Artificial Intelligence, Blockchain and Machine Learning globally are considered to be the holy trinity of technologies. Startups, SMEs and large corporations all together are looking at the various use case of these technologies. In a recent report, TMT predictions for 2018, Deloitte identified five key developments that will lead the popularity of machine learning in the future.
Reducing the Need of Training Data: Training a Machine learning solution needs tons of data elements and it can be quite time consuming and an expensive exercise. But as promising techniques are emerging, the time to train the ML model will significantly come down. Additionally synthetic data, which mimics the characteristics of real data, can open opportunities to crowdsourcing of data and add on to the purpose.
Another technique that could reduce the need for training data is transfer learning. With this approach, an ML model is pre-trained on one data set as a shortcut to learning new data set in similar data.
Explaining Results: Even though machine learning solutions are getting more and more impressive, it very difficult to explain how it takes the decisions. This is why Machine learning models are undesirable for various applications. However, Deloitte claims there are number of techniques being created that would help people understand how certain ML models work. Further, with this field of work ML solutions will be more interpretable and accurate.
Deploying Locally: Going forward, Deloitte also predicts more of ML will be introduced to smart phones and smart sensors. Furthermore the technology will also get deployed to smart cities, autonomous vehicles, wearable's and IoT products.Technology companies Google, Microsoft, Facebook and Apple are developing ML software models to undertake tasks such as image recognition and language translation on portable devices. While global firms like Intel and Qualcomm are developing in-house power efficient Artificial Intelligence chips to bring Machine Learning to phones.
Automating Data Science: Data exploration and feature engineering consume as much 8 per cent of data scientist's time. However, these tasks can be automated. A growing number of tools and techniques for data science automation, offered by established companies as well as venture-backed startups, should help shrink the time required to execute ML related proof of concept.
This will automate data scientist's job and improve productivity in the community. Additionally, the development will also help companies to double their work in machine learning space.
Accelerating Training: Globally, startups in the manufacturing domain are working to develop special hardware which would significantly reduce the time needed to train ML models, by using speed based calculations and transferring data to chips. "Early adopters of these specialized AI chips include major technology vendors and research institutions in data science and ML, but adoption is spreading retail, financial services and telecom."