Style-Controllable Speech-Driven Gesture Synthesis Using Normalising Flows

Automatic synthesis of realistic gestures promises to transform the fields of animation, avatars and communicative agents. In off-line applications, novel tools can alter the role of an animator to that of a director, who provides only high-level input for the desired animation; a learned network then translates these instructions into an appropriate sequence of body poses. In interactive scenarios, systems for generating natural animations on the fly are key to achieving believable and relatable characters. In this paper we address some of the core issues towards these ends. By adapting a deep learning-based motion synthesis method called MoGlow, we propose a new generative model for generating state-of-the-art realistic speech-driven gesticulation. Owing to the probabilistic nature of the approach, our model can produce a battery of different, yet plausible, gestures given the same input speech signal. Just like humans, this gives a rich natural variation of motion. We additionally demonstrate the ability to exert directorial control over the output style, such as gesture level, speed, symmetry and spacial extent. Such control can be leveraged to convey a desired character personality or mood. We achieve all this without any manual annotation of the data. User studies evaluating upper-body gesticulation confirm that the generated motions are natural and well match the input speech. Our method scores above all prior systems and baselines on these measures, and comes close to the ratings of the original recorded motions. We furthermore find that we can accurately control gesticulation styles without unnecessarily compromising perceived naturalness. Finally, we also demonstrate an application of the same method to full-body gesticulation, including the synthesis of stepping motion and stance.     

MIMO Channel Capacity and Configuration Selection for Switched Parasitic Antennas

Multiple-input multiple-output (MIMO) systems offer significant enhancements in terms of their data rate and channel capacity compared to traditional systems. However, correlation degrades the system performance and imposes practical limits on the number of antennas that can be incorporated into portable wireless devices. The use of switched parasitic antennas (SPAs) is a possible solution, especially where it is difficult to obtain sufficient signal decorrelation by conventional means. The covariance matrix represents the correlation present in the propagation channel, and has significant impact on the MIMO channel capacity. The results of this work demonstrate a significant improvement in the MIMO channel capacity by using SPA with the knowledge of the covariance matrix for all pattern configurations. By employing the “water-pouring algorithm” to modify the covariance matrix, the channel capacity is significantly improved compared to traditional systems, which spread transmit power uniformly across all the antennas. A condition number is also proposed as a selection metric to select the optimal pattern configuration for MIMO-SPAs.     

miR-16-5p Is a Stably-Expressed Housekeeping MicroRNA in Breast Cancer Tissues from Primary Tumors and from Metastatic Sites

For quantitative microRNA analyses in formalin-fixed paraffin-embedded (FFPE) tissue, expression levels have to be normalized to endogenous controls. To investigate the most stably-expressed microRNAs in breast cancer and its surrounding tissue, we used tumor samples from primary tumors and from metastatic sites. MiRNA profiling using TaqMan^® Array Human MicroRNA Cards, enabling quantification of 754 unique human miRNAs, was performed in FFPE specimens from 58 patients with metastatic breast cancer. Forty-two (72%) samples were collected from primary tumors and 16 (28%) from metastases. In a cross-platform analysis of a validation cohort of 32 FFPE samples from patients with early breast cancer genome-wide microRNA expression analysis using SurePrintG3 miRNA (8 × 60 K)^® microarrays from Agilent^® was performed. Eleven microRNAs could be detected in all samples analyzed. Based on NormFinder and geNorm stability values and the high correlation (rho ≥ 0.8) with the median of all measured microRNAs, miR-16-5p, miR-29a-3p, miR-126-3p, and miR-222-3p are suitable single gene housekeeper candidates. In the cross-platform validation, 29 human microRNAs were strongly expressed (mean log2-intensity > 10) and 21 of these microRNAs including miR-16-5p and miR-29a-3p were also stably expressed (CV < 5%). Thus, miR-16-5p and miR-29a-3p are both strong housekeeper candidates. Their Normfinder stability values calculated across the primary tumor and metastases subgroup indicate that miR-29a-3p can be considered as the strongest housekeeper in a cohort with mainly samples from primary tumors, whereas miR-16-5p might perform better in a metastatic sample enriched cohort.     

A Multiple Error-Correction Model of Housing Supply

This paper considers supply dynamics in the context of the Irish residential market. The analysis, in a multiple error-correction framework, reveals that although developers did respond to disequilibrium in supply, the rate of adjustment was relatively slow. In contrast, however, disequilibrium in demand did not impact upon supply, suggesting that inelastic supply conditions could explain the prolonged nature of the boom in the Irish market. Increased elasticity in the later stages of the boom may have been a contributory factor in the extent of the house price falls observed in recent years.     

Field D* path-finding on weighted triangulated and tetrahedral meshes

Classic shortest path algorithms operate on graphs, which are suitable for problems that can be represented by weighted nodes or edges. Finding a shortest path through a set of weighted regions is more difficult and only approximate solutions tend to scale well. The Field D* algorithm efficiently calculates an approximate, interpolated shortest path through a set of weighted regions and was designed for navigating robots through terrains with varying characteristics. Field D* operates on unit grid or quad-tree data structures, which require high resolutions to accurately model the boundaries of irregular world structures. In this paper, we extend the Field D* cost functions to 2D triangulations and 3D tetrahedral meshes: structures which model polygonal world structures more accurately. Since robots typically have limited resources available for computation and storage, we pay particular attention to computation and storage overheads when detailing our extensions. We begin by providing analytic solutions to the minimum of each cost function for 2D triangles and 3D tetrahedra. Our triangle implementation provides a 50 % improvement in performance over an existing triangle implementation. While our 3D extension to tetrahedra is the first full analytic extension of Field D* to 3D, previous work only provided an approximate minimization for a single cost function on a 3D cube with unit lengths. Each cost function is expressed in terms of a general function whose characteristics can be exploited to reduce the calculations required to find a minimum. These characteristics can also be exploited to cache the majority of cost functions, producing a speedup of up to 28 % in the 3D tetrahedral case. We demonstrate that, in environments composed of non-grid aligned data, Multi-resolution quad-tree Field D* requires an order of magnitude more faces and between 15 and 20 times more node expansions, to produce a path of similar cost to one produced by a triangle implementation of Field D* on a lower resolution triangulation. We provide examples of 3D pathing through models of complex topology, including pathing through anatomical structures extracted from a medical data set. To summarise, this paper details a robust and efficient extension of Field D* pathing to data sets represented by weighted triangles and tetrahedra, and also provides empirical data which demonstrates the reduction in storage and computation costs that accrue when one chooses such a representation over the more commonly used quad-tree and grid-based alternatives.     

Situation identification techniques in pervasive computing: A review

Pervasive systems must offer an open, extensible, and evolving portfolio of services which integrate sensor data from a diverse range of sources. The core challenge is to provide appropriate and consistent adaptive behaviours for these services in the face of huge volumes of sensor data exhibiting varying degrees of precision, accuracy and dynamism. Situation identification is an enabling technology that resolves noisy sensor data and abstracts it into higher-level concepts that are interesting to applications. We provide a comprehensive analysis of the nature and characteristics of situations, discuss the complexities of situation identification, and review the techniques that are most popularly used in modelling and inferring situations from sensor data. We compare and contrast these techniques, and conclude by identifying some of the open research opportunities in the area.