Search

Multi-Task Learning is a learning paradigm that uses correlated tasks to improve performance generalization. A common way to learn multiple tasks is through the hard parameter sharing approach, in which a single architecture is used to share the same subset of parameters, creating an inductive bias between them during the training process. Due to its simplicity, potential to improve generalization, and reduce computational cost, it has gained the attention of the scientific and industrial communities. However, tasks often conflict with each other, which makes it challenging to define how the gradients of multiple tasks should be combined to allow simultaneous learning. To address this problem, we use the idea of multi-objective optimization to propose a method that takes into account temporal behaviour of the gradients to create a dynamic bias that adjusts the importance of each task during backpropagation. The result of this method is to give more attention to tasks that are diverging or not being benefited during the last iterations, ensuring that the simultaneous learning is heading to the performance maximization of all tasks. As a result, we empirically show that the proposed method outperforms the state-of-the-art approaches on learning conflicting tasks. Unlike the adopted baselines, our method ensures that all tasks reach good generalization performances. [ABSTRACT FROM AUTHOR]

We investigate cheating in work groups, to empirically test the idea of an honest workplace environment as a determinant of performance. Three individuals receive team-based performance pay for executing a real-effort task. In addition, two of them have the opportunity to obtain a bonus in a dice game, which allows cheating without exposure by misreporting a secret die roll. We are particularly interested in the behavioral response of the bystander as the potential witness to the dishonest action. To identify the implications of lies at work, the rules of the bonus game were altered to randomly prevent cheating, or not, across treatment conditions while holding the monetary consequences constant. Survey data enables us to analyze effect heterogeneity and to explore mechanisms underlying behavioral responses. We begin our analysis by estimating the mean lying rate and find that the opportunity-to-cheat is exploited in roughly 42% of cases. The probability of misreporting increases if the cheater's partner in crime is male. Contrary to claims on the importance of honesty at work, we do not observe a reduction in performance when cheating takes place, neither for the bystander nor for the whole team. Bounded awareness could be an explanation, as we find substantial evidence for effect heterogeneity along the lines of information preferences. Bystanders with higher preferences for inconvenient information provide relatively low task performance, compared to those with lower information preferences, who seem to turn a blind eye to the dishonest action of their co-workers by putting increased effort into their work. [ABSTRACT FROM AUTHOR]

This experiment tested whether personal choice vs. external assignment of task characteristics moderates the effect of incidental affective stimulation on effort-related cardiovascular response. We expected strong action shielding and low receptivity for incidental affective influences when participants could choose themselves the stimulus color of an easy memory task. By contrast, when the stimulus color was assigned, we expected weak action shielding and high receptivity. As expected, participants in the assigned color condition showed stronger cardiac pre-ejection period reactivity when exposed to sad music than when exposed to happy music during task performance. These music effects did not appear among participants who could personally choose the stimulus color. Our results replicate previous research by showing that personal choice leads to action shielding, whereas individuals remain receptive for affective influences during volition when task characteristics are assigned. • Half the participants got to personally choose task characteristics; the other half got assigned to the task characteristics • Participants worked on an easy memory task and were presented with happy vs. sad music during task performance • PEP reactivity was stronger in the sad than in the happy music condition, but only when participants did not get to choose • The results show that personal choice leads to action shielding against incidental affective influences on action execution [ABSTRACT FROM AUTHOR]

Work on physiological and other behavioral correlates of motives often assumes that motives exert a direct effect on behavior once activated. Motivational intensity theory, however, suggests that this does not always apply. In the context of task engagement, motive strength should exert a direct effect on myocardial beta-adrenergic activity if task difficulty is unclear, but not if task difficulty is known. The presented study tested this prediction for the impact of the explicit achievement motive on myocardial beta-adrenergic activity—assessed as pre-ejection period (PEP) reactivity during task performance. Seventy-eight participants performed one of two versions of a mental arithmetic task. After having completed the achievement motive scale of the Personality Research Form, participants were either informed about the difficulty of the task or not before working on it. Participants' PEP reactivity during task performance provided evidence for the predicted moderating impact of clarity of task difficulty: PEP reactivity increased with increasing achievement motive strength if task difficulty was unclear, but not if it was clear. These findings demonstrate that the explicit achievement motive impact on myocardial beta-adrenergic activity is moderated by clarity of task difficulty and suggest that motive strength does not always translate into direct effects on physiology and behavior. • Explicit achievement motive (sanAch) strength affects PEP response. • Clarity of task demand moderates the sanAch-PEP relationship. • Higher sanAch is associated with stronger PEP response if task demand is unclear. [ABSTRACT FROM AUTHOR]

Previous studies have suggested that people are sensitive to anticipated cognitive processing demands when deciding which task to perform, but the influence of perceptual processing demands on voluntary task choice is still unclear. The present study tested whether voluntary task choice behavior may be influenced by unpredictable task-specific perceptual processing demands. Across four experiments using different voluntary task choice procedures, we randomly varied the perceptual discriminability of stimuli (easy vs. hard color discrimination) for one of the two tasks. We reasoned that people could only reactively adjust their task choice behavior to the unpredictable discriminability manipulation if they engaged in some perceptual processing before a task goal becomes sufficiently activated to select the task for further processing. The results confirmed this hypothesis: Task performance data demonstrated the presence of perceptual (discriminability effects) and cognitive (switch costs) processing demands. Participants' choice behavior was affected by both types of processing demands (as reflected in a task repetition bias and a bias to select the color task with easy compared to hard discriminations). Thus, the present findings indicate that both perceptual and cognitive processing demands influence voluntary task choice behavior. We propose that higher-level goal activations interact at least partially with early perceptual processes to influence task choice behavior, suggesting a locus of voluntary choices during or after the perceptual stage within the information-processing stream. [ABSTRACT FROM AUTHOR]

Over the past few decades, researchers have argued that playing action video games can substantially improve cognitive abilities and enhance learning. However, consensus has not been reached regarding the mechanisms through which action game experience facilitates superior performance on untrained perceptually and cognitively demanding transfer tasks. We argue that analysis of behaviors engaged in during transfer task performance may provide key insights into answering this question. In the current investigation, we examined potential action game effects in the context of a complex psychomotor task, the Space Fortress (SF) game, that allows for the detailed examination of player behaviors beyond aggregate score reports. Performance (game score) was compared between action video game players (VGPs) and non-gamers (nVGPs) in two different control interface conditions (keyboard or joystick), followed by analyses of behaviors associated with superior performance. Against expectations, VGPs displayed superior performance only in the keyboard condition, suggesting that the action gamer advantage may not generalize to less-familiar control interfaces. Performance advantages were specifically associated with more efficient ship control behaviors by VGPs. Findings highlight how process-tracing approaches may provide insight into the nature of, and mechanisms producing, action gamers' advantages on learning untrained tasks. [ABSTRACT FROM AUTHOR]

• Compared variance in d ′ for replicated performance across task type and difficulty. • Employed 2-AFC with Reminder (2-AFCR), Triangle, and Tetrad tasks. • On a difficult task, 2-AFCR had lower variance than the Triangle or Tetrad tasks. • On a mid-difficulty task, 2-AFCR had lower variance than the Triangle. • No differences in the variance of d ′ were observed for the easy task. Sensory scientists are commonly tasked with determining whether consumers will be able to detect changes in product formulations or ingredients. To this end, sensory panels are formed and discrimination tasks employed, of which numerous types exist. This research compares the performance of three discrimination tasks: the traditional triangle task and two more modern tasks, the two-alternative forced-choice with reminder (2-AFCR) and Tetrad tasks. Comparisons were in terms of observed performance without reliance on the binomial assumption implicit in most contemporary methods of analysis. Seven judges tasted aqueous solutions containing orange essence and were asked to identify the sample containing the greatest amount of essence. Orange essence concentrations were tailored to each individual judge during preliminary testing, and three concentrations of orange essence, greater than the fixed concentration of the reference sample were determined for each judge. Across judges, these were referred to as the easy, medium, and hard levels. To compare discriminability across the three tasks the sensitivity index dʹ was calculated. For the 2-AFCR task, this first required the decision strategy used by the judges to be determined. In general, the judges most likely used the β/τ strategy in 2-AFCR. Across the three tasks, average dʹ values for the same judges, discriminating between the same stimuli, were not significantly different. However, the 2-AFCR task had significantly lower sample variances of d ′ than the Triangle or Tetrad tasks at low levels of d ′ (the hard level), and than the Triangle task at the medium level of difficulty. No differences were observed at the easy level. [ABSTRACT FROM AUTHOR]

• Task complexity for skilled motor practice gave different effects on interlimb transfer between right- and left-handed. • The simple task failed to improve performance on the right hand after left-hand motor practice in right-handed subjects. • The complex task resulted in performance improvement on non-trained hands by contralateral motor practice. • A certain level of appropriate complexity is necessary to detect inter-limb transfers in motor learning in right-handed. Task complexity could affect acquisition efficiency of motor skills and interlimb transfer; however, how task complexity affects interlimb transfer remains unclear. We hypothesized that left- and right-handed participants may have different interlimb transfer efficiency depending on the task complexity. Left-hand (n = 28) and right-hand (n = 28) dominant participants (age = 24.70 ± 4.02 years, male:female = 28:28) performed a finger sequence test with two levels of complexity (simple: one-digit with four fingers vs. complex: two-digit with five fingers) before and after ten trials of 2-min practice each on the same apparatus. The speed and task errors were measured and analyzed. Right-handed participants failed to improve performance on their right hand (non-trained hand) after contralateral left-hand practice in the simple finger sequence task. In contrast, the left-handed participants improved performance on non-trained hands both right and left after contralateral practices. In the complex task, however, both the left- and right-handed participants improved performance on non-trained hands by contralateral practices. Our results showed that task complexity of skilled practice gave different effects on interlimb transfer between right- and left-handed subjects. It appears that a certain level of appropriate complexity is necessary to detect inter-limb transfers in motor learning in right-handed subjects. [ABSTRACT FROM AUTHOR]

In this paper, we focus on splitting a task into multiple sub-tasks to improve schedulability performance for a set of tasks subject to timing constraints. Targeting FPS (Fixed Priority Scheduling) as a scheduling algorithm and RTA (Response Time Analysis) as a schedulability analysis on a multiprocessor platform, we develop a systematic method to utilize the task split, which addresses (i) how to apply the task split without violating the timing requirement of the original tasks and (ii) how to apply each task so as to make an unschedulable task set to be schedulable. Our simulation results demonstrate that RTA with the method finds 16.1%–19.4% (on average) additional task sets that were not proven schedulable by the vanilla RTA. [ABSTRACT FROM AUTHOR]

It has been proposed that humans automatically compute the visual perspective of others. Evidence for this view comes from the Dot Perspective Task. In this task, participants view a room in which a human actor is depicted, looking either leftwards or rightwards. Dots can appear on either the left wall of the room, the right wall, or both. At the start of each trial, participants are shown a number. Their speeded task is to decide whether the number of dots visible matches the number shown. On consistent trials the participant and the actor can see the same number of dots. On inconsistent trials, the participant and the actor can see a different number of dots. Participants respond faster on consistent trials than on inconsistent trials. This self-consistency effect is cited as evidence that participants compute the visual perspective of others automatically, even when it impedes their task performance. According to a rival interpretation, however, this effect is a product of attention cueing: slower responding on inconsistent trials simply reflects the fact that participants' attention is directed away from some or all of the to-be-counted dots. The present study sought to test these rival accounts. We find that desk fans, a class of inanimate object known to cue attention, also produce the self-consistency effect. Moreover, people who are more susceptible to the effect induced by fans tend to be more susceptible to the effect induced by human actors. These findings suggest that the self-consistency effect is a product of attention cueing. [ABSTRACT FROM AUTHOR]

• Participants in a competitive task situation had higher competitive anxiety. • AUT and CRRAT performances in the paired-player mode were better. • CRRAT performance in a competitive task situation was better. Co-creativity has recently received increasing attention. However, few empirical studies explore individuals' creative performance in a group, and fewer have approached creative ideation in different task situations. This study recruited 156 participants to complete creativity tests on an online creativity task platform. Participants were randomly assigned to either cooperative or competitive task situations. Their performance was analyzed using two creativity tests: the Alternative Uses Test (AUT) and Chinese Radical Remote Associates Test (CRRAT). Participants completed tasks alone (i.e., in single player mode) and in a cooperative or competitive situation (i.e., in paired-player mode). The results revealed that participants in the competitive task situation showed higher levels of competitive anxiety. Moreover, their AUT and CRRAT performances in paired-player mode were better than those in single player mode. In the cooperative task situation, participants' CRRAT performance was significantly better than in the competitive task situation. This study had two main findings. First, it strengthens the understanding of how group work enhances individual online creative performance. Second, it distinguishes the influences of cooperative or competitive task situations on different creative performance. This study revealed the differences in creative performance in distinct task situations. [ABSTRACT FROM AUTHOR]

• Introduce task amenability for quantifying task-specific image quality assessment • Propose meta-reinforcement learning algorithms for learning label-efficient, adaptable task amenability • Select task amenable data for improving performance in three tasks from two clinical applications [Display omitted] In this paper, we consider image quality assessment (IQA) as a measure of how images are amenable with respect to a given downstream task, or task amenability. When the task is performed using machine learning algorithms, such as a neural-network-based task predictor for image classification or segmentation, the performance of the task predictor provides an objective estimate of task amenability. In this work, we use an IQA controller to predict the task amenability which, itself being parameterised by neural networks, can be trained simultaneously with the task predictor. We further develop a meta-reinforcement learning framework to improve the adaptability for both IQA controllers and task predictors, such that they can be fine-tuned efficiently on new datasets or meta-tasks. We demonstrate the efficacy of the proposed task-specific, adaptable IQA approach, using two clinical applications for ultrasound-guided prostate intervention and pneumonia detection on X-ray images. [ABSTRACT FROM AUTHOR]

Our actions cause manifold environmental changes. Monitoring these action effects serves at least two vital functions: While the validation of currently relevant effects assesses goal-achievement, screening for currently irrelevant effects accumulates knowledge about potential action-effect relationships. However, monitoring the perceptual consequences of our actions presumably impairs performance in concurrent tasks. Here, we investigated how effect relevance modulates monitoring costs by manipulating instructions in three dual-task experiments. We found performance decreases not only after validation of goal-relevant action effects but to a smaller extent also after screening of goal-irrelevant action effects. These results suggest that effect monitoring is a rather fundamental limitation of dual tasking. [ABSTRACT FROM AUTHOR]

The present study explored the influence of iconicity on sign lexical retrieval and whether it is modulated by the task at hand. Lexical frequency was also manipulated to have an index of lexical processing during sign production. Behavioural and electrophysiological measures (ERPs) were collected from 22 Deaf bimodal bilinguals while performing a picture naming task in Catalan Sign Language (Llengua de Signes Catalana , LSC) and a word-to-sign translation task (Spanish written-words to LSC). Iconicity effects were observed in the picture naming task, but not in the word-to-sign translation task, both behaviourally and at the ERP level. In contrast, frequency effects were observed in the two tasks, with ERP effects appearing earlier in the word-to-sign translation than in the picture naming task. These results support the idea that iconicity in sign language is not pervasive but modulated by task demands. As discussed, iconicity effects in sign language would be emphasised when naming pictures because sign lexical representations in this task are retrieved via semantic-to-phonological links. Conversely, attenuated iconicity effects when translating words might result from sign lexical representations being directly accessed from the lexical representations of the word. • The influence of iconicity and lexical frequency on sign production was explored. • Deaf LSC signers performed a picture naming and a word-to-sign translation task. • Iconicity related effects were observed only in the picture naming task. • Lexical frequency effects affected performance in both tasks. • Influence of iconicity on sign production was dependent on the task at hand. [ABSTRACT FROM AUTHOR]

The Hawai'i Space Exploration Analog and Simulation (HI-SEAS) project is a NASA-funded research program operating long-duration planetary analog surface mission simulations on Mauna Loa volcano, Hawai'i. During missions lasting from 4 to 12 months, crews of six analog astronaut participants live and work in an isolated habitat, communicating with a remote mission support team via a 20-min time delay. The main purpose of HI-SEAS is to study team effectiveness and adaptation over time in isolated, confined, and high autonomy mission scenarios. Among other duties, Crewmembers are tasked with routinely conducting geological fieldwork requiring extravehicular activity (EVA) in the environment surrounding the habitat. They must determine how they will accomplish these tasks, conduct the tasks themselves, and report results by a due date set by the remote science team. Here we describe the design, task parameters, and performance outcomes of HI-SEAS geology EVA tasks from four 6-person missions. We describe the assigned tasks, how the crews carried out their assignments, and the results of their work in terms of six performance metrics for each task: 1) number of days required for completion; 2) number of crewmembers participating; 3) number of EVAs required; 4) total EVA time required; 5) difference between required and planned EVA times; and 6) performance score evaluating how well crew met the task objective. We find weak evidence of a decrease in geology task performance during the third quarter of missions M2-M4. This dataset provides insights into varying crew performance over time for different mission durations. • Geologic field tasks were given to 6-person crews during 4 simulated long-duration planetary missions on Mauna Loa, Hawai'i. • Crews performed tasks in extra-vehicular activities that they planned and executed, reporting results to the science team. • Six metrics evaluating crew performance to study team effectiveness during isolated, confined, and high autonomy missions. • Crew performance across missions showed sensitivity to differences between tasks, crews, and changes within crews over time. • These tasks aid in investigating behavioral factors of operational performance in long-duration isolation and confinement. [ABSTRACT FROM AUTHOR]

Background and objectives: Transfer learning is a valuable approach to perform medical image segmentation in settings with limited cases available for training convolutional neural networks (CNN). Both the source task and the source domain influence transfer learning performance on a given target medical image segmentation task. This study aims to assess transfer learning-based medical segmentation task performance for various source task and domain combinations. Methods: CNNs were pre-trained on classification, segmentation, and self-supervised tasks on two domains: natural images and T1 brain MRI. Next, these CNNs were fine-tuned on three target T1 brain MRI segmentation tasks: stroke lesion, MS lesions, and brain anatomy segmentation. In all experiments, the CNN architecture and transfer learning strategy were the same. The segmentation accuracy on all target tasks was evaluated using the mIOU or Dice coefficients. The detection accuracy was evaluated for the stroke and MS lesion target tasks only. Results: CNNs pre-trained on a segmentation task on the same domain as the target tasks resulted in higher or similar segmentation accuracy compared to other source task and domain combinations. Pre-training a CNN on ImageNet resulted in a comparable, but not consistently higher lesion detection rate, despite the amount of training data used being 10 times larger. Conclusions: This study suggests that optimal transfer learning for medical segmentation is achieved with a similar task and domain for pre-training. As a result, CNNs can be effectively pre-trained on smaller datasets by selecting a source domain and task similar to the target domain and task. [ABSTRACT FROM AUTHOR]

• Children show lower task performance compared to adults when learning a 1D task. • Coordination analyses indicated greater degree of exploration in children. • Children show limited adaptability in matching exploration to task demands. Motor learning in novel tasks requires exploration to find the appropriate coordination patterns to perform the task. Prior work has shown that compared to adults, children show limited exploration when learning a task that required using upper body movements to control a 2D cursor on a screen. Here, by changing the task dimensionality to 1D, we examined two competing hypotheses: whether children show limited exploration as a general strategy, or whether children are suboptimal in adapting their exploration to task dimensionality. Two groups of children (9- and 12-year olds), and one group of adults learned a virtual task that involved learning to control a cursor on the screen using movements of the upper body. Participants practiced the task for a single session with a total of 232 reaching movements. Results showed that 9-year olds show worse task performance relative to adults, as indicated by higher movement times and path lengths. Analysis of the coordination strategies indicated that both groups of children showed lower variance along the first principal component, suggesting that they had greater exploration than adults which was suboptimal for the 1D task. These results suggest that motor learning in children is characterized not by limited exploration per se, but by a limited adaptability in matching motor exploration to task dimensionality. [ABSTRACT FROM AUTHOR]

Flow experience refers to a motivational state of complete concentration on a given task, which can immediately facilitate cognitive performance. Although research has identified feedback as strong antecedent of flow, differences between the effects of positive and negative feedback on flow and the moderating role of personality traits remain largely unexplored. In the present online-based experimental study, we investigated flow as mediator in the effects of normative feedback on performance in a logical reasoning task. Thereby, we also examined moderating effects of locus of control and conscientiousness on indirect effects of positive and negative feedback on task performance via flow. Our sample involved 240 participants, who were randomly assigned to three different feedback conditions (positive, negative, no feedback) after a logical reasoning task. Subsequently, the subjects solved a follow-up task for which they reported the degree of their flow experience. Moderated mediation path modelling showed that positive feedback exerted a positive indirect effect on task performance via flow for highly conscientious individuals. In contrast, negative feedback exerted negative indirect effects on task performance via flow for individuals, who exhibited an internal locus of control or low levels of conscientiousness. Finally, we derived theoretical implications for scholarly understanding of flow experience. [ABSTRACT FROM AUTHOR]

• Preschool children's private speech content was associated with task performance outcomes. • More frequent task-irrelevant private speech content was associated with lower task accuracy during a construction task. • More frequent planning and motivational private speech was associated with higher task accuracy during a construction task. • Children used less regulatory private speech in a computerised card sort compared to a construction task. Emerging in the preschool years, private speech provides a regulatory tool to support task performance. Here, we investigate the presence and nature of regulatory private speech content subtypes and their relationship to task performance in 3- to 5-year-old children. Seventy-one children completed a Duplo construction and an iPad-administered card sort task. Intelligible private speech utterances were coded according to task-relevance and across three regulatory phases – forethought content (planning, analytical and motivational), performance content (self-instructional, observational and attention focusing) and self-reflective content (evaluative and affect-laden). Children produced less task-relevant private speech during the card sort relative to the Duplo task. Children demonstrated all content subtypes with performance content being most prevalent. Forethought content was positively associated with Duplo construction accuracy when children were aged between 36 months to 54 months. In addition, there was a negative association between task-irrelevant content and Duplo construction accuracy. Our findings suggest that preschool children whose private speech contains more forethought content may experience greater success on similar tasks. [ABSTRACT FROM AUTHOR]

Working memory is fundamental to human cognitive functioning, and it is often measured with the n-back task. However, it is not clear whether the n-back task is a valid measure of working memory. Importantly, previous studies have found poor correlations with measures of complex span, whereas a recent study (Frost et al., 2019) showed that n-back performance was correlated with a transsaccadic memory task but dissociated from performance on the change detection task, a well-accepted measure of working memory capacity. To test whether capacity is involved in the n-back task we correlated a spatial version of the test with different versions of the change detection task. Experiment 1 introduced perceptual and cognitive disruptions to the change detection task. This impacted task performance, however, all versions of the change detection task remained highly correlated with one another whereas there was no significant correlation with the n-back task. Experiment 2 removed spatial and non-spatial context from the change detection task. This produced a correlation with n-back. Our results indicate that the n-back task is supported by faculties similar to those that support change detection, but that this commonality is hidden when contextual information is available to be exploited in a change detection task such that structured representations can form. We suggest that n-back might be a valid measure of working memory, and that the ability to exploit contextual information is an important faculty captured by some versions of the change detection task. • N-back performance is predicted by change detection when recall context is sparse. • N-back correlates poorly with the 'standard' change detection task. • Perceptual and cognitive change detection disruptions do not alter the n-back relationship. • Provides validation for n-back as a measure of unstructured working memory capacity [ABSTRACT FROM AUTHOR]