No abstract

In this paper, we propose VitaNet, a radio frequency based contactless approach that accurately estimates the PPG signal using radar for stationary participants. The main insight behind VitaNet is that the changes in the blood volume that manifest in the PPG waveform are correlated to the physical movements of the heart, which the radar can capture. To estimate the PPG waveform, VitaNet uses a self-attention architecture to identify the most informative reflections in an unsupervised manner, and then uses an encoder decoder network to transform the radar phase profile to the PPG sequence. We have trained and extensively evaluated VitaNet on a large dataset obtained from 25 participants over 179 full nights. Our evaluations show that VitaNet accurately estimates the PPG waveform and its derivatives with high accuracy, significantly improves the heart rate and heart rate variability estimates from the prior works, and also accurately estimates several useful PPG features. We have released the codes of VitaNet as well as the trained models and the dataset used in this paper.

Reinforcement Learning (RL) has brought forth ideas of autonomous robots that can navigate real-world environments with ease, aiding humans in a variety of tasks. RL agents have just begun to make their way out of simulation into the real world. Once in the real world, benchmark tasks often fail to transfer into useful skills. We introduce DoorGym, a simulation environment intended to be a first step to move RL from toy environments towards useful atomic skills that can be composed and extended towards a broader goal. DoorGym is an open-source door simulation framework designed to be highly configurable. We also provide a baseline Proximal Policy Optimization and Soft Actor-Critic implementation, which achieves a success rate of up to 70% for common tasks in this environment.

No abstract

In this paper, we introduce a method to compress intermediate feature maps of deep neural networks (DNNs) to decrease memory storage and bandwidth requirements during inference. Unlike previous works, the proposed method is based on converting fixed-point activations into vectors over the smallest GF(2) finite field followed by nonlinear dimensionality reduction (NDR) layers embedded into a DNN. Such an end-to-end learned representation finds more compact feature maps by exploiting quantization redundancies within the fixed-point activations along the channel or spatial dimensions. We apply the proposed network architectures derived from modified SqueezeNet and MobileNetV2 to the tasks of ImageNet classification and PASCAL VOC object detection. Compared to prior approaches, the conducted experiments show a factor of 2 decrease in memory requirements with minor degradation in accuracy while adding only bitwise computations.