DL-aided Active User Detection

Integrating 5G-based IoT networks with Deep Learning models, for solving some problems. I chose to solve the Active User Detection problem, wherein only actively transmitting end-point devices are to be identified, out of a massive number of devices in a network. I trained a custom Neural Network on synthetically generated training data, followed by performing the same task using a Bi-LSTM, as a reference point for well-established RNNs, and compared the performances of the two in correctly solving the problem.

I used Python and related libraries(NumPy, Pandas) to generate time-series training data modelling the situation in hand. In order to calculate the support of an activity matrix, I used a custom Neural Network with 6 hidden layers, and a Bi-LSTM net as well. The Bi-LSTM network achieved about the same max accuracy, albeit with a lot less training data (about 50% less).