AI Implementation in the Connected Home

AI and machine learning applications across verticals in the smart home ecosystem can draw on an array of data sources from the connected home. A device may constitute a single sensor, such as a motion sensor, though increasingly multi-sensors that bundle a collection of sensors in one device are becoming more common.??AI benefits from being able to gather a variety of sensor data in order to better understand the home context. As connected devices continue to be added, data sharing among them offer mutually beneficial value in facilitating artificial intelligence applications.

While the vision for AI typically involves sophisticated sensor fusion from an array of sensors, that is not necessarily required. A relatively small number of occupancy sensors can yield substantial information about the daily lifestyle patterns of a family

Companies seeking to leverage AI solutions have a variety of technology options available:

• Building on large public clouds with AI tools
• Developing hybrid public/on-premises infrastructure combining public and proprietary tools
• Contracting with software-as-a-service (SaaS) or AI-as-a-service (AIaaS) partners that pre-configure AI tools and services

In these various stack architectures, AI implementations are supported by a value chain of technology providers that begins with well-established algorithms and frameworks defined in academic settings and research labs, or more recently, by tech giant companies, such as Google?and Microsoft. Many of these building block tools have been published as open-source libraries with code written in Python or C++, the principal languages used for machine learning, though the original C programming language and a variety of newer languages such as Java, Scala, Squirrel, Lua, and R may also be used.

Public Clouds and AI Development Tools

Large-scale public clouds, such as Google?Cloud with TensorFlow, Microsoft?Azure, and Amazon?Web Services?(AWS) Machine Learning combine the most popular open-source tools with proprietary libraries to drive adoption of their core cloud services. The infrastructure, services, and tools have been optimized to provide scalable resources for companies developing AI-enabled solutions, such as smart home manufacturers.

AI Infrastructure: Data Tools

The foundation of the stack is based on tools for AI data management and AI computing. A data lake provides a repository for data stored in their raw form as well as their transformed form after processing by various applications. Data lakes can include unstructured data from documents and communications (email, text, PDFs), semi-structured data (time logs, CSV files, JSON, XML), relational databases, and binary data (images, audio and video). A separate relational database management system (RDBMS) with a row-based table structure typically use structured query language (SQL) to access the database. NoSQL tools enable storage and access to data by “non SQL” means other than row-based tables.

AI Infrastructure: Compute Tools

AI compute tools leverage high-powered, scalable CPUs and GPUs to deliver services such as batch processing, container orchestration, and serverless computing. These services enable the real-time, parallel data processing necessary for machine learning automation. Virtual machines apply the trained algorithms to new batches of data to make and test inferences leading to predictive analytics.

AI Services

Cognitive computing applications enable access and delivery of AI services via APIs. These services may provide generic AI computing capabilities or more customized computing for specialized business use cases. Conversational bot and personal assistant services extend AI computing into text and voice user interfaces through integration with desktop and mobile operating systems. For example, developers of smart home apps that seek to employ a conversational interface will find pre-configured tools that streamline the process of integrating those features into the app.

AI Analytics Tools

In addition to the infrastructure elements of the stack (data, processing, services), task-specific data analytics tools are tightly integrated with the platforms. These may range from wizards with easy-to-use graphical interfaces for dragging and dropping data sets and AI tasks into a workflow, to libraries of building block algorithms that can be further customized in a browser-based development environment.?Many of these AI tools also may be integrated into the hybrid cloud or on-premises configurations that are custom configured for the unique needs of companies.

This is an excerpt from Parks Associates research library. Thank you for reading our work. We welcome all comments.