Data and Intelligence published the monograph "Building Enterprise-level Recommendation Systems: Algorithms, Engineering Implementation and Case Analysis". Output 7 original articles on recommendation systems, data analysis, big data, and AI every week. "Data and Intelligence" (video account of the same name, Zhihu , Toutiao, Bilibili, Kuaishou, Douyin , Xiaohongshu and other self-media platform accounts) community focuses on the sharing and dissemination of knowledge in the fields of data and intelligence.
author | Harper
review | gongyouliu
editor | gongyouliu
In the previous content, I introduced you to the advantages of deep learning compared to machine learning, and also sorted out the current mainstream machine learning tools. Shared with everyone Content is a mainstream deep learning tool.
Deep learning is one of the most interesting areas in artificial intelligence , so there are many tools for creating deep learning artificial neural networks , these tools come in the form of deep learning frameworks. The Deep Learning Framework is an interface that enables developers to quickly and easily build and deploy deep learning AI models using a set of pre-built components.
One of the most popular deep learning frameworks is TensorFlow. Created in 2011 by the Google Brain team, Tensorflow is an end-to-end open source machine learning framework for developing, training, and deploying machine learning models. You can use it with most cloud-based machine learning service platforms, including Amazon SageMaker, IBM Watson, and Microsoft Azure. TensorFlow is cross-platform, meaning it can run on a variety of architectures—servers, smartphones, and even GPUs. Use the so-called "TensorFlow distributed execution engine" to abstract most of the hardware, use Python as the front-end application programming interface to build applications, and use high-performance C++ to execute these applications.
Then another framework is Microsoft Cognitive Tolkit (CNTK for short), which is an open source toolkit for commercial-grade distributed deep learning. As a competitor to TensorFlow, CNTK is another low-level deep learning framework for building and deploying deep learning models. CNTK appears to have advantages over TensorFlow in terms of processing speed, creating production-ready models, and supporting CPU and GPU calculations. TensorFlow scores higher in terms of ease of use, community support, and mobile deep learning. But these two frameworks are also under continuous development.
Another popular tool is Keras, which is an open source neural network library written in Python. You can run Keras on top of TensorFlow or CNTK (and other low-level deep learning libraries) and use it as a high-level API to simplify the process of building deep learning models. It is modular, scalable, user-friendly, and designed to enable rapid experimentation. If you are using deep learning artificial neural networks to recognize objects, there is also a deep learning framework called Caffe. This framework was developed by Berkeley AI research and a community of contributors. You can go to its official website to see more about caffe. Information.
Overall, if you are just starting to build your own deep learning artificial neural network, it is recommended to start with TensorFlow and Keras. As you gain more experience with machine learning algorithms, Python, and C++, there are more possibilities you can try. In addition to simplifying development and deployment, TensorFlow is one of the most popular frameworks for building deep learning networks on almost all popular cloud services. You can also run them on your own server, phone, computer or other computing device. It abstracts away most of the technical challenges involved in building these networks and provides the flexibility to run deep learning artificial neural networks on top of many existing technologies.
