Swiping the screen to switch videos is a unique browsing behavior for short videos, intended to facilitate viewers in quickly searching for content of interest. However, frequent video switching can result in nearly half of the data being used to transmit never-watched video data, leading to unnecessary network load via resource wastage. To tackle this problem, recent studies have utilized historical viewing data to predict the necessary length of videos to download based on viewing probability. Critically, the precision of the predictions plays a pivotal role in shaping both data consumption and user experience. This paper addresses issues that emerge with inaccurate predictions by proposing KBL (Kettle-style Buffer Loading), a novel algorithm to balance waste with a high-quality video experience, without requiring extensive training or prior knowledge. Inspired by tea kettle service, KBL reduces waste by setting the boundary of the respective video buffers, current and pre-loaded videos, based on an evaluation of the network conditions. Through extensive evaluation, KBL is demonstrated to reduce waste by up to 58% of data usage compared to state of the art short video strategies without incurring significant QoE degradation, even in the face of shifting user behavior.

Network throughput tests, commonly known as “speed tests” are widely used by consumers, regulators, and ISPs to measure and diagnose network performance. However, the tools used to conduct these tests are often costly in terms of data consumption. Moreover, the speed tests rely on data that is transferred to clients for the sole purpose of measuring throughput with the data being discarded and serving no other purpose. In this paper, we present rePurpose, a system that moves useful content (ads) to enable periodic speed tests by significantly offsetting the cost of network measurement, thereby avoiding harming user QoE. rePurpose can work within the existing ad ecosystem to pre-stage ads needed by users ahead of time. We evaluate the efficacy of rePurpose by emulating a common scenario where users watch videos and ads. Our evaluation shows that rePurpose can reduce the data cost of periodic speed tests by up to 90%. Moreover, by virtue of the time-shifted delivery courtesy of the periodic speed tests moving useful data, rePurpose improves video and ad QoE by reducing or eliminating start-up delay by up to five seconds.

Active network probing, commonly known as a speed test, is the prevalent network speed measurement and diagnostic method. Speed tests primarily measure achievable throughput by conducting bulk downloads that saturate the bottleneck link. However, the impact of speed tests on user Quality of Experience (QoE) has not been thoroughly explored. In this paper, we investigate the effects of active network probing on user QoE during two common activities: file downloading and video streaming, focusing on key QoE metrics such as download time, video bitrate, and buffering. Our analysis reveals that the standard speed test significantly extends download times (by up to 88% in WiFi and 46% in cellular networks) and adversely affects various video QoE metrics, particularly bitrate, resulting in an average bitrate reduction ranging from 46% to 60%. Moreover, we assess the outcomes of typical speed test scenarios, such as single and double tests, and establish that both variants impair QoE, with double tests causing greater disruptions. Our findings offer a comprehensive insight into the ramifications of active network probing on user applications and emphasize the necessity for approaches to alleviate its detrimental effects on QoE.

Short videos have recently emerged as a popular form of short-duration User Generated Content (UGC) within modern social media. Short video content is generally less than a minute long and predominantly produced in vertical orientation on smartphones. While still fundamentally being streaming, short video delivery is distinctly characterized by the deployment of a mechanism that pre-loads ahead of user request. Background pre-loading aims to eliminate start-up time, which is now prioritized higher in Quality of Experience (QoE) objectives, given that the application design facilitates instant 'swiping' to the next video in a recommended sequence. In this work, we provide a comprehensive comparison of four popular short video services. In particular, we explore content characteristics and evaluate the video quality across resolutions for each service. We next characterize the pre-loading policy adopted by each service. Last, we conduct an experimental study to investigate data consumption and evaluate achieved QoE under different network scenarios and application configurations.

In the past few years, unmanned aerial vehicles (UAVs) have drastically increased in popularity both from consumer and industry perspectives. A key component towards enabling the widespread usage of UAVs is the ability to stay in near-constant communication with the drone for command and control and conveying relevant instrumentation. The usage of cellular technology, namely LTE, seems to be a natural fit for addressing coverage and Line of Sight (LoS) issues. However, there is a relative dearth of data, specifically open source data that explores key performance aspects of cellular at altitudes typically envisioned for commercial UAV operation. The key contribution of this paper is to analyze data taken from numerous drone flights that include varying altitudes, locations, and multiple cellular carriers as recorded in a medium-sized Midwestern city. Further, we offer our data as an open-source repository for the community offering multiple vantage points for the various runs including the operating system, chipset (through MobileInsight), drone instrumentation, and server-side packet captures as part of the recorded data streams.

Imitation learning holds the promise to address challenging robotic tasks such as autonomous navigation. It however requires a human supervisor to oversee the training process and send correct control commands to robots without feedback, which is always prone to error and expensive. To minimize human involvement and avoid manual labeling of data in the robotic autonomous navigation with imitation learning, this paper proposes a novel semi-supervised imitation learning solution based on a multi-sensory design. This solution includes a suboptimalsensor policybased on sensor fusion to automatically label states encountered by a robot to avoid human supervision during training. In addition, arecording policyis developed to throttle the adversarial affect of learning too much from the suboptimal sensor policy. As a result, this solution allows the robot to learn a navigation policy in a self-supervised manner without human intervention after the initial data collection. With extensive experiments in indoor environments, this solution can achieve near human performance in most of the tasks and even surpasses human performance in case of unexpected events such as hardware failures or human operation errors. To best of our knowledge, this is the first work that synthesizes sensor fusion and imitation learning to enable robotic autonomous navigation in the real world without human supervision.

Video streaming using WiFi networks poses the challenge of variable network performance when multiple clients are present. Hence, it is important to continuously monitor and predict the network changes in order to ensure a higher user quality of experience (QoE) for video streaming. Existing approaches that aim to detect such network changes have several disadvantages. For example, active probing approaches are expensive so that generate more additional traffic flow during the testing. To overcome its shortcomings, we propose a passive, lightweight approach, CP-DASH, whereby queuing effects present in frame aggregation are leveraged to predict link congestion in the WiFi network. This approach allows the early detection which can be used to adapt our video appropriately. We conduct experiments simulating a WiFi network with multiple clients and compare CP-DASH with five contemporary rate selection mechanisms. We found that our proposed method significantly reduces the switch rates and stall rates from 22% to 5% and from 38% to 25% compared with an existing throughput-based algorithm, respectively.

Indoor human activity recognition has been extensively investigated. However, most of the solutions require sensors e.g. 9-axis IMU be equipped on human body or use image processing that presents privacy issues. This work proposes an ambient radar sensor based a solution to recognize the activities that humans normally perform in indoor environments. This solution uses a 7.8 GHz radar to emit 16 pulse signals every second and samples the reflected signals at 128 KHz to capture the fine dynamics of human activities. This solution designs a set of data preprocessing algorithms, including a data refining algorithm to filter outlier data, a contrastive divergence algorithm to remove background static reflection, and a transformation algorithm to convert the signal data into feature- rich spatial location changes. This solution also develops schemes to separate a collection of various activities into individuals. A lowpass frequency filter is designed to remove unwanted noisy data and the motion intensity is used to classify the activities into two high-level groups. It uses a slope-based approach and a k- means clustering to further finely recognize each activity. This solution has been extensively evaluated in a spacious research lab room and shows outstanding accuracy.

The world has entered into a “smart” era. One area becoming smart is the place where we live-homes. Smart homes are expected to be equipped with numerous sensors to continually monitor, sense, and actuate the space. The data from these sensors can be used to provide various types of services by automating common tasks while causing minimal disruption to daily life. In order to provide these services, a system must have sufficient intelligence to predict future events based on its observations. This paper first examines the requirements for smart home predictions. It then comprehensively reviews prediction algorithms and variations that have been proposed and investigated in smart environments, such as smart homes. It is these prediction algorithms that provide the intelligence required by a smart home. Comparisons are also made upon these prediction algorithms on their features and models.