Scholarly Works - Cambridge Neuroscience
Permanent URI for this collection
Browse
Recent Submissions
Item Open Access Accepted version Peer-reviewed Parallel swarm intelligence strategies for large-scale clustering based on MapReduce with application to epigenetics of aging(Elsevier, 2018) Benmounah, Z; Meshoul, S; Batouche, M; Lio, P; Lio, Pietro [0000-0002-0540-5053]Clustering is an important technique for data analysis and knowledge discovery. In the context of big data, it becomes a challenging issue due to the huge amount of data recently collected making conventional clustering algorithms inappropriate. The use of swarm intelligence algorithms has shown promising results when applied to data clustering of moderate size due to their decentralized and self-organized behavior. However, these algorithms exhibit limited capabilities when large data sets are involved. In this paper, we developed a decentralized distributed big data clustering solution using three swarm intelligence algorithms according to MapReduce framework. The developed framework allows cooperation between the three algorithms namely particle swarm optimization, ant colony optimization and artificial bees colony to achieve largely scalable data partitioning through a migration strategy. This latter reaps advantage of the combined exploration and exploitation capabilities of these algorithms to foster diversity. The framework is tested using amazon elastic map-reduce service (EMR) deploying up to 192 computer nodes and 30 gigabytes of data. Parallel metrics such as speed-up, size-up and scale-up are used to measure the elasticity and scalability of the framework. Our results are compared with their counterparts big data clustering results and show a significant improvement in terms of time and convergence to good quality solution. The developed model has been applied to epigenetics data clustering according to methylation features in CpG islands, gene body, and gene promoter in order to study the epigenetics impact on aging. Experimental results reveal that DNA-methylation changes slightly and not aberrantly with aging corroborating previous studies.Item Open Access Accepted version Peer-reviewed Soft Morphological Processing of Tactile Stimuli for Autonomous Category Formation(IEEE, 2018) Scimeca, Luca; Maiolino, Perla; Iida, Fumiya; Scimeca, Luca [0000-0002-2821-0072]; Iida, Fumiya [0000-0001-9246-7190]Sensor morphology is a fundamental aspect of tactile sensing technology. Design choices induce stimuli to be morphologically processed, changing the sensory perception of the touched objects and affecting inference at a later processing stage. We develop a framework to analyze the filtered sensor response and observe the correspondent change in tactile information. We test the morphological processing effects on the tactile stimuli by integrating a capacitive tactile sensor into a flat end-effector and creating three soft silicon-based filters with varying thickness (3mm, 6mm and 10mm). We incorporate the end-effector onto a robotic arm. We control the arm in order to apply a calibrated force onto 4 objects, and retrieve tactile images. We create an unsupervised inference process through the use of Principal Component Analysis and K-Means Clustering.We use the process to group the sensed objects into 2 classes and observe how different soft filters affect the clustering results. The sensor response with the 3mm soft filter allows for edges to be the feature with most variance (captured by PCA) and induces the association of edged objects. With thicker soft filters the associations change, and with a 10mm filter the sensor response results more diverse for objects with different elongation. We show that the clustering is intrinsically driven by the morphology of the sensor and that the robot’s world understanding changes according to it.Item Open Access Accepted version Peer-reviewed Balancing family with a successful career in neuroscience.(Wiley-Blackwell, 2016-07) Poirazi, P; Belin, D; Gräff, J; Hanganu-Opatz, IL; López-Bendito, G; Belin, David [0000-0002-7383-372X]After years of hard work as a student and postdoc, stressful negotiations and restless nights of agony regarding your academic future, you managed to secure a Principal Investigator (PI) position and establish your own laboratory. And just when you thought you could relax a bit and enjoy some time with your family, or start a family, you find yourself facing massive levels of responsibility added to your research, that demand most of your time and energy. The challenge to balance a successful career with a happy family life is not a trivial one.Item Open Access Accepted version Peer-reviewed Multi-task Learning Using Uncertainty to Weigh Losses for Scene Geometry and Semantics(IEEE, 2018) Cipolla, R; Gal, Y; Kendall, A; Cipolla, Roberto [0000-0002-8999-2151]; Kendall, Alex [0000-0003-1904-5885]Numerous deep learning applications benefit from multi-task learning with multiple regression and classification objectives. In this paper we make the observation that the performance of such systems is strongly dependent on the relative weighting between each task's loss. Tuning these weights by hand is a difficult and expensive process, making multi-task learning prohibitive in practice. We propose a principled approach to multi-task deep learning which weighs multiple loss functions by considering the homoscedastic uncertainty of each task. This allows us to simultaneously learn various quantities with different units or scales in both classification and regression settings. We demonstrate our model learning per-pixel depth regression, semantic and instance segmentation from a monocular input image. Perhaps surprisingly, we show our model can learn multi-task weightings and outperform separate models trained individually on each task.Item Open Access Accepted version Peer-reviewed Investigating diagrammatic reasoning with deep neural networks(Springer International Publishing AG, 2018) Wang, Duo; Jamnik, Mateja; Lio, Pietro; Jamnik, Mateja [0000-0003-2772-2532]; Lio, Pietro [0000-0002-0540-5053]Diagrams in mechanised reasoning systems are typically en- coded into symbolic representations that can be easily processed with rule-based expert systems. This relies on human experts to define the framework of diagram-to-symbol mapping and the set of rules to reason with the symbols. We present a new method of using Deep artificial Neu- ral Networks (DNN) to learn continuous, vector-form representations of diagrams without any human input, and entirely from datasets of dia- grammatic reasoning problems. Based on this DNN, we developed a novel reasoning system, Euler-Net, to solve syllogisms with Euler diagrams. Euler-Net takes two Euler diagrams representing the premises in a syl- logism as input, and outputs either a categorical (subset, intersection or disjoint) or diagrammatic conclusion (generating an Euler diagram rep- resenting the conclusion) to the syllogism. Euler-Net can achieve 99.5% accuracy for generating syllogism conclusion. We analyse the learned representations of the diagrams, and show that meaningful information can be extracted from such neural representations. We propose that our framework can be applied to other types of diagrams, especially the ones we don’t know how to formalise symbolically. Furthermore, we propose to investigate the relation between our artificial DNN and human neural circuitry when performing diagrammatic reasoning.Item Open Access Accepted version Peer-reviewed Evaluation of NeuroPage as a memory aid for people with multiple sclerosis: A randomised controlled trial.(Informa UK Limited, 2020-01) Goodwin, Rachel A; Lincoln, Nadina B; das Nair, Roshan; Bateman, Andrew; Goodwin, Rachel A [0000-0002-4084-3120]; Lincoln, Nadina B [0000-0001-5604-2339]; das Nair, Roshan [0000-0001-8143-7893]Memory problems are reported in 40%-60% of people with multiple sclerosis (MS). These problems affect independence and may limit the ability to benefit from rehabilitation. Our aim was to evaluate the effectiveness of NeuroPage for people with MS living in the community. A multicentre, single-blind, randomised controlled crossover trial was conducted. The intervention comprised the NeuroPage service, which sends reminder messages to mobile phones at pre-arranged times. In the control condition participants received "non-memory texts", that is, messages not aimed at providing a reminder; for example, supplying news headlines or sport updates. Outcome measures were completed using postal questionnaires after each condition. There were 38 participants aged 28 to 72 (mean 48, SD 11) and 10 (26%) were men. There were no significant differences between NeuroPage and control conditions on the Everyday Memory Questionnaire (p = 0.41, d = 0.02). The number of daily diary items forgotten in the NeuroPage condition was significantly less than in the control (9% vs. 31%, p = 0.01, d = -0.64). Psychological distress was less in the NeuroPage condition than control (p = 0.001, d = -0.84). Further evaluation of the effect on everyday memory is required.Item Open Access Accepted version Peer-reviewed Culture-specific links between maternal executive function, parenting, and preschool children's executive function in South Korea.(Wiley, 2018-06) Lee, Min Kyung; Baker, Sara; Whitebread, David; Lee, Min Kyung [0000-0001-8355-0574]BACKGROUND: Research on the relationships between parental factors and children's executive function (EF) has been conducted mainly in Western cultures. AIM: This study provides the first empirical test, in a non-Western context, of how maternal EF and parenting behaviours relate to child EF. SAMPLE: South Korean mothers and their preschool children (N = 95 dyads) completed EF tasks. METHOD: Two aspects of parental scaffolding were observed during a puzzle task: contingency (i.e., adjusting among levels of scaffolding according to the child's ongoing evidence of understanding) and intrusiveness (i.e., directive, mother-centred interactions). RESULTS AND CONCLUSIONS: Maternal EF and maternal contingency each accounted for unique variance in child EF, above and beyond child age, child language and maternal education. Maternal intrusiveness, however, was not significantly related to child EF. Additionally, no mediating role of parenting was found in the maternal and child EF link. However, child language was found to partially mediate the link between maternal contingency and child EF. These results complement prior findings by revealing distinctive patterns in the link between maternal EF, parenting behaviours, and child EF in the Korean context.Item Open Access Accepted version Peer-reviewed The Mirage of Action-Dependent Baselines in Reinforcement Learning(PMLR, 2018-07-31) Tucker, George; Bhupatiraju, Surya; Gu, Shixiang; Turner, Richard E; Ghahramani, Zoubin; Levine, Sergey; Ghahramani, Zoubin [0000-0002-7464-6475]Policy gradient methods are a widely used class of model-free reinforcement learning algorithms where a state-dependent baseline is used to reduce gradient estimator variance. Several recent papers extend the baseline to depend on both the state and action and suggest that this significantly reduces variance and improves sample efficiency without introducing bias into the gradient estimates. To better understand this development, we decompose the variance of the policy gradient estimator and numerically show that learned state-action-dependent baselines do not in fact reduce variance over a state-dependent baseline in commonly tested benchmark domains. We confirm this unexpected result by reviewing the open-source code accompanying these prior papers, and show that subtle implementation decisions cause deviations from the methods presented in the papers and explain the source of the previously observed empirical gains. Furthermore, the variance decomposition highlights areas for improvement, which we demonstrate by illustrating a simple change to the typical value function parameterization that can significantly improve performance.Item Open Access Accepted version Peer-reviewed Multiple motor memories are learned to control different points on a tool.(Springer Science and Business Media LLC, 2018-04) Heald, James B; Ingram, James N; Flanagan, J Randall; Wolpert, Daniel M; Heald, James B [0000-0002-7293-7914]; Ingram, James N [0000-0003-2567-504X]; Flanagan, J Randall [0000-0003-2760-6005]; Wolpert, Daniel M [0000-0003-2011-2790]Skillful object manipulation requires learning the dynamics of objects, linking applied force to motion 1 ,2 . This involves the formation of a motor memory 3 ,4 , which has been assumed to be associated with the object, independent of the point on the object that one chooses to control. Importantly, in manipulation tasks, different control points on an object, such as the rim of a cup when drinking or its base when setting it down, can be associated with distinct dynamics. Here we show that opposing dynamic perturbations, which interfere when controlling a single location on an object, can be learned when each is associated with a separate control point. This demonstrates that motor memory formation is linked to control points on the object, rather than the object per se . We also show that the motor system only generates separate memories for different control points if they are linked to different dynamics, allowing efficient use of motor memory. To account for these results, we develop a normative switching state-space model of motor learning, in which the association between cues (control points) and contexts (dynamics) is learned rather than fixed. Our findings uncover an important mechanism through which the motor system generates flexible and dexterous behavior.Item Open Access Accepted version Peer-reviewed Weakly supervised collective feature learning from curated media(2018) Mukuta, Y; Kimura, A; Adrian, DB; Ghahramani, Z; Ghahramani, Zoubin [0000-0002-7464-6475]The current state-of-the-art in feature learning relies on the supervised learning of large-scale datasets consisting of target content items and their respective category labels. However, constructing such large-scale fully-labeled datasets generally requires painstaking manual effort. One possible solution to this problem is to employ community contributed text tags as weak labels, however, the concepts underlying a single text tag strongly depends on the users. We instead present a new paradigm for learning discriminative features by making full use of the human curation process on social networking services (SNSs). During the process of content curation, SNS users collect content items manually from various sources and group them by context, all for their own benefit. Due to the nature of this process, we can assume that (1) content items in the same group share the same semantic concept and (2) groups sharing the same images might have related semantic concepts. Through these insights, we can define human curated groups as weak labels from which our proposed framework can learn discriminative features as a representation in the space of semantic concepts the users intended when creating the groups. We show that this feature learning can be formulated as a problem of link prediction for a bipartite graph whose nodes corresponds to content items and human curated groups, and propose a novel method for feature learning based on sparse coding or network fine-tuning.Item Open Access Published version Peer-reviewed The control of tonic pain by active relief learning.(eLife Sciences Publications, Ltd, 2018-02-27) Zhang, Suyi; Mano, Hiroaki; Lee, Michael; Yoshida, Wako; Kawato, Mitsuo; Robbins, Trevor W; Seymour, Ben; Zhang, Suyi [0000-0001-9028-6265]; Yoshida, Wako [0000-0001-9273-1617]; Seymour, Ben [0000-0003-1724-5832]Tonic pain after injury characterises a behavioural state that prioritises recovery. Although generally suppressing cognition and attention, tonic pain needs to allow effective relief learning to reduce the cause of the pain. Here, we describe a central learning circuit that supports learning of relief and concurrently suppresses the level of ongoing pain. We used computational modelling of behavioural, physiological and neuroimaging data in two experiments in which subjects learned to terminate tonic pain in static and dynamic escape-learning paradigms. In both studies, we show that active relief-seeking involves a reinforcement learning process manifest by error signals observed in the dorsal putamen. Critically, this system uses an uncertainty ('associability') signal detected in pregenual anterior cingulate cortex that both controls the relief learning rate, and endogenously and parametrically modulates the level of tonic pain. The results define a self-organising learning circuit that reduces ongoing pain when learning about potential relief.Item Open Access Published version Peer-reviewed Homeostasis, failure of homeostasis and degenerate ion channel regulation(Elsevier BV, 2018-04) O'Leary, T; O'Leary, Timothy [0000-0002-1029-0158]Most neurons express a wide variety of ion channels with diverse properties, providing a rich toolbox for tuning neural function. Coexpressed channel types are often degenerate: they share overlapping roles in shaping electrophysiological properties. This can allow one set of channels to compensate the role of others, thus making nervous systems robust to perturbations such as channel deletions and mutations, expression noise or environmental disturbances. In tandem, activity-dependent homeostatic mechanisms can actively regulate channel expression to counteract perturbations by sensing changes in physiological activity. However, recent work shows that in spite of degeneracy and homeostatic regulation, the compensatory outcome of a perturbation can be unpredictable. Sometimes a single mutation in an ion channel gene can be catastrophic, while in other contexts a similar loss of function might be compensated. Compensation sometimes fails even when there may be many potential ways to compensate using available channels. Theoretical models show how homeostatic mechanisms that regulate degenerate conductances can fail and even cause pathologies through aberrant compensation.Item Open Access Published version Peer-reviewed Location, number and factors associated with cerebral microbleeds in an Italian-British cohort of CADASIL patients.(Public Library of Science (PLoS), 2018) Nannucci, Serena; Rinnoci, Valentina; Pracucci, Giovanni; MacKinnon, Andrew D; Pescini, Francesca; Adib-Samii, Poneh; Bianchi, Silvia; Dotti, Maria Teresa; Federico, Antonio; Inzitari, Domenico; Markus, Hugh S; Pantoni, Leonardo; Pantoni, Leonardo [0000-0001-7357-8530]BACKGROUND AND PURPOSE: The frequency, clinical correlates, and risk factors of cerebral microbleeds (CMB) in Cerebral Autosomal Dominant Arteriopathy with Subcortical Infarcts and Leukoencephalopathy (CADASIL) are still poorly known. We aimed at determining the location and number of CMB and their relationship with clinical manifestations, vascular risk factors, drugs, and other neuroimaging features in CADASIL patients. METHODS: We collected clinical data by means of a structured proforma and centrally evaluated CMB on magnetic resonance gradient echo sequences applying the Microbleed Anatomical Rating Scale in CADASIL patients seen in 2 referral centers in Italy and United Kingdom. RESULTS: We evaluated 125 patients. CMB were present in 34% of patients and their presence was strongly influenced by the age. Twenty-nine percent of the patients had CMB in deep subcortical location, 22% in a lobar location, and 18% in infratentorial regions. After adjustment for age, factors significantly associated with a higher total number of CMB were hemorrhagic stroke, dementia, urge incontinence, and statins use (this latter not confirmed by multivariate analysis). Infratentorial and deep CMB were associated with dementia and urge incontinence, lobar CMB with hemorrhagic stroke, dementia, and statins use. Unexpectedly, patients with migraine, with or without aura, had a lower total, deep, and lobar number of CMB than patients without migraine. DISCUSSION: CMB formation in CADASIL seems to increase with age. History of hemorrhagic stroke, dementia, urge incontinence, and statins use are associated with a higher number of CMB. However, these findings need to be confirmed by longitudinal studies.Item Open Access Accepted version Peer-reviewed Efficient tagged memory(IEEE, 2017) Joannou, A; Woodruff, J; Kovacsics, R; Moore, SW; Bradbury, A; Xia, H; Watson, RNM; Chisnall, D; Roe, M; Davis, B; Napierala, E; Baldwin, J; Gudka, K; Neumann, PG; Mazzinghi, A; Richardson, A; Son, S; Markettos, AT; Moore, Simon [0000-0002-2806-495X]; Xia, Hongyan [0000-0002-8047-899X]; Richardson, Alexander [0000-0002-6372-217X]We characterize the cache behavior of an in-memory tag table and demonstrate that an optimized implementation can typically achieve a near-zero memory traffic overhead. Both industry and academia have repeatedly demonstrated tagged memory as a key mechanism to enable enforcement of powerful security invariants, including capabilities pointer integrity, watchpoints, and information-flow tracking. A single-bit tag shadowspace is the most commonly proposed requirement, as one bit is the minimum metadata needed to distinguish between an untyped data word and any number of new hardware-enforced types. We survey various tag shadowspace approaches and identify their common requirements and positive features of their implementations. To avoid non-standard memory widths, we identify the most practical implementation for tag storage to be an in-memory table managed next to the DRAM controller. We characterize the caching performance of such a tag table and demonstrate a DRAM traffic overhead below 5\% for the vast majority of applications. We identify spatial locality on a page scale as the primary factor that enables surprisingly high table cache-ability. We then demonstrate tag-table compression for a set of common applications. A hierarchical structure with elegantly simple optimizations reduces DRAM traffic overhead to below 1\% for most applications. These insights and optimizations pave the way for commercial applications making use of single-bit tags stored in commodity memory.Item Open Access Published version Peer-reviewed An error-tuned model for sensorimotor learning(Public Library of Science (PLoS), 2017-12-18) Ingram, James; Sadeghi, Mohsen; Flanagan, John Randall; Wolpert, DM; Sadeghi, Mohsen [0000-0003-2573-146X]; Wolpert, Daniel [0000-0003-2011-2790]Current models of sensorimotor control posit that motor commands are generated by combining multiple modules which may consist of internal models, motor primitives or motor synergies. The mechanisms which select modules based on task requirements and modify their output during learning are therefore critical to our understanding of sensorimotor control. Here we develop a novel modular architecture for multi-dimensional tasks in which a set of fixed primitives are each able to compensate for errors in a single direction in the task space. The contribution of the primitives to the motor output is determined by both top-down contextual information and bottom- up error information. We implement this model for a task in which subjects learn to manipulate a dynamic object whose orientation can vary. In the model, visual information regarding the context (the orientation of the object) allows the appropriate primitives to be engaged. This top-down module selection is implemented by a Gaussian function tuned for the visual orientation of the object. Second, each module's contribution adapts across trials in proportion to its ability to decrease the current kinematic error. Specifically, adaptation is implemented by cosine tuning of primitives to the current direction of the error, which we show to be theoretically optimal for reducing error. This error-tuned model makes two novel predictions. First, interference should occur between alternating dynamics only when the kinematic errors associated with each oppose one another. In contrast, dynamics which lead to orthogonal errors should not interfere. Second, kinematic errors alone should be sufficient to engage the appropriate modules, even in the absence of contextual information normally provided by vision. We confirm both these predictions experimentally and show that the model can also account for data from previous experiments. Our results suggest that two interacting processes account for module selection during sensorimotor control and learning.Item Open Access Published version Peer-reviewed Co-complex protein membership evaluation using Maximum Entropy on GO ontology and InterPro annotation.(Oxford University Press (OUP), 2018-06-01) Armean, Irina M; Lilley, Kathryn S; Trotter, Matthew WB; Pilkington, Nicholas CV; Holden, Sean B; Lilley, Kathryn [0000-0003-0594-6543]; Holden, Sean [0000-0001-7979-1148]MOTIVATION: Protein-protein interactions (PPI) play a crucial role in our understanding of protein function and biological processes. The standardization and recording of experimental findings is increasingly stored in ontologies, with the Gene Ontology (GO) being one of the most successful projects. Several PPI evaluation algorithms have been based on the application of probabilistic frameworks or machine learning algorithms to GO properties. Here, we introduce a new training set design and machine learning based approach that combines dependent heterogeneous protein annotations from the entire ontology to evaluate putative co-complex protein interactions determined by empirical studies. RESULTS: PPI annotations are built combinatorically using corresponding GO terms and InterPro annotation. We use a S.cerevisiae high-confidence complex dataset as a positive training set. A series of classifiers based on Maximum Entropy and support vector machines (SVMs), each with a composite counterpart algorithm, are trained on a series of training sets. These achieve a high performance area under the ROC curve of ≤0.97, outperforming go2ppi-a previously established prediction tool for protein-protein interactions (PPI) based on Gene Ontology (GO) annotations. AVAILABILITY AND IMPLEMENTATION: https://github.com/ima23/maxent-ppi. CONTACT: sbh11@cl.cam.ac.uk. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.Item Open Access Accepted version Peer-reviewed Neural mechanisms underlying valence inferences to sound: The role of the right angular gyrus.(Elsevier BV, 2017-07-28) Bravo, Fernando; Cross, Ian; Hawkins, Sarah; Gonzalez, Nadia; Docampo, Jorge; Bruno, Claudio; Stamatakis, Emmanuel Andreas; Cross, Ian [0000-0002-2404-7765]; Hawkins, Sarah [0000-0003-0591-7929]; Stamatakis, Emmanuel [0000-0001-6955-9601]We frequently infer others' intentions based on non-verbal auditory cues. Although the brain underpinnings of social cognition have been extensively studied, no empirical work has yet examined the impact of musical structure manipulation on the neural processing of emotional valence during mental state inferences. We used a novel sound-based theory-of-mind paradigm in which participants categorized stimuli of different sensory dissonance level in terms of positive/negative valence. Whilst consistent with previous studies which propose facilitated encoding of consonances, our results demonstrated that distinct levels of consonance/dissonance elicited differential influences on the right angular gyrus, an area implicated in mental state attribution and attention reorienting processes. Functional and effective connectivity analyses further showed that consonances modulated a specific inhibitory interaction from associative memory to mental state attribution substrates. Following evidence suggesting that individuals with autism may process social affective cues differently, we assessed the relationship between participants' task performance and self-reported autistic traits in clinically typical adults. Higher scores on the social cognition scales of the AQ were associated with deficits in recognising positive valence in consonant sound cues. These findings are discussed with respect to Bayesian perspectives on autistic perception, which highlight a functional failure to optimize precision in relation to prior beliefs.Item Open Access Published version Peer-reviewed Home-based neurologic music therapy for upper limb rehabilitation with stroke patients at community rehabilitation stage-a feasibility study protocol.(Frontiers Media SA, 2015) Street, Alexander J; Magee, Wendy L; Odell-Miller, Helen; Bateman, Andrew; Fachner, Jorg C; Bateman, Andrew [0000-0002-2547-5921]BACKGROUND: Impairment of upper limb function following stroke is more common than lower limb impairment and is also more resistant to treatment. Several lab-based studies with stroke patients have produced statistically significant gains in upper limb function when using musical instrument playing and techniques where rhythm acts as an external time-keeper for the priming and timing of upper limb movements. METHODS: For this feasibility study a small sample size of 14 participants (3-60 months post stroke) has been determined through clinical discussion between the researcher and study host in order to test for management, feasibility and effects, before planning a larger trial determined through power analysis. A cross-over design with five repeated measures will be used, whereby participants will be randomized into either a treatment (n = 7) or wait list control (n = 7) group. Intervention will take place twice weekly over 6 weeks. The ARAT and 9HPT will be used to measure for quantitative gains in arm function and finger dexterity, pre/post treatment interviews will serve to investigate treatment compliance and tolerance. A lab based EEG case comparison study will be undertaken to explore audio-motor coupling, brain connectivity and neural reorganization with this intervention, as evidenced in similar studies. DISCUSSION: Before evaluating the effectiveness of a home-based intervention in a larger scale study, it is important to assess whether implementation of the trial methodology is feasible. This study investigates the feasibility, efficacy and patient experience of a music therapy treatment protocol comprising a chart of 12 different instrumental exercises and variations, which aims at promoting measurable changes in upper limb function in hemiparetic stroke patients. The study proposes to examine several new aspects including home-based treatment and dosage, and will provide data on recruitment, adherence and variability of outcomes.