Changes to the community structure and the menu bar
Significant changes have happened to the structure of our community and our drop down menu bar. READ ABOUT IT HERE
Big Data

Machine Learning APIs power a more intelligent Internet of Things at the Edge

Machine Learning APIs power a more intelligent Internet of Things at the Edge



The Internet of Things is the evolution of machine-to-machine (M2M) technology and is the interconnection of devices and management platforms. Some IoT devices see, while others listen, sense, monitor, and/or control. A few are even capable of near-human-like interaction. In many cases these devices are continuously capturing and processing a lot of data that only gets used once before being archived in some sort of log-file or even dumped.

This isn’t necessarily a bad thing, since the data often has a very short shelf-life, and it isn’t always needed for historical reporting as much as it is needed for generating timely insights and alerts.

Where these IoT devices are in fact already doing some limited analytics at or very near the point of capture (as in the case with true Edge Computing systems), there is opportunity to create a more intelligent, more relevant, and more positive experience or outcome from the Internet of Things by using Haven OnDemand Machine Learning APIs to perform early analytics and computing that enhances or augments the data that is being acquired and aggregated at the edge.


Figure 1. Our shift left to the edge. Taken from the HPE Business whitepaper, Harness the power of IoT data: Real-time decision-making and control at the edge.

IoT devices are often built with hardware, firmware and connectivity that is somewhat future-proofed with lots of latent disabled ability, but enabled and sold as a Minimum Viable Product with application software that only tackles a very narrow use case. The problem here is that the device is often capturing data that undergoes minimal processing to generate one insight. The opportunity therefore is to extract more actionable insights before the data is archived or discarded.

Many IoT device manufacturers realize this and for that reason, they typically open up parts of the device platform to third party developers so that these developers can extend and customize the capabilities even more. In fact, in November of 2015, the IDC published a paper wherein they set the prediction that “2016 Will Become the 'Year of the IoT Developer' with More than 250,000 Unique IoT Applications Created by 2020.”

Haven OnDemand is gaining increasing popularity with these IoT developers, as the platform offers more than 70 APIs that enable developers to accelerate innovation and harness the power of Machine Learning, Cognitive Computing, Natural Language Processing/Search, Predictive Analytics, and Big Data.

There are many use cases for Machine Learning to make IoT devices more intelligent, but we’ll touch on a few to illustrate what is possible.

Detection of trends and anomalies

Some of the leading home automation products include a variety of sensors that can control electrical devices, detect when a door has been opened/closed, know the location of family members based on the GPS data from their connected smartphone, and even allow users to setup routines that automate how these devices behave and interact based on a given set of conditions. One of the challenges faced however, is that the apps typically don’t have sufficient insights to understand behavioral trends and anomalies, so the user experience is often degraded with false positives and annoying behaviors. In a case such as this, developers might consider using the trend analysis and anomaly detection APIs from Haven OnDemand. For example, by automatically recognizing new patterns (trends) for a household, the app can reduce the occurrence of false positive alerts. Furthermore, by understanding these new patterns, the app can also recognize the most likely anomalies thus enabling a better end user experience.

One startup used these APIs to build a route planning mobile app that provides the user with the statistically safest route to walk from point-A to point-B. It achieved this by analyzing local law enforcement open data crime statistics to detect specific crime trends and specific crime anomalies. Certain crimes that are trending up in frequency could result in an alternative route being proposed, while crimes that are relatively infrequent are discounted as anomalies in the overall route planning algorithm.

Location-specific insights and answers

IoT devices generally do not have built in data warehouses. They are however connected to the internet, and this opens up a whole new world of possibilities for developers using Haven OnDemand. The unstructured text indexing and search APIs enables the embedding of location-based search functionality, while the map coordinates API enables translating geo-coordinates into human readable location data that can be used to augment the data that is to be stored for deep analytics at a later date e.g. translating the GPS coordinates from a crime-scene captured by a law-enforcement body camera into location data that can be used to augment aggregated open data crime statistics.

Machine vision

As was mentioned earlier, some IoT devices can see. They have cameras and are capable of capturing both still images and video. Most IoT devices do nothing more than save these files to a database for future review. A more intelligent IoT solution would analyze still images to detect the presence of faces, recognize and extract text via Optical Character Recognition (OCR), identify corporate logos and even read barcodes. With video-enabled IoT devices, detecting scene changes with key-frame extraction along with license plate recognition would enable a slew of security related apps such as for Safe Cities.

In retail, in-store IoT enabled camera systems can yield incredible insights for retailers. Examples include counting customers, analyzing customer demographics, analyzing customer personal effects to detect logos and determine brand preferences, analyzing real-time social media check-in mentions for sentiment, and point-of-sale data trend analysis. Retailers can gain a significant competitive advantage through an intelligent IoT network that is powered by Machine Learning.

One developer had a bit of fun with the face detection API and used it to build an IoT holiday ornament that recognized when a face was present in the video stream, and then used the coordinates of the face in the image to create digital signage with a personalized holiday greeting. Another developer built a face-recognition Nerf-gun to toy around with his co-workers.

Command and control speech recognition

Wearable technology such as smart watches are also considered to be a part of the Internet of Things. Some of the more capable devices include built-in microphones, and this opens up the possibilities for command and control speech recognition enabled use cases. Christian Berg, Developer & Founder at Appsolutt AS, built an enterprise IoT solution for field service management, where he integrated speech-to-text in an Apple watch client to enable voice command updates to HP Service Manager.

These are only a few of the possibilities. Do you see a use for Haven OnDemand machine learning APIs in any of your IoT solutions? Please add your comments below. 

Follow Sean on Twitter:

About the Author


Sean Hughes is the head of Developer Marketing and Evangelism for HPE Haven OnDemand; responsible for engaging, enabling and supporting developers, data scientists, startups and businesses from all industries to accelerate innovation and harness the power of Machine Learning, Cognitive Computing, Natural Language Processing/Search, Predictive Analytics, and Big Data. Follow Sean on Twitter: