AIoT – Internet of Things & Machine Learning, is this a hot recipe for new-age solutions?

AIoT - scope, working and use case of Artificial Intelligence of Things
Image showing AI use-cases with IoT
Listen Article

AIoT is a fairly new acronym coined for Artificial Intelligence of Things“. As you have rightly guessed, it is a combination of AI and IoT, the two major technological advancements of this decade. Together they are playing a pivotal role in creating cost-effective, intelligent and scalable solutions for a sustainable interconnected world. In other words, large scale industrial automation, which is self-healing and self-managed, is made possible by leveraging IoT sensing and AI decision-making capabilities.

AIoT today and tomorrow

A recent report by Nasscom (an Indian IT & BPO trade association) suggests a tremendous increase (of almost six times) in IoT patents filed in India alone, in the last five years. A similar report by WIPO (world intellectual property organization), mention that patents filed in AI grew at 28% between 2012-17. One recent market research estimated market size of AI in IoT projects to grow from 5.1 Billion USD in 2019 to 16.2 Billion USD by 2024, growing at CAGR of 26%. Google Trends on web search keywords show that IoT and machine learning have emerged as the most popular searches trending in the past few years, worldwide.

Let’s have a look at key events to boost AI and IoT in India:

  • In June 2018 NITI Aayog which is policy institute, of Government Of India, published National Strategy for AI.  The purpose is to empower the Indian workers for jobs of tomorrow and boost the country’s economic & social growth.
  • In Jan 2020 NITI Aayog released approach paper AIRAWAT, i.e. AI Research, Analytics and knowledge Assimilation Platform. The goal is to invigorate the AI ecosystem in India by building India’s first specialized computing infrastructure for executing AI-based projects. In short, democratize AI and make it easily accessible to citizens – #AIFORALL
  • In Feb 2020 ASUS – Taiwan based electronic multinational announced “AIoT for India” as its new business vertical

In our recent article, we discussed how Industrial IoT is disrupting business landscape around us. But to fully utilise the potential of ever booming IoT, we must understand its combination with another disruptive tech set – Artificial intelligence.

Drivers of IoT growth enabled by AI

Source: PwC

Types of computing in the world of AIoT

  1. Fog Computing is a fairly new metaphor, which remains invisible and is essential tech in IoT age. It works on decentralized architecture to make IoT applications more efficient. With billions of IoT sensors connected, cloud servers couldn’t aptly handle huge IoT-workloads. Fog computing helps to transfer IoT computation workload from cloud servers to network edge devices. It works as a backend technology at proximity to end-user and deals with automation devices.
  2. Dew Computing works at the frontend. In a simplistic way, it helps websites remain available without an internet connection. It harnesses the potential of personal computers or mobile devices, which can be called as dew-computer or dew-phone. It not only detaches applications from the cloud but also collaborates with cloud-based services, efficiently.
  3. Cloud Computing is a well-known computing paradigm. It is a resource-rich cluster of computers, which are on-demand, centralized, and depends on an internet connection. Any device having an internet connection can easily access or use cloud-based applications or resources in the form of service, such as IAAS, PAAS, SAAS and so on.
  4. Edge Computing is quite relevant in the industrial IoT (IIoT) world. Here, computation happens at the edge-device (or device-node), it provides real-time analysis using local data available. With sensor intensive data workloads, it optimizes bandwidth, network latency and execution speed. Edge TPU by Google can run sophisticated AI models right at the edge.

How AIoT solution works?

Sequence of steps in any AIoT enabled solution can be defined as following:

AIOT – How it works?
  1. Capture Data: IoT devices sense and generate a stream of sensor data also called Telemetry data. There can be a variety of sensor data based on a type of sensor (or actuator). Each source of telemetry data results in a channel.
  2. Ingest: Data generated by a fleet of IoT sensors can be massive and unstructured. It requires a data pipeline to transform this data, for example, a voltage signal converted to a temperature unit. Data received from multiple devices is aggregated and enriched for example adding metadata such as weather or traffic to a data point. Processed data is stored at appropriate data storage for further analysis. Popular data ingestion tools for capturing streaming data are Apache Kafka, and Google Pub/Sub.
  3. Storage: Due to the enormous volume, ingested data is persisted at scalable cloud storage. NoSQL databases (such as Apache Cassandra, HBase, and MongoDB) are preferred due to low latency and high throughput. IoT data is mostly time-series data, and therefore specialized time-series databases such as InfluxDB, OpenTSDB, and Prometheus can also be used.
  4. AI Modeling: Collected telemetry data forms the basis of machine learning models. Models trained on historic dataset to generate predictions for the future. There could be a variety of algorithms that can be implemented depending on the use case and input data type. Few examples are a)Anomaly detection to find a technical glitch or fault, b) Time series forecasting to find a trend, c) Clustering to find patterns, d) Image Segmentation and Object Detection on image sensor feed, f) Text-based models such text classification and named entity recognition (NER), g) Tree-based models such as Decision tree, Random Forest and XGBoost, h) Probabilistic model such as Bayesian i) Neural Network models such as RNN, LSTM and GRU, j) Deep Reinforcement learning models, and so on.
    Model Tuning – There is a famous saying about ML models – Every model is wrong but some are useful!”. Machine learning models are continuously fine-tuned, optimized and retrained on a batch of new data, this is an iterative process.
  5. Insights: Data points, outcomes and predictions are used to perform real-time actions, visual plotting is done for creating in-depth reports. Insights and advance dashboards help in aligning business goals, fine-tune processes and developing strategies for the future. Data Visualization tools such as Tableau and Microsoft Power BI are proved very effective in visualizing a huge amount of data having millions of data points.

Innovative application of AIoT

Technology has the potential to transform any industry. Agriculture is one such industry where AIoT based solutions offer the opportunity to solve prominent problems.

AIoT in Agriculture – India is an agriculture country with more than 70% of the population still dependent on farming, to support livelihood. Technology to improve farming-efficiency becomes even more necessary on this land. Smart farming, also known as precision farming is a way to improve both quality and quantity of agriculture products leveraging cutting-edge modern-technologies. IoT based sensors collect spatial and temporal data and feed it to a cloud server. AI-based models use deep learning techniques to understand data, find patterns and generate automatic analysis and forecasting. One of the advance application in this area is ‘smart pest monitoring system‘.

  • Smart pest monitoring system – Based on a research paper published by National Taiwan University in Nov 2019, AI and IoT based technique is found effective for intelligent pest control on farms. It uses the environment and wireless image sensors installed inside greenhouses to monitor population density of pests in possible hotspots. AI-based techniques such as image segmentation, object detection are used to detect pest objects in an image sensor-feed. More robust deep-learning-based techniques such as YOLOv3 or (the more recent one) YOLOv4 used to detect a target object with even higher accuracies, even in poor lighting conditions. CNN based neural network techniques used to do the classification of an identified object as pest or not-pest. Later, pest object is classified into major pest categories such as fly, whitefly, thrips or gnat. Notification or alarm are triggered to the farmer and other stakeholders, based on pest density findings.

Agritech in India is an emerging trend, Nasscom 2019 report underlines this. One such startup in this space is Fasal, which is harnessing the power of AI, IoT and Cloud to produce cost-effective agriculture solutions for Indian farms.

There is almost nothing we can think of that cannot be made new, different, or more valuable by infusing it with some extra IQ. In fact, business plans for the next 10,000 startups are easy to forecast: Take X and add AI

Kevin Kelley – in his book The Inevitable

Combination of AI and IoT is a major breakthrough, leading us to a “smart revolution”. It will not only cause a fundamental shift and give a competitive advantage to companies but will definitely contribute to cost-effective alternatives and make our life easier!

If you have an interesting article, case-study, or experience to share, we are more than delighted to publish that, please get in touch at

Anubhav Agarwal
Anubhav is a data scientist by profession who loves to explore new technologies and tools. He is passionate about table tennis, AI and evolution of Galaxy:). He feels that unemployability can be reduced with the right skills and knowledge among the people.