Analysis of Dependence Tracking Algorithms for Task Dataflow Execution

Processor architectures has taken a turn towards many-core processors, which integrate multiple processing cores on a single chip to increase overall performance, and there are no signs that this trend will stop in the near future. Many-core processors are harder to program than multi-core and single-core processors due to the need of writing parallel or concurrent programs with high degrees of parallelism. Moreover, many-cores have to operate in a mode of strong scaling because of memory bandwidth constraints. In strong scaling increasingly finer-grain parallelism must be extracted in order to keep all processing cores busy.

Task dataflow programming models have a high potential to simplify parallel program- ming because they alleviate the programmer from identifying precisely all inter-task de- pendences when writing programs. Instead, the task dataflow runtime system detects and enforces inter-task dependences during execution based on the description of memory each task accesses. The runtime constructs a task dataflow graph that captures all tasks and their dependences. Tasks are scheduled to execute in parallel taking into account dependences specified in the task graph.

Several papers report important overheads for task dataflow systems, which severely limits the scalability and usability of such systems. In this paper we study efficient schemes to manage task graphs and analyze their scalability. We assume a programming model that supports input, output and in/out annotations on task arguments, as well as commutative in/out and reductions. We analyze the structure of task graphs and identify versions and generations as key concepts for efficient management of task graphs. Then, we present three schemes to manage task graphs building on graph representations, hypergraphs and lists. We also consider a fourth edge-less scheme that synchronizes tasks using integers. Analysis using micro-benchmarks shows that the graph representation is not always scalable and that the edge-less scheme introduces least overhead in nearly all situations.

Anterior thalmic lesions stop immediate early gene activation in selective laminae of the retrosplenial cortex: evidence of covert pathology in rats

Lesions involving the anterior thalamic nuclei stopped immediate early gene (IEG) activity in specific regions of the rat retrosplenial cortex, even though there were no apparent cytoarchitectonic changes. Discrete anterior thalamic lesions were made either by excitotoxin (Experiment 1) or radiofrequency (Experiment 2) and, following recovery, the rats foraged in a radial-arm maze in a novel room. Measurements made 6-12 weeks postsurgery showed that, in comparison with surgical controls, the thalamic lesions produced the same, selective patterns of Fos changes irrespective of method. Granular (caudal granular cortex and rostral granular cortex), but not dysgranular (dysgranular cortex), retrosplenial cortex showed a striking loss of Fos-positive cells. While a loss of between 79 and 89% of Fos-positive cells was found in the superficial laminae, the deeper layers appeared normal. In Experiments 3 and 4, rats 9-10 months postsurgery were placed in an activity box for 30 min. Anterior thalamic lesions (Experiment 3) led to a pronounced IEG decrease of both Fos and zif268 throughout the retrosplenial cortex that now included the dysgranular area. These IEG losses were found even though the same regions appeared normal using standard histological techniques. Lesions of the postrhinal cortex (Experiment 4) did not bring about a loss of retrosplenial IEG activity even though this region is also reciprocally connected with the retrosplenial cortex. This selective effect of anterior thalamic damage upon retrosplenial activity may both amplify the disruptive effects of anterior thalamic lesions and help to explain the posterior cingulate hypoactivity found in Alzheimer's disease.

Guiding contact by coupling the taus of gaps

Animals control contact with surfaces when locomoting, catching prey, etc. This requires sensorily guiding the rate of closure of gaps between effecters such as the hands, feet or jaws and destinations such as a ball, the ground and a prey. Control is generally rapid, reliable and robust, even with small nervous systems: the sensorimotor processes are therefore probably rather simple. We tested a hypothesis, based on general tau theory, that closing two gaps simultaneously, as required in many actions, might be achieved simply by keeping the taus of the gaps coupled in constant ratio. tau of a changing gap is defined as the time-to-closure of the gap at the current closure-rate. General tau theory shows that tau of a gap could, in principle, be directly sensed without needing to sense either the gap size or its rate of closure. In our experiment, subjects moved an effector (computer cursor) to a destination zone indicated on the computer monitor, to stop in the zone just as a moving target cursor reached it. The results indicated the subjects achieved the task by keeping tau of the gap between effector and target coupled to tau of the gap between the effector and the destination zone. Evidence of tau -coupling has also been found, for example, in bats guiding landing using echolocation. Thus, it appears that a sensorimotor process used by different species for coordinating the closure of two or more gaps between effecters and destinations entails constantly sensing the taus of the gaps and moving so as to keep the taus coupled in constant ratio.

Supported and liquid phase task specific ionic liquids for base catalysed Knoevenagel reactions

A series of Hunig's base tethered ammonium ionic liquids have been used to catalyse the Knoevenagel condensation of aldehydes/ketones with malononitrile and ethyl cyanoacetate. The reactions were performed under homogeneous and under biphasic, liquid-liquid and liquid-silica supported ionic liquid, conditions with the biphasic systems employing cyclohexene as the second phase. By increasing the distance between the ammonium head group and Hunig's base the activity of the catalyst was found to increase. Higher activity, in general, was found under homogeneous reaction conditions; however, the recyclability of the catalyst was improved by supporting the BIL under biphasic conditions. (c) 2007 Elsevier B.V. All rights reserved.

The role of the primary motor cortex during skill acquisition on a two-degrees-of-freedom movement task

One can partially eliminate motor skills acquired through practice in the hours immediately following practice by applying repetitive transcranial stimulation (rTMS) over the primary motor cortex. The disruption of acquired levels of performance has been demonstrated on tasks that are ballistic in nature. The authors investigated whether motor recall on a discrete aiming task is degraded following a disruption of the primary motor cortex induced via rTMS. Participants (N = 16) maintained acquired performance levels and patterns of muscle activity following the application of rTMS. despite a reduction in corticospinal excitability. Disruption of the primary motor cortex during a consolidation period did not influence the retention of acquired skill in this type of discrete visuomotor task.

Neuromuscular-skeletal constraints on the acquisition of skill in a discrete torque production task

The organisation of the human neuromuscular-skeletal system allows an extremely wide variety of actions to be performed, often with great dexterity. Adaptations associated with skill acquisition occur at all levels of the neuromuscular-skeletal system although all neural adaptations are inevitably constrained by the organisation of the actuating apparatus (muscles and bones). We quantified the extent to which skill acquisition in an isometric task set is influenced by the mechanical properties of the muscles used to produce the required actions. Initial performance was greatly dependent upon the specific combination of torques required in each variant of the experimental task. Five consecutive days of practice improved the performance to a similar degree across eight actions despite differences in the torques required about the elbow and forearm. The proportional improvement in performance was also similar when the actions were performed at either 20 or 40% of participants' maximum voluntary torque capacity. The skill acquired during practice was successfully extrapolated to variants of the task requiring more torque than that required during practice. We conclude that while the extent to which skill can be acquired in isometric actions is independent of the specific combination of joint torques required for target acquisition, the nature of the kinetic adaptations leading to the performance improvement in isometric actions is influenced by the neural and mechanical properties of the actuating muscles.

Neuromuscular adaptation during skill acquisition on a two degree-of-freedom target-acquisition task: Isometric torque production

In this study we attempted to identify the principles that govern the changes in neural control that occur during repeated performance of a multiarticular coordination task. Eight participants produced isometric flexion/extension and pronation/supination torques at the radiohumeral joint, either in isolation (e.g., flexion) or in combination (e.g., flexion - supination), to acquire targets presented by a visual display. A cursor superimposed on the display provided feedback of the applied torques. During pre- and postpractice tests, the participants acquired targets in eight directions located either 3.6 cm (20% maximal voluntary contraction [MVC]) or 7.2 cm (40% MVC) from a neutral cursor position. On each of five consecutive days of practice the participants acquired targets located 5.4 cm (30% MVC) from the neutral position. EMG was recorded from eight muscles contributing to torque production about the radiohumeral joint during the pre- and posttests. Target-acquisition time decreased significantly with practice in most target directions and at both target torque levels. These performance improvements were primarily associated with increases in the peak rate of torque development after practice. At a muscular level, these changes were brought about by increases in the rates of recruitment of all agonist muscles. The spatiotemporal organization of muscle synergies was not significantly altered after practice. The observed adaptations appear to lead to performances that are generalizable to actions that require both greater and smaller joint torques than that practiced, and may be successfully recalled after a substantial period without practice. These results suggest that tasks in which performance is improved by increasing the rate of muscle activation, and thus the rate of joint torque development, may benefit in terms of the extent to which acquired levels of performance are maintained over time.