The Future of Software Innovation? Hardware-Enabled AI & ML Innovation

#Discussion(General) [ via IoTForIndiaGroup ]

You’re probably thinking right now, “All I ever read about is how AI, ML, blockchain, and XR are ready to revolutionize the world” and that’s exactly the point. There are some amazing software technologies coming out, but this “new” software has hardware in its DNA. Until recently, smart technologies have largely been limited by their access points: computers, tablets, smartphones, etc. Going forward, hardware innovations will become increasingly integral and valuable as the interface for tomorrow’s software. Hardware will capture the data through wearables, hearables, cameras and an increasing variety of sensors and will then be leveraged as the outputs to interact with the world as robots, drones, and the myriad of other IoT devices that are being developed.

As the ecosystem of devices, computation, connection, and data evolve, the platforms, tools and systems are naturally finding more synergy and lowering the barriers of integration. The line between what is a hardware or software product will continue to blur. The sensor technologies leading the way are cameras and microphones. If there is a camera, there’s a good chance there’s an AI stack behind it, with self-driving cars being the most prominent example. On the microphone/speaker side, the Smart Home assistants Amazon Echo, Google Home, and others are obvious and ubiquitous.

The beauty of hardware enabled AI/ML is that it not only crosses the boundary between the physical and virtual, but also between analog and digital. It’s particularly valuable when interacting with the world and dealing with its messy data. The next generation of AI hardware startups will take all that messy analog data and transform it into productive and executable knowledge that provide better experiences all the way from shopping to cancer treatments that enable personalized health care at scale.

While the future is clear, the hurdles are as well. Processing power, robustness, generalization and cost are all tradeoffs future hardware products will need to balance. Unlike the on-demand and scalable cloud and other services software enjoys, each hardware product will have onboard processing, sensors, connectivity tech, and other requirements that all make their way into the product cost.

We’re seeing a lot of edge computing such as NVIDIA’s Jetson line and Google’s Edge TPUs. TensorFlow is probably the most common AI framework, since it has such broad support for hardware deployment, including Raspberry Pi. ROS is still fairly popular despite being a jumble of mismatched and complicated software, and people have done ports to OpenAI’s Gym environments.

Read More..