This research work employs deep-learning neural-networks in aiming to unveil the possible existence of a relation between mean rates of seismic activity among consecutive large seismic events and their interim time-intervals. The research is conducted in possibly discrete seismic areas of the steady flow of input strain energy, identified in the southern front of the seismic Hellenic arc. Periods with low-level seismicity in terms of activity result in accumulation of strain energy, which is being stored in under-ground geological faults in the Earth’s crust, which in turn are acting as energy storage elements. On the contrary, the occurrence of strong earthquakes acts as a decongestion mechanism which causes the release of significant amounts of the stored energy to the surface of the Earth. Accounting for mean seismicity rates on regular time intervals, e.g., monthly, results in a tool that enables monitoring the underlying management system of the stored seismic energy. The captured information from the processed big-data, regarding not just quantity, but mostly type variation. It was channelled to a deep-learning model whose purpose is the identification and simulation. It was aided by parallel processing training algorithms, of the potential relation of mean seismicity rates, recorded in between consecutive strong earthquakes and their interim time-intervals, in a particular possibly discrete seismic area. Successful training yields a real time dynamic mechanism able to estimate the duration period between the last recorded and next upcoming strong earthquake. The proposed model achieves noteworthy approximations, in the range of approximately two weeks to four months, of the time interval between successive large earthquakes, which reside within the foreshock-aftershock time period of each main seismic event. The obtained range falls well-short of the observed mean seismic recurrence times laying in between 1.5 to 2 years.