Neural video codecs have recently become competitive with standard codecs such as HEVC in the low-delay setting. However, most neural codecs are large floating-point networks that use pixel-dense warping operations for temporal modeling, making them too computationally expensive for deployment on mobile devices. Recent work has demonstrated that running a neural decoder in real time on mobile is feasible, but shows this only for 720p RGB video.This work presents the first neural video codec that decodes 1080p YUV420 video in real time on a mobile device. Our codec relies on two major contributions. First, we design an efficient codec that uses a block-based motion compensation algorithm available on the warping core of the mobile accelerator, and we show how to quantize this model to integer precision. Second, we implement a fast decoder pipeline that concurrently runs neural network components on the neural signal processor, parallel entropy coding on the mobile GPU, and warping on the warping core. Our codec outperforms the previous on-device codec by a large margin with up to 48 % BD-rate savings, while reducing the MAC count on the receiver side by 10×. We perform a careful ablation to demonstrate the effect of the introduced motion compensation scheme, and ablate the effect of model quantization.

This demonstration showcases our innovations on efficient, accurate, and temporally consistent video semantic segmentation on mobile device. We employ our test-time unsupervised scheme, AuxAdapt, to enable the segmentation model to adapt to a given video in an online manner. More specifically, we leverage a small auxiliary network to perform weight updates and keep the large, main segmen-tation network frozen. This significantly reduces the computational cost of adaptation when compared to previous methods (e.g., Tent, DVP), and at the same time, prevents catastrophic forgetting. By running AuxAdapt, we can considerably improve the temporal consistency of video segmentation while maintaining the accuracy. We demonstrate how to efficiently deploy our adaptive video segmentation algorithm on a smartphone powered by a Snapdragon® Mobile Platform11Snapdragon is a product of Qualcomm Technologies, Inc. and/or its subsidiaries., Rather than simply running the entire algorithm on the GPU, we adopt a crossunit deployment strategy. The main network, which will be frozen during test time, will perform inferences on a highly optimized AI accelerator unit, while the small auxiliary net-work, which will be updated on the fly, will run forward passes and back-propagations on the GPU. Such a deployment scheme best utilizes the available processing power on the smartphone and enables real-time operation of our adaptive video segmentation algorithm. We provide example videos in supplementary material.

The metabolic syndrome is one of the major public health challenges worldwide. Prevalence of metabolic syndrome vastly increases the risk of type 2 diabetes and cardiovascular diseases (CVDs). Metabolic Syndrome in general is under-diagnosed and often goes undetected for years. In this paper we present a machine learning based method for early detection of metabolic syndrome which uses only non-invasive features. We train and test our model based on data collected from a German population consisting of 2314 subjects (male = 918, female = 1396). Out of 2314 subjects 941 were diagnosed with metabolic syndrome (male = 441, female = 500). Features we consider include different anthropometric features (such as height, weight, waist circumference), medications, age, gender etc.; machine learning techniques we employed included gradient boosting machines, random forest, logistic regression and an ensemble model. We compare our models against the ones that were proposed in previous literature and outperform them in our cohort. We achieve area under the curve values (AUCs) of up to 0.90 with the ensemble classifier. The results achieved suggest that machine learning can be a valuable tool to predict metabolic syndrome with high discriminative power without relying on any invasive bio-markers, which significantly facilitates early detection.

Low back pain (LBP) is a leading cause of activity limitation. Objective assessment of the spinal motion plays a key role in diagnosis and treatment of LBP. We propose a method that facilitates clinical assessment of lower back motions by means of a wireless inertial sensor network. The sensor units are attached to the right and left side of the lumbar region, the pelvis and the thighs, respectively. Since magnetometers are known to be unreliable in indoor environments, we use only 3D accelerometer and 3D gyroscope readings. Compensation of integration drift in the horizontal plane is achieved by estimating the gyroscope biases from automatically detected initial rest phases. For the estimation of sensor orientations, both a smoothing algorithm and a filtering algorithm are presented. From these orientations, we determine three-dimensional joint angles between the thighs and the pelvis and between the pelvis and the lumbar region. We compare the orientations and joint angles to measurements of an optical motion tracking system that tracks each skin-mounted sensor by means of reflective markers. Eight subjects perform a neutral initial pose, then flexion/extension, lateral flexion, and rotation of the trunk. The root mean square deviation between inertial and optical angles is about one degree for angles in the frontal and sagittal plane and about two degrees for angles in the transverse plane (both values averaged over all trials). We choose five features that characterize the initial pose and the three motions. Interindividual differences of all features are found to be clearly larger than the observed measurement deviations. These results indicate that the proposed inertial sensor-based method is a promising tool for lower back motion assessment.

For test application, data path circuits are divided into subcircuits. At the inputs and outputs of every subcircuit, accumulators to generate patterns and to compact test responses are configured from the other available modules. Test registers are inserted only at positions where appropriate paths from and to accumulators do not exist. A small set of test configurations is chosen so that all the modules of the data path are tested and the hardware overhead is as low as possible. For each test configuration, we construct a test schedule with minimum test application time. The presented scheduling approach is based on loop folding and uses an integer linear programming formulation. The scheduling procedure always finds a schedule with maximum throughput which reduces test application times significantly.

In recent years, accumulators have been shown to be efficient pattern generators and response compactors for built-in self-test. Many circuits contain modules which can be configured as accumulators just by controlling these modules adequately. This paper presents algorithms that find all possible accumulator configurations in a circuit and optimize the control of the accumulator test application or or inexpensive test control implementation.

Accumulators based on addition or subtraction can be used as test pattern generators. Some circuits, however, require long test lengths if the parameters of the accumulator are not properly adapted. This paper presents two different methods to minimize the test length without sacrificing fault coverage. The simulation-based reseeding method is suited to random pattern testable circuits and uses forward and reverse order simulation to skip ineffective patterns. The analytical method is appropriate for circuits with "hard" faults that are detected only by few test patterns. This method searches for an optimal input value of the accumulator and calculates the best seed analytically. The results show significant test length reductions. The proposed pattern generators can be implemented very efficiently in hardware using available blocks of a data path or in software using an embedded processor.

The authors describe how mandatory and discretionary access mediation are performed in the trusted mach (TMach) kernel, a system that uses message passing as its primary means of communication both between tasks and with the kernel. As a consequence, control of interprocess communication in the TMach kernel is a central concern whereas controlled sharing of segments is the central focus in trusted systems with more traditional architectures. The TMach kernel is not a complete trusted system. It is a reference monitor of basic system abstractions, providing a small, well-controlled base on which the rest of a trusted computing base and operating system can be constructed. The TMach kernel provides simple and elegant mandatory access control for port access rights. The TMach kernel's MAC (mandatory access control) mechanisms clearly control the flow of information according to a mandatory security policy based on a Bell and La Padula model. DAC (discretionary access control) mechanisms are provided in the TMach kernel to implement TCSEC (trusted computer system evaluation criteria) requirements and to support DAC in servers to be built on the kernel.>