The expected results of EDL are:
- Substantial improvement of compute and energy efficiency of DL methods, at algorithmic, implementation and hardware levels.
- Embedded applications of deep learning become obtainable; less need for using many high-end GPUs.
- On device data analysis reduces the need of data to be transported over the network and solves privacy, network latency, and data-bandwidth issues; it saves also a lot of energy.
- Deep learning becomes accessible for the non-experts: the DL technology should be as efficient and intuitive as building a website with Word Press.
- Less data required: training a DL model requires a huge amount of annotated data. EDL addresses this problem, reducing training time and making DL widely applicable to new problems.
EDL results will be prototyped and demoed with real-life use cases, favoring economic impact, as shown below:
