Issue link: https://resources.mouser.com/i/1437726
33 INTEL 2021 appropriate ML/DL model best suited for the data to be analyzed. Types of ML models that may be considered include NNs, decision trees, support vector machines, regression analysis, genetic algorithms, Bayesian networks, and federated learning. NNs are amongst the most common model types and are central to DL algorithms. In addition to the frameworks, many of the Intel-optimized algorithms are offered as libraries for many popular languages such as Numpy in Python and Cart for R. 3. Toolkits for Application Developers: At the highest level of the developer stack is toolkits that wrap up all the underlying mathematical/analytics functionality and ML/DL frameworks into simple to use software development kits that allow the developer to focus on building solutions vice fundamental technologies. Analytics Zoo allows developers to build AI applications atop the Apache Spark analytics engine with built-in deep learning pre-trained models for object detection, image classification, text classification, and recommendation use cases. Analytics Zoo supports popular ML/DL frameworks and libraries such as TensorFlow, Keras, and BigDL. Furthermore, it helps in scaling models in order to handle training and inferencing at scale. OpenVINO is a deep learning toolkit intended to target convolutional neural network-based models to Intel edge computing hardware such as the Movidius Neural Compute Stick (NCS). Though initially intended for applications that are heavy into real-time machine vision, OpenVINO also supports audio and language recognition workloads across a wide range of Intel hardware. AI Analytics Toolkit is built specifically for those familiar with Python and provides tools to accelerate deep neural networks training on CPUs. It can also integrate trained models into Python-based inferencing applications by leveraging Intel hardware optimized implementations of TensorFlow and PyTorch. Lastly, it is incorporated into many popular Python packages (Pandas, NumPy, SciPy, Scikit-learn, and XGboost) to help expedite Python-based data analytics applications. DAAL with those tools and libraries. Using Intel hardware and these libraries within higher-level code frameworks such as TensorFlow and PyTorch allows ML applications to run at peak performance. nGraph: An open-source C++ library and runtime/ compiler suite permits data scientists to use their preferred deep learning framework (e.g., TensorFlow, Caffe2, MXNet, Keras, PyTorch) to any number of Intel hardware architectures. Intel Distribution for Python: Intel's distribution of CPython contains numerous computational-heavy packages (e.g., NumPy, SciPy, and scikit-learn) that leverage MKL and DAAL. The result yields significantly improved performance when run on Intel hardware. Intel Distribution for Python is powered by the Anaconda data science platform and is available on many popular package management utilities such as conda, Docker, APT GET, YUM, and PIP. 2. Optimized ML Frameworks: Intel has worked with many leading developers of ML frameworks, including TensorFlow, PyTorch, Caffe2, ONXX, BigDL, and MXNet, to incorporate their hardware-optimized mathematical and analytical libraries into those various frameworks. At this level, developers expect to have tools that allow them to build deep learning models such as convolutional neural networks (CNN) while abstracting away the complexity of the underlying algorithms encapsulated in the aforementioned mathematical and analytical libraries. In other words, we want to build an appropriate model for our particular dataset, but we don't want to worry about the nuances of the math that drives a specific type of model. The focus is more on engineering and design than it is on mathematical theory and computer science. Indeed, though, the latter's appreciation is needed to select the Figure 5: Analytics Zoo helps to bring AI solutions to Big Data problems. (Source: Intel)