Temporal Convolutional Network-Based Signal Detection for Magnetotactic Bacteria Communication System.

Molecular communication (MC) aims to use signaling molecules as information carriers to achieve communication between biological entities. However, MC systems severely suffer from inter symbol interference (ISI) and external noise, making it virtually difficult to obtain accurate mathematical models. Specifically, the mathematically intractable channel state information (CSI) of MC motivates the deep learning (DL) based signal detection methods. In this paper, a modified temporal convolutional network (TCN) is proposed for signal detection for a special MC communication system which uses magnetotactic bacteria (MTB) as information carriers. Results show that the TCN-based detector demonstrates the best overall performance. In particular, it achieves better bit error rate (BER) performance than sub-optimal maximum a posteriori (MAP) and deep neural network (DNN) based detectors. However, it behaves similarly to the bidirectional long short term memory (BiLSTM) based detector that has been previously proposed and performs worse than the optimal MAP detector. When both BER performance and computational complexity are taken into account, the proposed TCN-based detector outperforms BiLSTM-based detectors. Furthermore, in terms of robustness evaluation, the proposed TCN-based detector outperforms all other DL-based detectors.

Addressing Hindsight Bias in Multigoal Reinforcement Learning.

Multigoal reinforcement learning (RL) extends the typical RL with goal-conditional value functions and policies. One efficient multigoal RL algorithm is the hindsight experience replay (HER). By treating a hindsight goal from failed experiences as the original goal, HER enables the agent to receive rewards frequently. However, a key assumption of HER is that the hindsight goals do not change the likelihood of the sampled transitions and trajectories used in training, which is not the fact according to our analysis. More specifically, we show that using hindsight goals changes such a likelihood and results in a biased learning objective for multigoal RL. We analyze the hindsight bias due to this use of hindsight goals and propose the bias-corrected HER (BHER), an efficient algorithm that corrects the hindsight bias in training. We further show that BHER outperforms several state-of-the-art multigoal RL approaches in challenging robotics tasks.