We live in a digital world. It is not just a marketing buzzword anymore. From banking to healthcare, from government to education, from work to entertainment, from communication to transportation – digital technologies have penetrated almost every aspect of modern life.
In the past, building a website was a serious endeavor. You had to design it using a complex web development tool like Dreamweaver or FrontPage and then went through a laborious process of uploading files to a server and writing HTML code for it. Today, with the advent of “cloud computing” – a term used for hosted services provided on the internet – website creation has become as easy as making a simple Google search.
A: Cloud computing for machine learning
The ability of computers to run machine learning algorithms has been highly dependent on their hardware capabilities. Machine learning is an area of data analysis where the machines learn from data and give useful solutions based on it. To give an example, say you want to analyze data related to the purchase of mobile phones in the US. The machine learning algorithm will analyze this data and present insights like “people who bought iPhone 7 were more likely to buy Samsung S8 than those who bought older iPhones”. This kind of data analysis can be done by anyone with access to the internet.
Cloud computing has changed the way machine learning is done today. Before cloud computing, it used to take days or weeks to train a machine learning model that could only be used locally on the computer where it was trained. Now, with cloud computing, it is possible to make use of resources across the globe and train sophisticated machine learning models in minutes by using tools like Google Cloud ML.
Cloud computing has made it possible to make use of AI-powered apps like TensorFlow (Google), MXNet (Amazon), and Caffe2 (Microsoft) to build machine learning models for any business problem.
A: Section introduction
Let us look at an example where cloud computing is used to train a machine learning model for image recognition. For our example, we will consider face recognition by artificially creating images that are distorted using two or more random transformations. A machine learning algorithm can be trained using these distorted images to identify the original image after applying the same transformations. To train this machine learning model, we could do so locally on our own computer but it would take time and would require large amounts of memory, both of which are costly. Alternatively, we could also upload these images to cloud servers like Google Cloud ML or Microsoft Azure Machine Learning Studio where they are stored in the cloud. Because of the high computing power available on these servers, it takes less time to train the model and hence, saving money on local hardware costs. Once the model is trained, we can use it to recognize faces in real-time by uploading images of faces to compute its predictions locally on our device without having to upload anything to the cloud again.
A: Section conclusion
Cloud computing has lowered the barriers to entry for developing deep learning models for problems like image recognition, natural language processing, voice recognition, recommendation engines, etc. It has also made developing machine learning models accessible even for smaller companies that cannot afford expensive hardware. Furthermore, because of the high scalability offered by cloud computing platforms, it is possible to deploy machine learning models at larger scales without incurring prohibitive costs.