Search

Recently, the COMbinatorial Multi-Armed Bandits (COMMAB) problem has arisen as an active research field. In systems interacting with humans, those reinforcement learning approaches use a feedback strategy as their reward function. On the study of those strategies, this paper present three contributions : 1) We model a feedback strategy as a three-step process, namely : Feedback Identification, Feedback Retrieval and Reward Computing, where each step influences the performances of an agent. 2) Based on this model, we propose a novel Reward Computing process, BUSBC, which significantly increases the global accuracyreachedbyoptimisticCOM-MABalgorithms;3) We conduct an empirical analysis on our approach and several feedback strategies from the literature. Our experiments, conducted on three real-world datasets, confirm our propositions with significant results whether full or partial feedback vector are used., Récemment, le problème des Bandits-Manchots COMbinatoires(COM-MAB)a été sujet de nombreux travaux de recherche. Au sein de systèmes en interaction avec des humains, ces techniques, basées sur un apprentissage par renforcement, exploitent une stratégie de considération du retour utilisateur en guise de fonction de récompense. Dans l’étude de ces stratégies, cet article présente les contributions suivantes : 1) Nous proposons un modèle général de stratégie en trois étapes : Feedback Identification, Feedback Retrieval et Reward Computing, chacune influant sur les performances d’un agent; 2) Suivant ce modèle, nous proposons une nouvelle méthode de Reward Computing, BUSBC, améliorant significativement la précision globale des algorithmes optimistes; 3) Nous réalisons une analyse empirique de notre approche et d’autres stratégies issues de la littérature. Nos expérimentations, réalisées sur trois jeux de données issus d’applications réelles, confirment nos propositions avec des retours utilisateurs complets ou partiels.

The extracellular matrix comprises a network of macromolecules such as collagens, proteoglycans and glycoproteins. VWA1 (von Willebrand factor A domain containing 1) encodes a component of the extracellular matrix that interacts with perlecan/collagen VI, appears to be involved in stabilizing extracellular matrix structures, and demonstrates high expression levels in tibial nerve. Vwa1-deficient mice manifest with abnormal peripheral nerve structure/function; however, VWA1 variants have not previously been associated with human disease. By interrogating the genome sequences of 74 180 individuals from the 100K Genomes Project in combination with international gene-matching efforts and targeted sequencing, we identified 17 individuals from 15 families with an autosomal-recessive, non-length dependent, hereditary motor neuropathy and rare biallelic variants in VWA1. A single disease-associated allele p.(G25Rfs*74), a 10-bp repeat expansion, was observed in 14/15 families and was homozygous in 10/15. Given an allele frequency in European populations approaching 1/1000, the seven unrelated homozygote individuals ascertained from the 100K Genomes Project represents a substantial enrichment above expected. Haplotype analysis identified a shared 220 kb region suggesting that this founder mutation arose >7000 years ago. A wide age-range of patients (6-83 years) helped delineate the clinical phenotype over time. The commonest disease presentation in the cohort was an early-onset (mean 2.0 ± 1.4 years) non-length-dependent axonal hereditary motor neuropathy, confirmed on electrophysiology, which will have to be differentiated from other predominantly or pure motor neuropathies and neuronopathies. Because of slow disease progression, ambulation was largely preserved. Neurophysiology, muscle histopathology, and muscle MRI findings typically revealed clear neurogenic changes with single isolated cases displaying additional myopathic process. We speculate that a few findings of myopathic changes might be secondary to chronic denervation rather than indicating an additional myopathic disease process. Duplex reverse transcription polymerase chain reaction and immunoblotting using patient fibroblasts revealed that the founder allele results in partial nonsense mediated decay and an absence of detectable protein. CRISPR and morpholino vwa1 modelling in zebrafish demonstrated reductions in motor neuron axonal growth, synaptic formation in the skeletal muscles and locomotive behaviour. In summary, we estimate that biallelic variants in VWA1 may be responsible for up to 1% of unexplained hereditary motor neuropathy cases in Europeans. The detailed clinical characterization provided here will facilitate targeted testing on suitable patient cohorts. This novel disease gene may have previously evaded detection because of high GC content, consequential low coverage and computational difficulties associated with robustly detecting repeat-expansions. Reviewing previously unsolved exomes using lower QC filters may generate further diagnoses.

Nowadays, in most fields of activities, companies are strengthening their digitization process and offer new services to their users. In recent years, many of these services have relied on machine learning techniques. Concerning combinatorial multi-armed bandit algorithms, which are particularly employed for recommendation, user feedbacks play a crucial role for online learning. However, strategies for considering those feedbacks are essentially based on the observation of a full rewards vector which can be hard to acquire when users must be directly and too frequently solicited. Herein, we propose a novel approach which overcomes these limitations, while providing a level of global accuracy similar to that obtained by classical competitive methods., Aujourd'hui, dans de nombreux secteurs d'activités, les entreprises renforcent leur numérisation et proposent de nouveaux services à leurs usagers. Ces dernières années, nombre de ces services ont reposé sur des techniques d'ap-prentissage automatique. Pour les algorithmes de bandits-manchots combinatoires, particulièrement employés pour la recommandation, le retour utilisateur joue un rôle crucial dans l'apprentissage en ligne. Cependant, les straté-gies de prise en compte de ce retour reposent essentielle-ment sur l'observation d'un vecteur de récompenses com-plet. Celui-ci reste difficile à acquérir lorsque l'utilisateur doit être directement et trop fréquemment sollicité. Dans cet article, nous proposons une nouvelle approche permet-tant de pallier cette problématique et maintenant une pré-cision globale proche de celles des méthodes classiques.

See Arribat (doi.10.1093/brain/awaa464) for a scientific commentary on this article. In a series of 17 individuals with neuropathy and rare biallelic variants in VWA1, Pagnamenta et al. identify a single frameshift present in the majority of patients. Haplotype analysis suggests that this founder mutation, which is likely to have evaded detection due to its high GC content, may have arisen >7000 years ago., The extracellular matrix comprises a network of macromolecules such as collagens, proteoglycans and glycoproteins. VWA1 (von Willebrand factor A domain containing 1) encodes a component of the extracellular matrix that interacts with perlecan/collagen VI, appears to be involved in stabilizing extracellular matrix structures, and demonstrates high expression levels in tibial nerve. Vwa1-deficient mice manifest with abnormal peripheral nerve structure/function; however, VWA1 variants have not previously been associated with human disease. By interrogating the genome sequences of 74 180 individuals from the 100K Genomes Project in combination with international gene-matching efforts and targeted sequencing, we identified 17 individuals from 15 families with an autosomal-recessive, non-length dependent, hereditary motor neuropathy and rare biallelic variants in VWA1. A single disease-associated allele p.(G25Rfs*74), a 10-bp repeat expansion, was observed in 14/15 families and was homozygous in 10/15. Given an allele frequency in European populations approaching 1/1000, the seven unrelated homozygote individuals ascertained from the 100K Genomes Project represents a substantial enrichment above expected. Haplotype analysis identified a shared 220 kb region suggesting that this founder mutation arose >7000 years ago. A wide age-range of patients (6–83 years) helped delineate the clinical phenotype over time. The commonest disease presentation in the cohort was an early-onset (mean 2.0 ± 1.4 years) non-length-dependent axonal hereditary motor neuropathy, confirmed on electrophysiology, which will have to be differentiated from other predominantly or pure motor neuropathies and neuronopathies. Because of slow disease progression, ambulation was largely preserved. Neurophysiology, muscle histopathology, and muscle MRI findings typically revealed clear neurogenic changes with single isolated cases displaying additional myopathic process. We speculate that a few findings of myopathic changes might be secondary to chronic denervation rather than indicating an additional myopathic disease process. Duplex reverse transcription polymerase chain reaction and immunoblotting using patient fibroblasts revealed that the founder allele results in partial nonsense mediated decay and an absence of detectable protein. CRISPR and morpholino vwa1 modelling in zebrafish demonstrated reductions in motor neuron axonal growth, synaptic formation in the skeletal muscles and locomotive behaviour. In summary, we estimate that biallelic variants in VWA1 may be responsible for up to 1% of unexplained hereditary motor neuropathy cases in Europeans. The detailed clinical characterization provided here will facilitate targeted testing on suitable patient cohorts. This novel disease gene may have previously evaded detection because of high GC content, consequential low coverage and computational difficulties associated with robustly detecting repeat-expansions. Reviewing previously unsolved exomes using lower QC filters may generate further diagnoses.

Histone lysine methylation, mediated by mixed-lineage leukemia (MLL) proteins, is now known to be critical in the regulation of gene expression, genomic stability, cell cycle and nuclear architecture. Despite MLL proteins being postulated as essential for normal development, little is known about the specific functions of the different MLL lysine methyltransferases. Here we report heterozygous variants in the gene KMT2B (also known as MLL4) in 27 unrelated individuals with a complex progressive childhood-onset dystonia, often associated with a typical facial appearance and characteristic brain magnetic resonance imaging findings. Over time, the majority of affected individuals developed prominent cervical, cranial and laryngeal dystonia. Marked clinical benefit, including the restoration of independent ambulation in some cases, was observed following deep brain stimulation (DBS). These findings highlight a clinically recognizable and potentially treatable form of genetic dystonia, demonstrating the crucial role of KMT2B in the physiological control of voluntary movement.

SummaryBackgroundIf carotid artery narrowing remains asymptomatic (ie, has caused no recent stroke or other neurological symptoms), successful carotid endarterectomy (CEA) reduces stroke incidence for some years. We assessed the long-term effects of successful CEA.MethodsBetween 1993 and 2003, 3120 asymptomatic patients from 126 centres in 30 countries were allocated equally, by blinded minimised randomisation, to immediate CEA (median delay 1 month, IQR 0·3–2·5) or to indefinite deferral of any carotid procedure, and were followed up until death or for a median among survivors of 9 years (IQR 6–11). The primary outcomes were perioperative mortality and morbidity (death or stroke within 30 days) and non-perioperative stroke. Kaplan-Meier percentages and logrank p values are from intention-to-treat analyses. This study is registered, number ISRCTN26156392.Findings1560 patients were allocated immediate CEA versus 1560 allocated deferral of any carotid procedure. The proportions operated on while still asymptomatic were 89·7% versus 4·8% at 1 year (and 92·1% vs 16·5% at 5 years). Perioperative risk of stroke or death within 30 days was 3·0% (95% CI 2·4–3·9; 26 non-disabling strokes plus 34 disabling or fatal perioperative events in 1979 CEAs). Excluding perioperative events and non-stroke mortality, stroke risks (immediate vs deferred CEA) were 4·1% versus 10·0% at 5 years (gain 5·9%, 95% CI 4·0–7·8) and 10·8% versus 16·9% at 10 years (gain 6·1%, 2·7–9·4); ratio of stroke incidence rates 0·54, 95% CI 0·43–0·68, p