Toward Multi-Modal Deep Learning-Assisted Task Offloading for Consumer Electronic Devices Over an IoT-Fog Architecture

Internet of Things (IoT) devices along with associated software have proliferated at an unprecedented pace, presenting the challenge of high energy use combined with latency during complex, time-sensitive transactions. Fog computing, i.e., a distributed computing paradigm, may be a potential remedy. However, despite these efforts, it is highly strenuous to regulate service latency and energy efficiency within the fog computing layer for IoT centered consumer electronics. In our research, we propose an algorithm called Dynamic Deep Reinforcement learning-based Task Offloading (DDTO). Therefore, in multi-modal IoT-Fog systems, the intelligent distribution of fog computing resources by DDTO is considered on the basis of resources constraints and the timeliness of completion of tasks. We use a log-normal distribution to describe delay and energy consumption so as to make up for varying modalities. Besides, a task prioritization problem, which is described as an integer programming problem, that minimizes service latency and energy consumption in fog servers is further described. DDTO yields better performance than a conventional Q–learning with respect to long-term expected rewards since it uses task priority weights derived from statistical latency and energy data. Experimental results demonstrate the benefits of DDTO on reducing service latency and consumption of energy, when compared to benchmark strategies, discussing issues with multiple modalities within IoT-Fog systems.