methods, or pattern recognition methods, draw inspiration from statistics where one can use descriptive statistical methods to transform raw data and help a machine learn hidden patterns. Some of the examples include Naive Bayesian classifiers [13], K-nearest neighbors (KNNs), and discriminant analysis. Finally, NNs, described above, have advanced significantly since its inception as a perceptron. By adding multiple layers to the networks (e.g., multilayer feed-forward network with back-propagation learning), NNs are currently widely used for medical imaging applications. More-over, with the advent of new technology and computational resources like graphical processing units (GPUs), Google tensor processing unit (TPU), cloud computing, a plethora of NNs have been developed, including AlexNet [14], ResNet [15], and U-Net [16]. See Table 5.1 for examples of common convolutional neural network architectures.
Artificial Intelligence Software and Hardware Platforms
Published in Mazin Gilbert, Artificial Intelligence for Autonomous Networks, 2018
Rajesh Gadiyar, Tong Zhang, Ananth Sankaranarayanan
Recent announcements on custom AI hardware (e.g., Google tensor processing unit [TPU] and Intel neural network processor [NNP]) have raised awareness that custom solutions might leapfrog the performance of both CPUs and GPUs for some neural network applications. However, users may want to consider the types of network models that will be trained when they make the hardware choice. For example, benefits of special-purpose hardware that performs tensor operations at reduced precision apply to only few types of neural architectures like CNNs. It is important to understand if your work requires the use of those specific types of neural architectures.
Machine learning solutions in the management of a contemporary business organisation
Published in Journal of Decision Systems, 2020
Nowadays, contemporary business organisations put into practice multiple machine (ML) learning and artificial intelligence solutions. Machine learning and also reinforcement learning with such frameworks as Cafe, Google tensor flow might play a crucial role in the improvement of the functionality of enterprise, and it perfects and increase the business value of the entire decision-making process. What is more such innovative solutions have a positive impact on the optimisation of business processes and contribute to leveraging competitive advantage of a particular enterprise in the global market. The vital issue is the quality of input data, especially big data sets which may constitute the basis for model training. What is more, the hybrid solutions combine machine learning with data analytics such as Business Intelligence systems including managerial dashboards, increases the efficiency of the whole solution.