Differential Learning Research Articles

A motor control model of task-specific dystonia and its rehabilitation.

Task-specific dystonia is a painless deficit of motor control specific to a particular motor skill. In this article we present a motor control model which integrates risk factors for the disorder with the neuroscientific literature of skill learning in health. We particularly focus on the idea that the amount and type of movement variability is critical and show how retraining therapies such as Differential Learning which reintroduces variability into practice can restore motor performance.

Acute Effects of Instructed and Self-Created Variable Rope Skipping on EEG Brain Activity and Heart Rate Variability.

The influence of physical activity on brain and heart activity dependent on type and intensity of exercise is meanwhile widely accepted. Mainly cyclic exercises with longer duration formed the basis for showing the influence on either central nervous system or on heart metabolism. Effects of the variability of movement sequences on brain and heart have been studied only sparsely so far. This study investigated effects of three different motor learning approaches combined with a single bout of rope skipping exercises on the spontaneous electroencephalographic (EEG) brain activity, heart rate variability (HRV) and the rate of perceived exertion (RPE). Participants performed repetitive learning (RL) and two extremely variable rope skipping schedules according to the differential learning approach. Thereby one bout was characterized by instructed variable learning (DLi) and the other by self-created variable learning (DLc) in randomized order each on three consecutive days. The results show higher RPE after DLi and DLc than after RL. HRV analysis demonstrates significant changes in pre–post exercise comparison in all training approaches. No statistically significant differences between training schedules were identified. Slightly greater changes in HRV parameters were observed in both DL approaches indicating a higher activation of the sympathetic nervous system. EEG data reveals higher parietal alpha1 and temporal alpha2 power in RL compared to both DL schedules immediately post exercise. During the recovery of up to 30 min, RL shows higher temporal and occipital theta, temporal, parietal and occipital alpha, temporal and occipital beta and frontal beta3 power. In conclusion, already a single bout of 3 min of rope skipping can lead to brain states that are associated with being advantageous for cognitive learning. Combined with additional, cognitively demanding tasks in form of the DL approach, it seems to lead to an overload of the mental capacity, at least on the short term. Further research should fathom the reciprocal influence of cardiac and central-nervous strain in greater detail.

A descriptive profile of state Child Care and Development Fund policies in states with high populations of low-income Hispanic children

The Child Care and Development Fund (CCDF) serves dual goals of promoting low-income parents' employment, education, and training, and supporting parents’ use of high quality child care and education. While changes in CCDF policies have aimed to improve the accessibility and functionality of CCDF subsidies for all eligible low-income families, utilization among Hispanic families, who represent the highest growing proportion of income poor families, remains low. Our study explores the ways in which the state-level child care policy context may differentially affect utilization among Hispanic families and contribute to racial/ethnic disparities in program utilization. We use the child care accommodation model and administrative exclusion to offer an expanded conceptual framework. We consider how policies and administrative practices may interact with demographic and community characteristics common among low-income Hispanic families and impose differential learning, psychological, and compliance costs for accessing government assistance programs such as CCDF. Our analysis of the 13 states in which over 80% of the low-income Hispanic child population resides, identifies policies around eligibility, documentation requirements, receipt prioritization, and the online user experience that vary across states and may serve to either facilitate or constrain access to a child care subsidy. Implications for the design of child care policies to equitably support economic and child well-being for all eligible families are discussed.

Prefrontal Theta Oscillations Promote Selective Encoding of Behaviorally Relevant Events.

The ability to capture the most relevant information from everyday experiences without constantly learning unimportant details is vital to survival and mental health. While decreased activity of the medial prefrontal cortex (mPFC) is associated with failed or inflexible encoding of relevant events in the hippocampus, mechanisms used by the mPFC to discern behavioral relevance of events are not clear. To address this question, we chemogenetically activated excitatory neurons in the mPFC of male rats and examined its impact on local network activity and differential associative learning dependent on the hippocampus. Rats were exposed to two neutral stimuli in two environments whose contingency with an aversive stimulus changed systematically across days. Over 2 weeks of differential and reversal learning, theta band activity began to ramp up toward the expected onset of the aversive stimulus, and this ramping activity tracked the subsequent shift of the set (stimulus modality to environment) predictive of the aversive stimulus. With chemogenetic mPFC activation, the ramping activity emerged within a few sessions of differential learning, which paralleled faster learning and stronger correlations between the ramping activity and conditioned responses. Chemogenetic mPFC activity, however, did not affect the adjustment of ramping activity or behavior during reversal learning or set-shifting, suggesting that the faster learning was not because of a general enhancement of attention, sensory, or motor processing. Thus, the dynamics of the mPFC network activation during events provide a relevance-signaling mechanism through which the mPFC exerts executive control over the encoding of those events in the hippocampus.

The Effects of Differential Learning Trainings on Technical Development of Basketball School Players.

There are several different methods of learning motor skills, like traditional (linear) and differential (nonlinear) learning training. The traditional motor learning approach proposes that learners improve a skill just by repeating it. According to the teaching principles, exercises are selected along continua from easy to hard and from simple to complex. The differential learning approach is mainly characterized by taking advantage, for the purpose of learning, of fluctuations that occur, without movement repetitions and without corrections during the skill acquisition process. The purpose of this study is to examine the effects of differential training on technical development of between 10-12 year-old basketball players who have been continuing basketball education. Twenty-three (23) basketball players who were 10-12 years old from the youth basketball team (Under 13) of Istanbul Marmara University Basketball School were tested voluntarily in this study. In this study, the German Heidelberger Basketball Tests; tourniquet test, dribbling test, shooting test, and passing test were applied on the basketball field with synthetic surface of the Istanbul Marmara University in 2018. The Wilcoxon test for the comparison of pre- and post-tests of the groups were used for statistical analyses. In conclusion, the findings suggest that the differential learning group showed the improvement in performance. These findings may be useful for trainers and physical education teachers in the selection process of players and in preparing basketball training programs. Article visualizations:

A bio-geography-based algorithm for optimal siting and sizing of distributed generators with an effective power factor model

In this paper, a bio-geography-based optimization techniqueis proposed for optimal sizing and placement of multiple distributed generators to reduce system loss and to improve system voltage profiles in electric distribution systems. Distributed generation paves the way for installing generating sources in the vicinity of the loads, for improving the power factor of the system, and thereby significantly reducing total system losses. An effective power factor model is proposed to pre-set the power factor of each distributed generator placed at various locations in a distribution system. The bio-geography optimization algorithm is enhanced with a differential learning scheme to deal with the problems of high dimensionality and complex constraints. A sensitivity factor approach is presented to reduce the search space for the placement of distributed generators in the system. Numerical simulations are performed on the IEEE 33-bus and IEEE 69-bus systems and the results of the proposed method are compared with other methods reported in the literature.

Differential Learning for Outliers: A Case Study of Water Demand Prediction

Predicting water demands is becoming increasingly critical because of the scarcity of this natural resource. In fact, the subject was the focus of numerous studies by a large number of researchers around the world. Several models have been proposed that are able to predict water demands using both statistical and machine learning techniques. These models have successfully identified features that can impact water demand trends for rural and metropolitan areas. However, while the above models, including recurrent network models proposed by the authors are able to predict normal water demands, most have difficulty estimating potential deviations from the norms. Outliers in water demand can be due to various reasons including high temperatures and voluntary or mandatory consumption restrictions by the water utility companies. Estimating these deviations is necessary, especially for water utility companies with a small service footprint, in order to efficiently plan water distribution. This paper proposes a differential learning model that can help model both over-consumption and under-consumption. The proposed differential model builds on a previously proposed recurrent neural network model that was successfully used to predict water demand in central Indiana.

Motor Learning of Gross-Motor Skills under Variable Practice Conditions

Research background and hypothesis. For both the athlete and the coach, the purpose and goal of training is the same: to enhance performance. Research aim. This study investigated the effect of differential learning on basketball free throw and volleyball strike. Research methods. In the basketball experiment, in pre-, post- and retention test design, the free throw performance was measured (number of successful shots). Aiming to investigate transfer performance, jump shots were tested. In the volleyball experiment, movement variability during the strike was further increased by the application of an elastic constraint. The second intervention and quasi-control group trained under constant practice conditions. Ball velocity and accuracy were analysed with a constant and a variable test.Research results. No signifi cant differences were observed for either the free throw (p > 0.05) or the transfer performance (p > 0.05). However, a positive trend for the variable group was observed in the transfer situation. For the strike in volleyball, the differential learning group had a signifi cant advantage with respect to velocity in a variable test situation (p < 0.05) whereas in the constant situation (p > 0.05) and measurements for accuracy (p > 0.05) it did not reveal similar results.Discussion and conclusions. In both experiments, the set variability leads to benefi ts in variable (transfer) situations. However, as a practical consequence, especially for constant situations, certain moderator variables such as training age or background in other sports or activities must be kept in mind to adjust the amount of external or intervention-induced variability.Keywords: differential learning, variability, adaptability, ball sports.

Phylogenetic patterns in learning and decision making in pit vipers (Viperidae: Crotalinae)

Learning is a key behavioural adaptation allowing animals to respond to complex and changing environments. Although the field of animal behaviour has seen an increase in the taxonomic breadth of learning studies in recent decades, investigations within an explicit, broad phylogenetic framework are rare, curtailing our understanding of the evolution of learning and advanced cognition. Pit vipers (Viperidae: Crotalinae) represent a particularly interesting taxon in which to study patterns in learning as they are widely studied, have had learning documented in many species, occupy a range of ecological niches and possess vast and varied natural history traits. We investigated the latency to decision in a laboratory thermal maze in 13 species of pit viper (7 species of rattlesnakes and 6 species of nonrattlesnake pit viper) and one true viper species (the puff adder, Bitis arietans (Viperidae: Viperinae), as an outgroup comparison). Relative to the other pit vipers, rattlesnakes uniformly and quickly (after only one trial) decreased their latency to respond to thermal stress; their latency remained low and uniform across the remaining 11 trials. All other pit viper species failed to modify their behaviour across trials and maintained a consistent latency across all 12 trials, as did the puff adder. This pattern could reflect differential learning abilities between rattlesnake and nonrattlesnake pit vipers. Our results underscore the advantages of incorporating a broad phylogenetic perspective when investigating learning and comparative cognition.

From statistical inference to a differential learning rule for stochastic neural networks.

Stochastic neural networks are a prototypical computational device able to build a probabilistic representation of an ensemble of external stimuli. Building on the relationship between inference and learning, we derive a synaptic plasticity rule that relies only on delayed activity correlations, and that shows a number of remarkable features. Our delayed-correlations matching (DCM) rule satisfies some basic requirements for biological feasibility: finite and noisy afferent signals, Dale’s principle and asymmetry of synaptic connections, locality of the weight update computations. Nevertheless, the DCM rule is capable of storing a large, extensive number of patterns as attractors in a stochastic recurrent neural network, under general scenarios without requiring any modification: it can deal with correlated patterns, a broad range of architectures (with or without hidden neuronal states), one-shot learning with the palimpsest property, all the while avoiding the proliferation of spurious attractors. When hidden units are present, our learning rule can be employed to construct Boltzmann machine-like generative models, exploiting the addition of hidden neurons in feature extraction and classification tasks.

Left-shifting prism adaptation boosts reward-based learning

Visuospatial cognition has an inherent lateralized bias. Individual differences in the direction and magnitude of this bias are associated with asymmetrical D2/3 dopamine binding and dopamine system genotypes. Dopamine level affects feedback-based learning and dopamine signaling asymmetry is related to differential learning from reward and punishment. High D2 binding in the left hemisphere is associated with preference for reward. Prism adaptation (PA) is a simple sensorimotor technique, which modulates visuospatial bias according to the direction of the deviation. Left-deviating prism adaptation (LPA) induces rightward bias in healthy subjects. It is therefore possible that the right side of space increases in saliency along with left hemisphere dopaminergic activity. Right-deviating prism adaptation (RPA) has been used mainly as a control condition because it does not modulate behavior in healthy individuals. Since LPA induces a rightward visuospatial bias as a result of left hemisphere modulation, and higher dopaminergic activity in the left hemisphere is associated with preference for rewarding events we hypothesized that LPA would increase the preference for learning with reward. Healthy volunteers performed a computer-based probabilistic classification task before and after LPA or RPA. Consistent with our predictions, PA altered the preference for rewarded versus punished learning, with the LPA group exhibiting increased learning from reward. These results suggest that PA modulates dopaminergic activity in a lateralized fashion.

The statistical structures of reinforcement learning with asymmetric value updates

Reinforcement learning (RL) models have been broadly used in modeling the choice behavior of humans and other animals. In standard RL models, the action values are assumed to be updated according to the reward prediction error (RPE), i.e., the difference between the obtained reward and the expected reward. Numerous studies have noted that the magnitude of the update is biased depending on the sign of the RPE. The bias is represented in RL models by differential learning rates for positive and negative RPEs. However, which aspect of behavioral data that the estimated differential learning rates reflect is not well understood. In this study, we investigate how the differential learning rates influence the statistical properties of choice behavior (i.e., the relation between past experiences and the current choice) based on theoretical considerations and numerical simulations. We clarify that when the learning rates differ, the impact of a past outcome depends on the subsequent outcomes, in contrast to standard RL models with symmetric value updates. Based on the results, we propose a model-neutral statistical test to validate the hypothesis that value updates are asymmetric. The asymmetry in the value updates induces the autocorrelation of choice (i.e., the tendency to repeat the same choice or to switch the choice irrespective of past rewards). Conversely, if an RL model without an intrinsic autocorrelation factor is fitted to data that possess an intrinsic autocorrelation, a statistical bias to overestimate the difference in learning rates arises. We demonstrate that this bias can cause a statistical artifact in RL-model fitting leading to a “pseudo-positivity bias” and a “pseudo-confirmation bias.”

Small differences in learning speed for different food qualities can drive efficient collective foraging in ant colonies

Social insects frequently make important collective decisions, such as selecting the best food sources. Many collective decisions are achieved via communication, for example by differential recruitment depending on resource quality. However, even species which only rarely recruit can respond to a changing environment on a collective level by tracking food source quality. We hypothesised that an apparent collective decision to focus on the highest quality food source can be explained by differential learning of food qualities. Overall, ants may learn the location of higher quality food faster, with most ants finally congregating at the best food source. To test the effect of reward quality and motivation on learning in Lasius niger, we trained individual ants to find sucrose molarities of varying concentrations on one arm of a T-maze in spring and in autumn after 1 or 4 days of food deprivation. As predicted, ants learned fastest in spring and lowest in autumn, with reduced food deprivation leading to slower learning. Surprisingly, the effect of food quality and motivation on the learning speed was small. However, persistence rates varied dramatically: All ants in spring made all (6) return visits to all food qualities, in contrast to only 33% of ants in autumn after 1 day of food deprivation. Fitting the empirical findings into an agent-based model revealed that even a weak tendency of ants to memorise routes to high-quality food sources faster can result in ecologically sensible colony-level behaviour. This collective-seeming decision is non-interactive, and thus resembles an annealing process. Collective decisions of insects are often achieved via communication and/or other interactions between individuals. However, animals can also make collective decisions in the absence of communication. We show that foraging motivation and food quality can affect both route memory formation speed and the likelihood to return to the food source and thus mediate selective food exploitation. An agent-based model, implemented with our empirical findings, demonstrates that at the collective level even small differences in learning lead to ecologically sensible behaviour: mildly food-deprived colonies are selective for high-quality food while highly food-deprived colonies exploit all food sources equally. We therefore suggest that non-interactive factors such as individual learning and the foraging motivation of a colony can mediate or even drive group-level behaviour in a process resembling annealing. Instead of accounting collective behaviour exclusively to social interactions, a possible contribution of individual processes should also be considered.

Food and water restriction lead to differential learning behaviors in a head-fixed two-choice visual discrimination task for mice.

Head-fixed behavioral tasks can provide important insights into cognitive processes in rodents. Despite the widespread use of this experimental approach, there is only limited knowledge of how differences in task parameters, such as motivational incentives, affect overall task performance. Here, we provide a detailed methodological description of the setup and procedures for training mice efficiently on a two-choice lick left/lick right visual discrimination task. We characterize the effects of two distinct restriction regimens, i.e. food and water restriction, on animal wellbeing, activity patterns, task acquisition, and performance. While we observed reduced behavioral activity during the period of food and water restriction, the average animal discomfort scores remained in the ‘sub-threshold’ and ‘mild’ categories throughout the experiment, irrespective of the restriction regimen. We found that the type of restriction significantly influenced specific aspects of task acquisition and engagement, i.e. the number of sessions until the learning criterion was reached and the number of trials performed per session, but it did not affect maximum learning curve performance. These results indicate that the choice of restriction paradigm does not strongly affect animal wellbeing, but it can have a significant effect on how mice perform in a task.

The Effects of Differential Grammar Learning Tasks on L2 Learners' Grammatical Knowledge and Their Beliefs about Grammar Instruction

The Effects of Differential Grammar Learning Tasks on L2 Learners' Grammatical Knowledge and Their Beliefs about Grammar Instruction

The effects of an enrichment training program for youth football attackers.

The aim of this study was to identify the effects of a complementary training program based on differential learning approach in the physical, technical, creative and positioning performance of youth football attackers. Fifteen players were allocated into the control (U15C = 9, age: 13.9±0.5 years; U17C = 6, age: 16.1±0.7 years) and the experimental (U15E = 9, age: 14.2±0.8 years; U17E = 6, age: 15.8±0.5 years) groups. The experimental groups participated in 10-weeks of a complementary training program based on differential learning approach to improve physical literacy and players’ tactical behavior. Variables studied encompassed: motor (vertical jump, speed and repeated change-of direction), technical (pass, dribble and shot), creative (fluency, attempts, versatility) and positioning-related variables (stretch index, spatial exploration index and regularity of the lateral and longitudinal movements). Results revealed that U15E improved both the jump and repeated change-of-direction performance, while the U17E have only improved the jump performance. The U15E showed improvements in all technical variables (small to large effects), and in the fluency and versatility (moderate effects), while the U17 have only improved the successful shots (large effects). From a positional perspective, there was a moderate increase in the stretch index, and decreased longitudinal and lateral regularity (small to moderate effects) in the U15E compared to the U15C. In turn, the U17E revealed a moderate increase of the spatial exploration index and a small decrease in the stretch index. Overall, the results suggest that the complementary training program was effective for the development of the overall performance of the U15E attackers, while more time and/or variability may be needed for older age groups. Nevertheless, the overall higher values found in experimental groups, may suggest that this type of complementary training program improves performance.

The Effects of Differential Learning and Traditional Learning Trainings on Technical Development of Football Players

There are several different methods of learning motor skills, like traditional (linear) and differential (nonlinear) learning training. The traditional motor learning approach proposes that learners improve a skill just by repeating it. According to the teaching principles, exercises are selected along continua from easy to hard and from simple to complex. The differential learning approach is mainly characterized by taking advantage, for the purpose of learning, of fluctuations that occur, without movement repetitions and without corrections during the skill acquisition process. The purpose of this study is to examine the effects of differential and traditional training on technical development of 15-year-old football players who have been continuing football education. Twelve (12) football players who were 15 years old from the youth football team (Under 15) of Istanbul Kavacik Club were tested voluntarily in this study. In this study, the Mor-Christian soccer passing test, German Football Association agility/dribbling test, and feet-juggling test were applied on the football field with synthetic grass of the Istanbul Kavacik Sports Club in 2016. The Mann-Whitney U test for paired comparison of the groups and the Wilcoxon test for the comparison of pre- and post-tests of the groups were used for statistical analyses. In conclusion, the findings suggest that the results of the technique tests for groups offer no clear evidence for the superiority of the differential learning approach in comparison to the classical training approach. However, participants of the Differential Group (DG) were able to improve their performance in all tests and techniques compared to those of the Classical Group (CG). These findings may be useful for trainers and physical education teachers in the selection process of players and in preparing football training programs.

Diferenciniu mokymo metodu grindžiamų aerobinio krūvio pratybų poveikis širdies ir kraujagyslių sistemos funkcinei būklei

Seeking for optimal and potentially more effective modes of exercising for health purposes remains an urgent problem today. It is well known that exercising must be individualized, that it is not recommended for exceptionally high intensity of workouts requiring high efforts, especially when a person chooses to exercise again after a longer dropout. The purpose of this study was to compare the changes in features of the cardiovascular system functional state in recovery phase after health-enhancing exercising, when this exercising are performed in easy aerobic mode and when such aerobic exercising are complemented by so-called “Differential Learning” (DL) methodology based tasks. Eight male volunteers were examined twice, i.e. in randomly completed sequence, performed two 20-minutes duration exercising sessions with a three-day break between them. It was the relatively ease workload (25W) on bicycle ergometer. One of the tasks was to assess the impact of exercising in aerobic mode, and the next on the exercising with the DL tasks while the participants during exercising was asked to perform simple tasks that required some coordinating efforts, and each subsequent task was a new variation of the task given before. The evaluations of the functional condition of the cardiovascular system were performed before the exercising, after 5 minutes and 30 minutes after the exercise session. The 12-lead standard ECG was registered continuously and the second-order matrix analysis (algebraic data cointegration approach) was used for assessment the dynamics of concatenation between RR interval and QRS complex duration and between the JT interval. These indices were chosen as reflecting different levels of myocardial regulation and features of metabolism. Results obtained during the study revealed that there was no significant differences (p > 0.05) while comparing the values of durations of RR intervals, JT intervals and the duration of QRS complex. Both aerobic exercising and exercising while the additional tasks based on DL methodology was added caused the same changes of registered electrocardiogram indices. Both, aerobic exercise exercising and the exercising with the DL tasks caused significant increase of concatenation between ECG indices (p > 0.05). Even after 30 minutes of exercising the increased concatenation remained significantly changed in comparison to the initial values recorded before exercising (p < 0.05). The assessment of concatenation between two electrocardiographic indices provides qualitatively new information about features of participant’s functional state. It is purposeful to apply the algebraic data cointegration approach for evaluation complexity features of sportsman body functioning during exercising, it allows to reveal the new features of functional state. It is concluded that aerobic exercising complemented with tasks based on DL methodology has made the stronger effect on the functional state of cardiovascular system. The increased concatenation between ECG indices remained substantially stronger in the recovery phase after exercising. Keywords: functional state, electrocardiography, differential learning. DOI: http://dx.doi.org/10.15823/sm.2018.5

Is Inquiry Science Instruction Effective for English Language Learners? A Meta-Analytic Review.

Despite being among the fastest growing segments of the student population, English Language Learners (ELLs) have yet to attain the same academic success as their English-proficient peers, particularly in science. In an effort to support the pedagogical needs of this group, educators have been urged to adopt inquiry approaches to science instruction. Whereas inquiry instruction has been shown to improve science outcomes for non-ELLs, systematic evidence in support of its effectiveness with ELLs has yet to be established. The current meta-analysis summarizes the effect of inquiry instruction on the science achievement of ELLs in elementary school. Although an analysis of 26 articles confirmed that inquiry instruction produced significantly greater impacts on measures of science achievement for ELLs compared to direct instruction, there was still a differential learning effect suggesting greater efficacy for non-ELLs compared to ELLs. Contextual factors that moderate these effects are identified and discussed.

Randomised controlled trial on differential learning of toothbrushing in 6- to 9-year-old children.

This study aimed to evaluate the effect of differential learning of toothbrushing compared to habitual toothbrushing and instruction/demonstration of toothbrushing on plaque reduction and gingival inflammation of primary school children. Children (6-9years) were subjected to a toothbrushing training consisting of 15days (3 × 5 days, interval 2days, 3min/day) and randomly assigned to one of three groups (each n = 18): habitual toothbrushing/control, instruction/demonstration of toothbrushing, differential learning of toothbrushing. The differential learning approach comprised 15 different movement exercises (1/day), while instruction/demonstration of toothbrushing was based on repetitive practice of toothbrushing. Plaque (Turesky-modified Quigley-Hein plaque index (T-QHI)) and papilla bleeding (papilla bleeding index (PBI)) scores were assessed prior to the first toothbrushing (t0, baseline) and 21 (t1), 42 (t2) and 63 (t3)days after beginning of the study. Primary statistical analyses were performed with cumulative logit regression models with repeated measures. The global significance level was set to 5% two-sided. At all time points, PBI and T-QHI were significantly reduced by differential learning compared to instruction/demonstration of toothbrushing and habitual toothbrushing. Compared to baseline, differential learning improved PBI and T-QHI significantly at all time points. In the other groups, no significant reduction of PBI compared to baseline was observed. T-QHI was significantly reduced by habitual toothbrushing only at t1 and by instruction/demonstration of toothbrushing at t1 and t2. Differential learning had an improved effect on posterior compared to anterior teeth. Differential learning of toothbrushing improved plaque reduction and reduced gingival inflammation compared to habitual toothbrushing and instruction/demonstration of toothbrushing in 6- to 9-year-old children in mixed dentition. Toothbrushing skills of children might be improved by differential learning. ISRCTN14951343, https://www.isrctn.com/ISRCTN14951343?