Methods to accelerate the learning of bayesian network structures

R. Daly and Q. Shen. Methods to accelerate the learning of bayesian network structures. Proceedings of the Proceedings of the 2007 UK Workshop on Computational Intelligence.

Exploiting Phonological Constraints for Handshape Inference in ASL Video

Handshape is a key articulatory parameter in sign language, and thus handshape recognition from signing video is essential for sign recognition and retrieval. Handshape transitions within monomorphemic lexical signs (the largest class of signs in signed languages) are governed by phonological rules. For example, such transitions normally involve either closing or opening of the hand (i.e., to exclusively use either folding or unfolding of the palm and one or more fingers). Furthermore, akin to allophonic variations in spoken languages, both inter- and intra- signer variations in the production of specific handshapes are observed. We propose a Bayesian network formulation to exploit handshape co-occurrence constraints, also utilizing information about allophonic variations to aid in handshape recognition. We propose a fast non-rigid image alignment method to gain improved robustness to handshape appearance variations during computation of observation likelihoods in the Bayesian network. We evaluate our handshape recognition approach on a large dataset of monomorphemic lexical signs. We demonstrate that leveraging linguistic constraints on handshapes results in improved handshape recognition accuracy. As part of the overall project, we are collecting and preparing for dissemination a large corpus (three thousand signs from three native signers) of American Sign Language (ASL) video. The video have been annotated using SignStream® [Neidle et al.] with labels for linguistic information such as glosses, morphological properties and variations, and start/end handshapes associated with each ASL sign.

An alignment based similarity measure for hand detection in cluttered sign language video

Locating hands in sign language video is challenging due to a number of factors. Hand appearance varies widely across signers due to anthropometric variations and varying levels of signer proficiency. Video can be captured under varying illumination, camera resolutions, and levels of scene clutter, e.g., high-res video captured in a studio vs. low-res video gathered by a web cam in a user’s home. Moreover, the signers’ clothing varies, e.g., skin-toned clothing vs. contrasting clothing, short-sleeved vs. long-sleeved shirts, etc. In this work, the hand detection problem is addressed in an appearance matching framework. The Histogram of Oriented Gradient (HOG) based matching score function is reformulated to allow non-rigid alignment between pairs of images to account for hand shape variation. The resulting alignment score is used within a Support Vector Machine hand/not-hand classifier for hand detection. The new matching score function yields improved performance (in ROC area and hand detection rate) over the Vocabulary Guided Pyramid Match Kernel (VGPMK) and the traditional, rigid HOG distance on American Sign Language video gestured by expert signers. The proposed match score function is computationally less expensive (for training and testing), has fewer parameters and is less sensitive to parameter settings than VGPMK. The proposed detector works well on test sequences from an inexpert signer in a non-studio setting with cluttered background.

Combinations of Deformable Shape Prototypes

We propose to investigate a model-based technique for encoding non-rigid object classes in terms of object prototypes. Objects from the same class can be parameterized by identifying shape and appearance invariants of the class to devise low-level representations. The approach presented here creates a flexible model for an object class from a set of prototypes. This model is then used to estimate the parameters of low-level representation of novel objects as combinations of the prototype parameters. Variations in the object shape are modeled as non-rigid deformations. Appearance variations are modeled as intensity variations. In the training phase, the system is presented with several example prototype images. These prototype images are registered to a reference image by a finite element-based technique called Active Blobs. The deformations of the finite element model to register a prototype image with the reference image provide the shape description or shape vector for the prototype. The shape vector for each prototype, is then used to warp the prototype image onto the reference image and obtain the corresponding texture vector. The prototype texture vectors, being warped onto the same reference image have a pixel by pixel correspondence with each other and hence are "shape normalized". Given sufficient number of prototypes that exhibit appropriate in-class variations, the shape and the texture vectors define a linear prototype subspace that spans the object class. Each prototype is a vector in this subspace. The matching phase involves the estimation of a set of combination parameters for synthesis of the novel object by combining the prototype shape and texture vectors. The strengths of this technique lie in the combined estimation of both shape and appearance parameters. This is in contrast with the previous approaches where shape and appearance parameters were estimated separately.

Broadband electromagnetic cloaking with smart metamaterials.

The ability to render objects invisible with a cloak that fits all objects and sizes is a long-standing goal for optical devices. Invisibility devices demonstrated so far typically comprise a rigid structure wrapped around an object to which it is fitted. Here we demonstrate smart metamaterial cloaking, wherein the metamaterial device not only transforms electromagnetic fields to make an object invisible, but also acquires its properties automatically from its own elastic deformation. The demonstrated device is a ground-plane microwave cloak composed of an elastic metamaterial with a broad operational band (10-12 GHz) and nearly lossless electromagnetic properties. The metamaterial is uniform, or perfectly periodic, in its undeformed state and acquires the necessary gradient-index profile, mimicking a quasi-conformal transformation, naturally from a boundary load. This easy-to-fabricate hybrid elasto-electromagnetic metamaterial opens the door to implementations of a variety of transformation optics devices based on quasi-conformal maps.

Differential lifetime success and performance of native and non-native Atlantic salmon examined under communal natural conditions.

The lifetime success and performance characteristics of communally reared offspring of wild native Burrishoole (native), ranched native (ranched) and non-native (non-native) Atlantic salmon Salmo salar from the adjacent Owenmore River were compared. Non-native year parr showed a substantial downstream migration, which was not shown by native and ranched parr. This appears to have been an active migration rather than competitive displacement and may reflect an adaptation to environmental or physiographic conditions within the Owenmore River catchment where the main nursery habitat is downstream of the spawning area. There were no differences between native and ranched in smolt output or adult return. Both of these measures, however, were significantly lower for the non-native group. A greater proportion of the non-native Atlantic salmon was taken in the coastal drift nets compared to the return to the Burrishoole system, probably as a result of the greater size of the non-native fish. The overall lifetime success of the non-native group, from fertilized egg to returning adult, was some 35% of native and ranched. The ranched group showed a significantly greater male parr maturity, a greater proportion of 1+ year smolts, and differences in sex ratio and timing of freshwater entry of returning adults compared to native, which may have fitness implications under specific conditions.

Hylambatin and (Thr)11-hylambatin from the skin secretion of the African hyperoliid frog, Kassina maculata: Structural characterisation, precursor cDNA cloning and preliminary pharmacological testing

The tachykinins hylambatin and (Thr)11-hylambatin have been isolated from the defensive skin secretion of the African hyperoliid frog, Kassina maculata,. Hylambatin (DPPDPNRFYGMMamide) is revised in structure from the original sequence by a single site substitution (Asn/Asp at position 6), and (Thr)11-hylambatin, a novel tachykinin, differs in structure from hylambatin by a single Thr/Met substitution. (Thr)11-hylambatin is five- to ten-fold more abundant than hylambatin in secretions. Synthetic replicates of both peptides were active in smooth muscle preparations including the rat tail artery, rat ileum and bovine trachea. While hylambatin displayed activity consistent with an NK1-receptor ligand, (Thr)11-hylambatin was more active than either substance P or neurokinin A in both NK1- and NK-2 receptor rich preparations. Incorporation of a threoninyl residue rather than the canonical leucyl residue at the penultimate position in both substance P and neurokinin A, generated active ligands in both arterial and intestinal smooth muscle preparations. Hylambatin precursor cDNAs, designated HYBN-1 and HYBN-2, respectively, were cloned from a skin library by 3'- and 5'-RACE reactions. Both were highly-homologous containing open-reading frames of 66 amino acids encoding single copies of either hylambatin or (Thr)11-hylambatin. These data reveal a hitherto unrecognized structure/activity attribute of mammalian tachykinin receptors revealed though discovery of a novel amphibian skin-derived, site-substituted peptide ligand.

Behaviours in video

With a significant increment of the number of digital cameras used for various purposes, there is a demanding call for advanced video analysis techniques that can be used to systematically interpret and understand the semantics of video contents, which have been recorded in security surveillance, intelligent transportation, health care, video retrieving and summarization. Understanding and interpreting human behaviours based on video analysis have observed competitive challenges due to non-rigid human motion, self and mutual occlusions, and changes of lighting conditions. To solve these problems, advanced image and signal processing technologies such as neural network, fuzzy logic, probabilistic estimation theory and statistical learning have been overwhelmingly investigated.

Kruppel-associated Box (KRAB)-associated co-repressor (KAP-1) Ser-473 phosphorylation regulates heterochromatin protein 1β (HP1-β) mobilization and DNA repair in heterochromatin

The DNA damage response encompasses a complex series of signaling pathways that function to regulate and facilitate the repair of damaged DNA. Recent studies have shown that the repair of transcriptionally inactive chromatin, named heterochromatin, is dependent upon the phosphorylation of the co-repressor, Krüppel-associated box (KRAB) domain-associated protein (KAP-1), by the ataxia telangiectasia-mutated (ATM) kinase. Co-repressors, such as KAP-1, function to regulate the rigid structure of heterochromatin by recruiting histone-modifying enzymes, such HDAC1/2, SETDB1, and nucleosome-remodeling complexes such as CHD3. Here, we have characterized a phosphorylation site in the HP1-binding domain of KAP-1, Ser-473, which is phosphorylated by the cell cycle checkpoint kinase Chk2. Expression of a nonphosphorylatable S473A mutant conferred cellular sensitivity to DNA-damaging agents and led to defective repair of DNA double-strand breaks in heterochromatin. In addition, cells expressing S473A also displayed defective mobilization of the HP1-ß chromodomain protein. The DNA repair defect observed in cells expressing S473A was alleviated by depletion of HP1-ß, suggesting that phosphorylation of KAP-1 on Ser-473 promotes the mobilization of HP1-ß from heterochromatin and subsequent DNA repair. These results suggest a novel mechanism of KAP-1-mediated chromatin restructuring via Chk2-regulated HP1-ß exchange from heterochromatin, promoting DNA repair.

Protocolized versus non-protocolized weaning for reducing the duration of mechanical ventilation in critically ill adult patients (Review)

BACKGROUND: Reducing weaning time is desirable in minimizing potential complications from mechanical ventilation. Standardized weaning protocols are purported to reduce time spent on mechanical ventilation. However, evidence supporting their use in clinical practice is inconsistent.

OBJECTIVES: To assess the effects of protocolized weaning from mechanical ventilation on the total duration of mechanical ventilation for critically ill adults; ascertain differences between protocolized and non-protocolized weaning in terms of mortality, adverse events, quality of life, weaning duration, intensive care unit (ICU) and hospital length of stay (LOS); and explore variation in outcomes by type of ICU, type of protocol and approach to delivering the protocol.

SEARCH STRATEGY: We searched the Cochrane Central Register of Controlled Trials (The Cochrane Library Issue 1, 2010), MEDLINE (1950 to 2010), EMBASE (1988 to 2010), CINAHL (1937 to 2010), LILACS (1982 to 2010), ISI Web of Science and ISI Conference Proceedings (1970 to 2010), Cambridge Scientific Abstracts (inception to 2010) and reference lists of articles. We did not apply language restrictions.

SELECTION CRITERIA: We included randomized and quasi-randomized controlled trials of protocolized weaning versus non-protocolized weaning from mechanical ventilation in critically ill adults.

DATA COLLECTION AND ANALYSIS: Three authors independently assessed trial quality and extracted data. A priori subgroup and sensitivity analyses were performed. We contacted study authors for additional information.

MAIN RESULTS: Eleven trials that included 1971 patients met the inclusion criteria. The total duration of mechanical ventilation geometric mean in the protocolized weaning group was on average reduced by 25% compared with the usual care group (N = 10 trials, 95% CI 9% to 39%, P = 0.006); weaning duration was reduced by 78% (N = 6 trials, 95% CI 31% to 93%, P = 0.009); and ICU LOS by 10% (N = 8 trials, 95% CI 2% to 19%, P = 0.02). There was significant heterogeneity among studies for total duration of mechanical ventilation (I(2) = 76%, P <0.01) and weaning duration (I(2) = 97%, P <0.01), which could not be explained by subgroup analyses based on type of unit or type of approach.

AUTHORS' CONCLUSIONS: There is some evidence of a reduction in the duration of mechanical ventilation, weaning duration and ICU LOS with use of standardized protocols, but there is significant heterogeneity among studies and an insufficient number of studies to investigate the source of this heterogeneity. Although some study authors suggest that organizational context may influence outcomes, these factors were not considered in all included studies and therefore could not be evaluated.

The Forgiving Graph: a distributed data structure for low stretch under adversarial attack

We consider the problem of self-healing in peer-to-peer networks that are under repeated attack by an omniscient adversary. We assume that, over a sequence of rounds, an adversary either inserts a node with arbitrary connections or deletes an arbitrary node from the network. The network responds to each such change by quick “repairs,” which consist of adding or deleting a small number of edges. These repairs essentially preserve closeness of nodes after adversarial deletions, without increasing node degrees by too much, in the following sense. At any point in the algorithm, nodes v and w whose distance would have been l in the graph formed by considering only the adversarial insertions (not the adversarial deletions), will be at distance at most l log n in the actual graph, where n is the total number of vertices seen so far. Similarly, at any point, a node v whose degree would have been d in the graph with adversarial insertions only, will have degree at most 3d in the actual graph. Our distributed data structure, which we call the Forgiving Graph, has low latency and bandwidth requirements. The Forgiving Graph improves on the Forgiving Tree distributed data structure from Hayes et al. (2008) in the following ways: 1) it ensures low stretch over all pairs of nodes, while the Forgiving Tree only ensures low diameter increase; 2) it handles both node insertions and deletions, while the Forgiving Tree only handles deletions; 3) it requires only a very simple and minimal initialization phase, while the Forgiving Tree initially requires construction of a spanning tree of the network.

The Forgiving Graph:A distributed data structure for low stretch under adversarial attack

We consider the problem of self-healing in peer-to-peer networks that are under repeated attack by an omniscient adversary. We assume that, over a sequence of rounds, an adversary either inserts a node with arbitrary connections or deletes an arbitrary node from the network. The network responds to each such change by quick "repairs," which consist of adding or deleting a small number of edges. These repairs essentially preserve closeness of nodes after adversarial deletions,without increasing node degrees by too much, in the following sense. At any point in the algorithm, nodes v and w whose distance would have been - in the graph formed by considering only the adversarial insertions (not the adversarial deletions), will be at distance at most - log n in the actual graph, where n is the total number of vertices seen so far. Similarly, at any point, a node v whose degreewould have been d in the graph with adversarial insertions only, will have degree at most 3d in the actual graph. Our distributed data structure, which we call the Forgiving Graph, has low latency and bandwidth requirements. The Forgiving Graph improves on the Forgiving Tree distributed data structure from Hayes et al. (2008) in the following ways: 1) it ensures low stretch over all pairs of nodes, while the Forgiving Tree only ensures low diameter increase; 2) it handles both node insertions and deletions, while the Forgiving Tree only handles deletions; 3) it requires only a very simple and minimal initialization phase, while the Forgiving Tree initially requires construction of a spanning tree of the network. © Springer-Verlag 2012.

Protocolized versus non-protocolized weaning for reducing the duration of mechanical ventilation in critically ill adult patients

Background: This is an update of a review last published in Issue 5, 2010, of The Cochrane Library. Reducing weaning time is desirable in minimizing potential complications from mechanical ventilation. Standardized weaning protocols are purported to reduce time spent on mechanical ventilation. However, evidence supporting their use in clinical practice is inconsistent. Objectives: The first objective of this review was to compare the total duration of mechanical ventilation of critically ill adults who were weaned using protocols versus usual (non-protocolized) practice.The second objective was to ascertain differences between protocolized and non-protocolized weaning in outcomes measuring weaning duration, harm (adverse events) and resource use (intensive care unit (ICU) and hospital length of stay, cost).The third objective was to explore, using subgroup analyses, variations in outcomes by type of ICU, type of protocol and approach to delivering the protocol (professional-led or computer-driven). Search methods: We searched the Cochrane Central Register of Controlled Trials (CENTRAL) (The Cochrane Library Issue 1, 2014), MEDLINE (1950 to January 2014), EMBASE (1988 to January 2014), CINAHL (1937 to January 2014), LILACS (1982 to January 2014), ISI Web of Science and ISI Conference Proceedings (1970 to February 2014), and reference lists of articles. We did not apply language restrictions. The original search was performed in January 2010 and updated in January 2014.Selection criteriaWe included randomized controlled trials (RCTs) and quasi-RCTs of protocolized weaning versus non-protocolized weaning from mechanical ventilation in critically ill adults. Data collection and analysis: Two authors independently assessed trial quality and extracted data. We performed a priori subgroup and sensitivity analyses. We contacted study authors for additional information. Main results: We included 17 trials (with 2434 patients) in this updated review. The original review included 11 trials. The total geometric mean duration of mechanical ventilation in the protocolized weaning group was on average reduced by 26% compared with the usual care group (N = 14 trials, 95% confidence interval (CI) 13% to 37%, P = 0.0002). Reductions were most likely to occur in medical, surgical and mixed ICUs, but not in neurosurgical ICUs. Weaning duration was reduced by 70% (N = 8 trials, 95% CI 27% to 88%, P = 0.009); and ICU length of stay by 11% (N = 9 trials, 95% CI 3% to 19%, P = 0.01). There was significant heterogeneity among studies for total duration of mechanical ventilation (I2 = 67%, P < 0.0001) and weaning duration (I2 = 97%, P < 0.00001), which could not be explained by subgroup analyses based on type of unit or type of approach. Authors' conclusions: There is evidence of reduced duration of mechanical ventilation, weaning duration and ICU length of stay with use of standardized weaning protocols. Reductions are most likely to occur in medical, surgical and mixed ICUs, but not in neurosurgical ICUs. However, significant heterogeneity among studies indicates caution in generalizing results. Some study authors suggest that organizational context may influence outcomes, however these factors were not considered in all included studies and could not be evaluated. Future trials should consider an evaluation of the process of intervention delivery to distinguish between intervention and implementation effects. There is an important need for further development and research in the neurosurgical population.

Coupled Effects of a Perturbation in a Complex Structure Observed with SDO/AIA, SDO/HMI and ROSA/HARDcam

We study properties of intensity fluctuations in NOAA Active Region 11250 observed on 13 July 2011 starting at UT 13:32. Included are data obtained in the EUV bands of the Atmospheric Imaging Assembly on board the Solar Dynamics Observatory (SDO/AIA) as well as nearly simultaneous observations of the chromosphere made, at much higher spatial and temporal resolution, with the Rapid Oscillations in the Solar Atmosphere (ROSA) and Hydrogen-Alpha Rapid Dynamics camera (HARDcam) systems at the Dunn Solar Telescope. A complex structure seen in both the ROSA/HARDcam and SDO data sets comprises a system of loops extending outward from near the boundary of the leading sunspot umbra. It is visible in the ROSA Ca II K and HARDcam Hα images, as well as the SDO 304 Å, 171 Å and 193 Å channels, and it thus couples the chromosphere, transition region and corona. In the ground-based images the loop structure is 4.1 Mm long. Some 17.5 Mm, can be traced in the SDO/AIA data. The chromospheric emissions observed by ROSA and HARDcam appear to occupy the inner, and apparently cooler and lower, quarter of the loop. We compare the intensity fluctuations of two points within the structure. From alignment with SDO/HMI images we identify a point "A" near the loop structure, which sits directly above a bipolar magnetic feature in the photosphere. Point "B" is characteristic of locations within the loops that are visible in both the ROSA/HARDcam and the SDO/AIA data. The intensity traces for point A are quiet during the first part of the data string. At time ~ 19 min they suddenly begin a series of impulsive brightenings. In the 171 Å and 193 Å coronal lines the brightenings are localized impulses in time, but in the transition region line at 304 Å they are more extended in time. The intensity traces in the 304 Å line for point B shows a quasi-periodic signal that changes properties at about 19 min. The wavelet power spectra are characterized by two periodicities. A 6.7 min period extends from the beginning of the series until about 25 minutes, and another signal with period ~3 min starts at about 20 min. The 193 Å power spectrum has a characteristic period of 5 min, before the 20 min transition and a 2.5 min periodicity afterward. In the case of HARDcam Hα data a localized 4 min periodicity can be found until about 7 min, followed by a quiet regime. After ~20 min a 2.3 min periodicity appears. Interestingly a coronal loop visible in the 94 Å line that is centrally located in the AR, running from the leading umbra to the following polarity, at about time 20 min undergoes a strong brightening beginning at the same moment all along 15 Mm of its length. The fact that these different signals all experience a clear-cut change at time about 20 min suggests an underlying organizing mechanism. Given that point A has a direct connection to the photospheric magnetic bipole, we conjecture that the whole extended structure is connected in a complex manner to the underlying magnetic field. The periodicities in these features may favor the wave nature rather than upflows and interpretations will be discussed.

La mise en registre automatique des surfaces acquises à partir d’objets déformables

La mise en registre 3D (opération parfois appelée alignement) est un processus de transformation d’ensembles de données 3D dans un même système de coordonnées afin d’en aligner les éléments communs. Deux ensembles de données alignés ensemble peuvent être les scans partiels des deux vues différentes d’un même objet. Ils peuvent aussi être deux modèles complets, générés à des moments différents, d’un même objet ou de deux objets distincts. En fonction des ensembles de données à traiter, les méthodes d’alignement sont classées en mise en registre rigide ou non-rigide. Dans le cas de la mise en registre rigide, les données sont généralement acquises à partir d’objets rigides. Le processus de mise en registre peut être accompli en trouvant une seule transformation rigide globale (rotation, translation) pour aligner l’ensemble de données source avec l’ensemble de données cible. Toutefois, dans le cas non-rigide, où les données sont acquises à partir d’objets déformables, le processus de mise en registre est plus difficile parce qu’il est important de trouver à la fois une transformation globale et des déformations locales. Dans cette thèse, trois méthodes sont proposées pour résoudre le problème de mise en registre non-rigide entre deux ensembles de données (représentées par des maillages triangulaires) acquises à partir d’objets déformables. La première méthode permet de mettre en registre deux surfaces se chevauchant partiellement. La méthode surmonte les limitations des méthodes antérieures pour trouver une grande déformation globale entre deux surfaces. Cependant, cette méthode est limitée aux petites déformations locales sur la surface afin de valider le descripteur utilisé. La seconde méthode est s’appuie sur le cadre de la première et est appliquée à des données pour lesquelles la déformation entre les deux surfaces est composée à la fois d’une grande déformation globale et de petites déformations locales. La troisième méthode, qui se base sur les deux autres méthodes, est proposée pour la mise en registre d’ensembles de données qui sont plus complexes. Bien que la qualité que elle fournit n’est pas aussi bonne que la seconde méthode, son temps de calcul est accéléré d’environ quatre fois parce que le nombre de paramètres optimisés est réduit de moitié. L’efficacité des trois méthodes repose sur des stratégies via lesquelles les correspondances sont déterminées correctement et le modèle de déformation est exploité judicieusement. Ces méthodes sont mises en oeuvre et comparées avec d’autres méthodes sur diverses données afin d’évaluer leur robustesse pour résoudre le problème de mise en registre non-rigide. Les méthodes proposées sont des solutions prometteuses qui peuvent être appliquées dans des applications telles que la mise en registre non-rigide de vues multiples, la reconstruction 3D dynamique, l’animation 3D ou la recherche de modèles 3D dans des banques de données.