In this work, we explore the possibility of a new delivery method for lossless data, namely compressive transmission. It aims at minimizing the transmission data volume at runtime by exploiting the tailored information shared between the sender and the receiver. There are two approaches to leverage shared information for compression: 1) using a DNN-based codec as a proxy for shared information and 2) applying redundancy elimination using deduplication. However, these approaches have not been studied in depth to utilize the trade-off between the compression rate and the amount of shared information. Compared to these approaches, compressive transmission is unique as it fully leverages the abundance of information available on both sides, which is chosen and placed purposely. To bring the concept to reality, we propose nCTX, a neural network-powered Compressive Transmission System that adaptively exploits a generative model and matching blocks. nCTX extracts the optimal semantic data from the input data, exploiting shared information to closely imitate the original and compensate it with the offset (i.e., difference). Extensive evaluations in mobile platforms confirm that nCTX reduces the transmission volume significantly by 25.8% and 23.3% compared to FLIF and RC, the state-of-the-art image codecs, respectively, in comparable or shorter computation times.

Offloading neural network inferences from resource-constrained mobile devices to an edge server over wireless networks is becoming more crucial as the neural networks get heavier. To this end, recent studies have tried to make this offloading process more efficient. However, the most fundamental question on extracting and offloading the minimal amount of necessary information that does not degrade the inference accuracy has remained unanswered. We call such an ideal offloading semantic offloading and propose N-epitomizer, a new offloading framework that enables semantic offloading, thus achieving more reliable and timely inferences in highly-fluctuated or even low-bandwidth wireless networks. To realize N-epitomizer, we design an autoencoder-based scalable encoder trained to extract the most informative data and scale its output size to meet the latency and accuracy requirements of inferences over a network. We also accelerate N-epitomizer by exploiting light-weight knowledge distillation for the encoder design and decoder slimming for the decoder design, reducing its overall computation time significantly. Moreover, we extend our N-epitomizer to support multiple DNNs by extracting and offloading the union of the essential information required for each DNN. Our evaluation shows that N-epitomizer achieves exceptionally high compression for images without compromising inference accuracy, which is 21$\times$, 77$\times$, and 192$\times$higher than JPEG compression, and 20$\times$, 55$\times$, and 86$\times$higher than the state-of-the-art DNN-aware image compression GRACE for semantic segmentation, depth estimation, and classification, respectively. Our results show N-epitomizer’s strong potential as the first semantic offloading system to guarantee end-to-end latency even under highly varying cellular networks.

In this paper, we present a new transportation mode recognition system for smartphones called DeepVehicleSense, which is widely applicable to mobile context-aware services. DeepVehicleSense aims at achieving three performance objectives: high accuracy, low latency, and low power consumption at once by exploiting sound characteristics captured from the built-in microphone while being on candidate transportations. To attain high energy efficiency, DeepVehicleSense adopts hierarchical accelerometer-based triggers that minimize the activation of the microphone of smartphones. Further, to achieve high accuracy and low latency, DeepVehicleSense makes use of non-linear filters that can best extract the transportation sound samples. For recognition of five different transportation modes, we design a deep learning based sound classifier using a novel deep neural network architecture with multiple branches. Our staged inference technique can significantly reduce runtime and energy consumption while maintaining high accuracy for the majority of samples. Through 263-hour datasets collected by seven different Android phone models, we demonstrate that DeepVehicleSense achieves the recognition accuracy of 97.44% with only sound samples of 2 seconds at the power consumption of 35.08 mW on average for all-day monitoring.

As video calls and personal broadcasting become popular, the demand for mobile live streaming over cellular uplink channels is growing fast. However, current live streaming solutions are known to suffer from frequent uplink throughput fluctuations causing unnecessary video stalls and quality drops. As a remedy to this problem, we propose PERCEIVE, a deep learning-based uplink throughput prediction framework. PERCEIVE exploits a 2-stage LSTM (Long Short Term Memory) design and makes throughput predictions for the next 100ms. Our extensive evaluations show that PERCEIVE, trained with LTE network traces from three major operators in the U.S., achieves high accuracy in the uplink throughput prediction with only 7.67% mean absolute error and outperforms existing prediction techniques. We integrate PERCEIVE with WebRTC, a popular video streaming platform from Google, as a rate adaptation module. Our implementation on the Android phone demonstrates that it can improve PSNR by up to 6dB (4x) over the default WebRTC while providing less streaming latency.

Robotic rollators (walkers with motor) require appropriate assist performance to improve the quality of life of the elderly and disabled. In this pilot study, the performance of commercial robotic rollators was evaluated by analyzing the effect of load condition on muscle activities and knee and hip joint angle when walking on a slope. Results showed that, the highest muscle activity occurred under a load of 5 kg but there was no difference in joint angle according to load condition. In future studies, we plan to test the performance with many subjects. In addition, we will perform motion analysis and muscle activity analysis according to changes in assist, brake, and speed levels.

A new transportation mode recognition system for smartphones, VehicleSense that is widely applicable to mobile context-aware services is proposed. VehicleSense aims at achieving three performance objectives: high accuracy, low latency, and low power consumption at once by exploiting sound characteristics captured from the built-in microphone while being on candidate transportations. To attain high energy efficiency, VehicleSense adopts hierarchical accelerometer-based triggers that minimize the activation of the microphone of smartphones. Further, to attain high accuracy and low latency, VehicleSense makes use of non-linear filters that can best extract the transportation sound samples. Our 186-hour log of sound and accelerometer data collected by seven different Android smartphone models confirms that VehicleSense achieves the recognition accuracy of 98.2% with only 0.5 seconds of sound sampling at the power consumption of 26.1 mW on average for all day monitoring.