News & Events

AI Trends in 2019

2018.12.21

As the 2018 comes to an end, AI is advancing and it is doing it rapidly, opening up incredible new sceneries for the years to come. It goes without saying that, once the Bottos model and data exchange will be fully operating, the AI development will advance at a more incredible speed.

The main trends to take in consideration at this point for the next year are at least five.

 

Increasing Cooperation Between AI and IoT

 

In 2019, AI may start to be deployed on the IoT at the edge computing layer. In fact, many models that will be trained in the public cloud or even exchanged on the Bottos platform, will be deployed at the edge.

One of the main application of AI and IoT will be the industrial setting, where we can more easily have artificial intelligence that can perform root cause analysis, outlier detection and predictive maintenance of the equipment.

Advanced machine learning models based on deep neural networks may eventually be optimized to run at the edge, thus increasing its computing performances. They will likely capable of dealing with time-series data, speech synthesis, video frames and unstructured data generated by connected devices such as sensors.

IoT will probably become one of the biggest driver of artificial intelligence in the enterprises, that may effectively harness the power of artificial intelligence.

 

Machine Learning Turns Automatic

 

Auto Machine Learning (AutoML) will be a trend that will fundamentally change the face of ML-based solutions. It has the potential to empower developers with the opportunity to evolve machine learning models that can be able to address complex scenarios without going through the traditional, time and resource consuming process of training models.

 

AutoML perfectly sits in between cognitive APIs and machine learning platforms, providing the right level of customization without forcing the developers to go through elaborate workflows and unsurmountable costs. Unlike cognitive APIs, AutoML presents the same degree of flexibility but with custom data together with portability. It will be not unlikely to find a good source of models trained using AutoML on the Bottos exchange that can be furtherly developed, thus to speed up the innovation trends by 1000x.

 

AI gets Its Own Chip

 

Contrary to other types of software, AI heavily relies on specialized processors that cooperate with the CPU, and usually even the fastest and most advanced CPU may not be really effective in improving the speed of training an AI model. While processing, the model often need additional hardware to perform complex mathematical computations to speed up tasks such as facial recognition and object detection.

In 2019, chip manufacturers will start to produce and sell specialized chips that will make the execution of AI- enabled applications much faster. Most of these chips will be specially built for specific use cases and scenarios related to speech recognition, natural language processing and computer vision. Next generation applications from the healthcare and automobile industries will most likely rely on these chips for deploying AI to end-users.

 

Automated Developer Operations

 

Existing applications and devices are generating log data that is used for searching, analytics and indexing. The massive data sets obtained from several systems can be aggregated and correlated to find insights and patterns. When machine learning models are applied to these enormous data sets, IT operations can turn from being reactive to predictive. The great point is that all these data can be easily exchanged on Bottos platform.

When the power of AI is applied to operations, it will redefine the way infrastructure is managed. The application of machine learning and AI in IT operations will deliver intelligence to organizations. It will give a hand the operation teams to perform accurate root cause analysis.

 

Neural Network Cooperations

 

A critical problem in developing neural network models is the choice of the right framework. Data scientists and developers have to choose from countless choices that include TensorFlow, PyTorch, Caffe2, Microsoft Cognitive Toolkit, and Apache MXNet. However, once a model is trained in a specific framework, it is tough to port it to another framework.

The lack of interoperability among neural network toolkits is slowing down the adoption of AI, and the same goes for blockchain, problems that Bottos is trying to solve.

 

However, a trend in 2019 to watch actively is the development of a common framework that may speed up the creation and the adoption of AI models.