Nowadays, deep learning faces the challenges of data collection and the complexity of the computations. Due to the last problem, a big part of the innovation in hardware is aimed at speeding up the deep learning experiments, like new GPUs with a greater number of cores and different from today’s architecture that are now under development. According to Marc Edgar, a senior information scientist at GE Research, in the next 3–5 years, deep training will shorten the development time of software solutions from several months to several days. This will lead to better functional characteristics, increased productivity and reduced product costs.
Speaking of data collection, now almost all large firms have realized its importance and influence on the effectiveness of the work. In the coming year, companies will start using even more data, and the success will depend on the ability to combine disparate data. In 2018, companies will collect customer data via CRM, ticket systems, BMP and DMP, omnichannel platforms. Also, there is a rise in popularity in collecting data on specialized sensors like LIDAR. The integration of existing systems and the integration of all types of client data into a single information pool will definitely be a trend. Moreover, startups will continue to create new methods for gathering and using data, and therefore the costs for it will be reducing.
EmoticonEmoticon