KDnuggets Home » News » 2016 » Nov » Opinions, Interviews » Continuous improvement for IoT through AI / Continuous learning ( 16:n42 )

Silver BlogContinuous improvement for IoT through AI / Continuous learning


 
 
http likes 132

In reality, especially for IoT, it is not like once an analytics model is built, it will give the results with same accuracy till the end of time. Data pattern changes over the time which makes it absolutely important to learn from new data and improve/recalibrate the models to get correct result. Below article explain this phenomenon of continuous improvement in analytics for IoT.



Synopsis

Continuous improvement for IoT through AI

In this post, we discuss how we could implement Continuous improvement and Continuous learning in IoT systems. Continuous improvement is a well-known concept (ex from Kaizen or Six Sigma). For IoT, we propose that to implement continuous improvement, we have to implement continuous learning.  Specifically, this means we train a model in one environment and deploy it to multiple points in the IoT ecosystem (Device level, work flow level and system level).  Models could be built and deployed across devices to capture learning at various points. This learning is then consolidated through a central model. The new data is continuously monitored and when data patterns change,the model is re-calibrated to accommodate the latest knowledge. Thus, this model ties to Enterprise AI. We discuss these ideas in the forthcoming Implementing Enterprise AI course starting in January 2017

IoT Data Analytics – the flow of Data and application of analytics

Before we address the question of Continuous Improvement and Continuous Learning, we first discuss the question:  At which points could you apply analytics to the IoT ecosystem?

Typically, data arising from sensors is in time series format and is often geo-tagged. This means, there are two forms of analytics for IoT: Time series and Spatial analytics. Time series analytics typically lead to insights like Anomaly detection. Thus, classifiers (used to detect anomalies) are commonly used for IoT analytics to detect abnormal patterns.  In addition, by looking at historical trends, streaming, combining data from multiple events (sensor fusion), we can get new insights.

IoT devices create a large amount of Data. Typically, the goal of IoT analytics is to analyze the data as close to the event as possible because we need Real Time interaction. Hence, for IoT analytics, we often need to perform Edge Processing – a topic I have discussed in IoT Edge analytics.

Also, you could correlate data in multiple IoT streams. Typically, in stream processing, we are trying to find out what happened now (as opposed to what happened in the past).  Sensor data could be combined to infer an event (Complex event processing).  Finally, you could apply these models at the Data Lake.

Thus, IoT analytics (Data Science for IoT) could be applied to multiple places in the IoT ecosystem: Edge, Stream, Data Lake and in modalities like Complex Event Processing

IoT analytics: Implementing continuous improvement

The flow of IoT data and the application of IoT analytics are shown in the right hand side of the diagram. The left hand side speaks of Continuous improvement. Continuous improvement is not a new concept. It is an integral part of Kaizen and Six Sigma through techniques like Feedback loops

Here, we propose that at the implementation level for IoT, continuous improvement is tied to continuous learning. Specifically,

  • Every device has a state and an ‘intelligence’ – capacity to build and / or deploy a model. Models could be built and deployed aross devices to capture learing at various points.
  • This learning is then consolidated through a central model.

Continuous learning could be implemented at three levels for IoT devices

  • Device level analytics are primarily concerned with the state of the device and in reporting its status  Monitoring conditions ex thresholds
  • Workflow level analytics are primarily concerned with the state of the chained process across one or more physical devices.
  • System level analytics are primarily concerned with historical trends and inferring new analytics

In this context, an analytic is a model (Algorithm trained on dataset). An analytic can be typically developed at the system level. Once developed, it can be deployed to a device (Operational level) or chained with other analytics to create a workflow (probably spanning multiple devices).  Analytics can be orchestrated at multiple levels.

Implementation of Continuous Learning and Continuous Improvement

The idea of implementing continuous improvement through models which are continuously learning is complex but not purely conceptual. There are many real-life examples of this model in the industry most notably in Predix (GE) and Siemens (Industrie 4.0)

We also see implementations of creating models on one platform and deploying them in other platforms in various places such as Salesforce.com Einstein, Dell Edge analytics with statistica, IBM with PMML implementations, PMML support for Databricks(Spark), PMML support in Python. H2O.ai with POJO. We could also see it in terms of devops as a model of continuous improvement i.e. a blurring of roles and an integration of development and operations into a single working relationship with analytics at the centre. It also fits in with a model I have been speaking of i.e. a logical  AI layer for the Enterprise

Conclusions

Here, we propose that at the implementation level for IoT, continuous improvement is tied to continuous learning. Specifically,

  1. Continuous improvement needs continuous learning
  2. Every device has a state and an ‘intelligence’ – and potentially a capacity to build and / or deploy a model.
  3. Models could be built and deployed aross devices to capture learing at various points. This learning is then consolidated through a central model.
  4. The idea of implementing continuous improvement through models which are continuously learning is complex but not purely conceptual. There are many real-life examples of this model in the industry

We discuss these ideas in the forthcoming Implementing Enterprise AI course starting in January 2017

Bio: Ajit Jaokar‘s work spans research, entrepreneurship and academia relating to IoT, predictive analytics and Mobility. His current research focus is on applying data science algorithms to IoT applications. This includes Time series, sensor fusion and deep learning(mostly in R/Apache Spark). This research underpins his teaching at Oxford University (Data Science for Internet of Things) and ‘City sciences’ program at UPM(Madrid). Ajit is also the Director of the newly founded AI/Deep Learning labs for Future cities at UPM(University of Madrid).

Related: