By Gopal Rajaraman
Investment Principal, Motorola Solutions Venture Capital
The history of the Motorola Solutions Venture Capital (MSVC) team began in 1999 when Motorola Inc. leaders centralized corporate venture capital (CVC) investing activities under one banner. In the early 2000s, the focus of the CVC unit was primarily mobile and telecom technologies. After the spinoffs and divestiture of the Motorola Mobility and Networks business units, the CVC team was renamed Motorola Solutions Venture Capital. MSVC has invested in more than 200 companies since 1999 and is continuing the tradition of investing to inject innovative technology into Motorola Solutions.
Motorola Solutions’ primary market is public safety (law enforcement, fire, ems), so MSVC devotes the majority of its efforts on investing in the future of public safety. One of Motorola Solutions’ core competencies has always been “mission-critical communications.” Mission critical simply means that the technology works seamlessly, every time, everywhere and for everyone. Bleeding-edge consumer technology always takes time to get adopted by government or public safety agencies, because in most cases, the technology is not optimized for mission-critical needs. However, consumer technologies have enormous potential to vastly enhance the productivity of public safety – and enable law enforcement personnel to better maintain the safety and security of civilians.
Realizing this possibility, Motorola Solutions has intentionally and deliberately evolved its mission from providing mission-critical communications to developing “mission-critical intelligence.” MSVC and Motorola Solutions’ Chief Technology Office (CTO) are at the forefront of this mission. Consequently, MSVC’s primary thesis is to invest in companies having “bleeding edge” consumer technology that can be enhanced to meet the rigorous performance standard of mission-critical intelligence, which is required for public safety applications. The figure below shows some of the technologies that MSVC has recently invested in and how they fit within Motorola Solutions’ vision for mission-critical intelligence.
Right now, one such technology that could potentially be a game changer for public safety is “Deep Learning” (DL). DL has received substantial media coverage during the past few months. Many VCs are investing heavily (more than $300 million in each of the last two years according to CB-Insights) and numerous online articles explaining the technology and its potential have popped up. We recently interviewed Dan Law, chief data scientist of Motorola Solutions, to understand this “buzzword” and how deep learning can be applied to the public safety domain. An excerpt of the interview follows:
Deep learning is a class of emerging machine learning algorithms that are driving transformation in artificial intelligence, pattern recognition, prediction, language processing, computer vision, robotics, etc. For many hard problems in these disciplines, DL seems to be the first machine learning approach that works well.
Why Deep Learning?
For three main reasons. First, DL algorithms are fairly general purpose. That is, the basic algorithmic structures can be applied broadly across many types of data and use cases, from learning to identify objects in video to identifying malware in files. Second, unlike traditional machine learning, one does not have to tell DL algorithms what features in data are important to learn from – for example, facial measurements, eye color, etc. for facial recognition. Such features traditionally determined by human experts. DL algorithms, on the other hand, figure out what features are important themselves automatically. This not only saves human time and effort, it also mitigates human biases. Third, DL algorithms are demonstrating high accuracy – very high accuracy. So much so that they are starting beating humans at some challenging tasks.
Inspired by neurological studies of animals (especially the brain), artificial neural networks models have been used, since the early 1950s to predict behavior of complex systems. Artificial neural networks (ANNs) are generally represented as systems of interconnected “neurons” which exchange messages between each other. However, the last few years have seen useful advances in related technology that have made DL useful and hence interesting again. This is because DL algorithms are maturing (for general learning) at the same time that big data is proliferating (for effective training) and at the same time that massive parallel processing is increasing in power and decreasing in cost, for example using GPUs (so we can train in reasonable amounts of time!). Thanks to this confluence of events, DL can solve hard problems cheaply and effectively!
What is it being used for?
Lots of interesting things. DL can be used to learn many things. We all heard about AlphaGo beating humans at Go. AlphaGo used deep learning. DL is being used to train robots to walk. Virtual assistants are learning from humans to perform tasks. Even cars can learn to drive themselves with DL, possibly safer and faster than humans can. DL is also learning to translate languages, to identify sounds in audio, to identify and track objects video, to identify threats in cyber data, to paint paintings in the style of the masters, and many more use cases. There are likely countless use cases where deep learning can solve hard problems we haven’t thought of yet!
How can Motorola Solutions innovate using DL?
One of the interesting use-cases we are looking at right now is what we internally call “intelligent edge.” By edge, we mean a device that is deployed in the field – such as a smartphone or a two-way radio, or a fixed IP camera, or any of the wide variety of sensors out there. Historically, these Internet of Things (IoT) sensors have been “dumb” and they send data to the “cloud” and all the big-data analytics run on the cloud. However, for public safety, sometimes this is too slow. With all the advances in computing and sensors, sensors have now become more intelligent and now the capability exists to do most of the analytics right on these edge devices to help users to make rapid decisions, or provide them with insights in real-time. For example, DL can be employed for computer vision to detect objects such as weapons (guns) or detect events such as fights or crimes in progress. Hence, edge devices can now be provided with the intelligence to detect these objects or events in real-time. This intelligence can be quickly disseminated to law enforcement personnel to help save lives or prevent crimes.