Machine Intelligence brings a potent combination of AI and IoT to businesses. This domain includes industrial automation, application of deep learning to sensor telemetry, edge computing, intelligent video analytics and cognitive computing.
Here are 5 trends of 2019 that influenced the field of machine intelligence:
1. AIoT Emerges as the Future of Industry 4.0
2019 saw the convergence of industrial IoT (IIoT) and AI. First-generation IoT platforms and solutions relied on a basic rules engine to orchestrate the workflow among the connected sensors and actuators. These rules engines were comparable to the functionality provided by IFTTT.
With AIoT, the complex event processing engines and stream analytics are natively integrated with AI. Instead of relying on static rules, AIoT-based engines learn from the historical data to derive baseline metrics for each of the connected devices. This integration helps correlate inbound telemetry ingested by various sensors to find hidden patterns. AIoT can perform anomaly detection with increased accuracy and precision which directly translates to enhanced predictive maintenance of industrial equipment.
Major IoT and edge computing platform providers including Amazon, Microsoft, IBM, Google and FogHorn support the native integration of ML models with stream analytics and stream analytics services.
2. The Rise of AutoML
AutoML is changing the way deep learning models are built and consumed. With the advancements in deep learning and the application of transfer learning, ML platforms are able to reduce the number of steps it takes to train and optimize a model.
Though AutoML initially targeted the use cases of image classification and object detection, it is now extended to support language understanding and even time-series data structured in a tabular form. Businesses can input a dataset to an AutoML platform and walk away with a fully trained model. Behind the scenes, AutoML engine will perform necessary steps such as data preparation, feature extraction, normalization, model selection, hyperparamater tuning and even deployment.
2019 saw the rise of AutoML for vision, language, and time-series data. Amazon SageMaker Autopilot, Azure ML Services, Amazon Comprehend Custom Label, Google Cloud AutoML, DataRobot are some of the platforms to support AutoML.
3. Rapid Growth of AI Accelerator Chips
2019 witnessed rapid growth in the purpose-built processors and chips that accelerate AI training and inference.
AI Accelerators complement CPU by taking over the intense mathematical computations involved in training and inferencing ML models. While GPUs are enabling accelerated training of deep learning models, custom AI accelerators and enhancing the inferencing of ML models.
In 2019, we have seen Intel launch the next iteration of Movidius and Myriad Vision Processing Unit (VPU) V2, NVIDIA launching Jetson Nano, and pre-announcing Jetson Xavier NG, the entry of Snapdragon 855 Mobile Platform from Qualcomm, Google shipping the Edge TPU and Huawei announcing the Ascend family of chips.
4. AI Inferencing Moves to the Edge
Edge computing has become the logical destination for AI and ML models. If the public cloud is the preferred environment to train deep learning models, the edge is becoming the target for running ML models for inference.
Combined with the purpose-built AI accelerators, specialized software is being built to optimize the models to run at the edge. The OpenVINO Toolkit from Intel, TensorRT from NVIDIA, TensorFlow Lite from Google, Qualcomm Neural Processing SDK and ONNX Runtime simplify and speedup the inference at the Edge.
5. Increased Focus on AI Platforms and Developer Tools
During 2019, the machine learning platform as a service (ML PaaS) offerings focused on increasing developer productivity. By combining platform-specific SDKs with Jupyter Notebooks and building plug-ins for mainstream IDEs, ML PaaS vendors attempted to make developing ML models simple for data scientists and developers.
Google announced a comprehensive AI Platform that includes AI Hub, ML Engine, and other components targeted at the data scientists. It also launched TensorFlow Enterprise to integrate the training and deployment workflow with GCP. Microsoft revamped Azure ML Platform to offer a visual designer and streamlined developer experience. More recently, Amazon launched SageMaker Studio, an end-to-end tool to train, debug, optimize, deploy and monitor ML models. Startups such as Paperspace have built hybrid ML PaaS products that run both in the public cloud as well as an enterprise data center.