AI is the superset of various techniques that allow machines to be artificially intelligent. For an analogy, think of a Russian nesting doll: machine learning is a subset of AI, and deep learning is a subset of machine learning

Machine learning refers to a machine’s ability to think without being externally programmed. While devices have traditionally been programmed with a set of rules for how to act, machine learning enables devices to learn directly from the data itself and become more intelligent over time as more data is collected.

Deep learning is a machine learning technique that uses multiple neural network layers to progressively extract higher level features from the raw input data. For example, in image processing, lower layers of the neural network may identify edges, while higher layers may identify the concepts relevant to a human such as letters or faces.

An AI assistant is a program powered by machine learning that can respond to you, provide information, anticipate your needs, and perform tasks at your request. While these assistants are most commonly thought of in terms of smartphones and smart home speakers, they can exist in a range of devices and will become common in XR glasses, home appliances, connected cars, and more. With the 5th generation Qualcomm® AI Engine on Qualcomm Snapdragon 865, we’re enabling AI assistants with advanced capabilities to enhance user experiences while meeting the power and thermal constraints of mobile devices.

A: The biggest challenges AI models face are how to be more power efficient, how to learn from less data, how to learn from unlabeled data (unsupervised learning), and how to generalize across multiple tasks. The industry is also focused on making AI unbiased and explainable, such that we know how it works, where it fails, and how to quantify confidence levels. For example, we want to understand how the AI used for autonomous driving is deciding how to drive safely on a road under various environments and weather conditions.

A: AI is currently benefiting from more data and more efficient hardware, as well as better AI tools and networks/algorithms. Advancements in state-of-the-art accuracy for various tasks happen regularly due to the collaborative nature of the AI research community through papers and workshops. For example, Qualcomm AI Research has published many papers in the areas of power efficiency, personalization, and efficient learning. These advancements are being applied in more areas, integrated into different types of devices, and enhancing our user experiences.

A: When there’s bias in the data set that trains an AI model, the model will contain the same bias. A way to address this bias is by collecting robust and diverse data. For example, in distributed learning on smartphones, you may want to sample data from different geographies and demographics. If bias is noticed in algorithms, the data should be examined to determine whether new data should be added.

A: Edge computing is computing that happens at the edge cloud or edge device, while cloud computing occurs in the central cloud. Where the processing is located may result in different levels of performance, latency, or privacy. Both offer different benefits and are complimentary to each other.

A: Computer vision involves generating feature detectors, which traditionally have been hand-crafted by humans. With the help of large data sets of labeled images or videos, machine learning, and specifically deep learning, can learn these feature detectors automatically and more accurately than humans.

Distributed intelligence is the result of running artificial intelligence algorithms across various computing devices, such as a phone and a server in the cloud.

A: Edge computing has multiple meanings, but it is generally a specific location where the computing occurs. Distributed intelligence means that the intelligence is being created from running the AI algorithms in multiple locations by splitting the processing workloads.

A: We began researching AI more than a decade ago. Today, more and more intelligence is moving to end devices, and mobile is becoming the pervasive AI platform. Building on the smartphone foundation and the scale of mobile, Qualcomm Technologies envisions making AI ubiquitous—expanding beyond mobile and powering other end devices, machines, vehicles, and things. We are inventing, developing, and commercializing power-efficient on-device AI, cloud AI, and 5G to make this a reality.