Search

International audience; Networks, mainly composed of cables, are prone to failures. Faults in cables are initially benign (so-called soft faults) and then evolve over time to a much more severe form of faults as short or open circuits (so-called hard faults). It is thus of prime importance to detect and localize efficiently the soft faults. Residuals may be calculated by comparing online computed transmission coefficients with reference transmission coefficients. In this paper, the sensitivity of the residuals with respect to the fault is studied for a simulated but realistic Yshaped network. The residuals are represented in the residual space. Using intensive simulations, the residuals are analyzed with respect to the position of the fault in a branch, the severity of the fault and the signal-to-noise ratio.

The detection of abnormal events in surveillance footage remains a challenge and has been the scope of various research works. Having observed that the state-of-the-art performance is still unsatisfactory, this paper provides a novel solution to the problem, with four-fold contributions: 1) upon the work of Sultani et al., we introduce one iterative learning framework composed of two experts working in the weak and self-supervised paradigms and providing additional amounts of learning data to each other, where the novel instances at each iteration are filtered by a Bayesian framework that supports the iterative data augmentation task; 2) we describe a novel term that is added to the baseline loss to spread the scores in the unit interval, which is crucial for the performance of the iterative framework; 3) we propose a Random Forest ensemble that fuses at the score level the top performing methods and reduces the EER values about 20% over the state-of-the-art; and 4) we announce the availability of the ”UBI-Fights” dataset, fully annotated at the frame level, that can be freely used by the research community. The code, details of the experimental protocols and the dataset are publicly available at http://github.com/DegardinBruno/ .

It is known that the kinematics of the human body skeleton reveals valuable information in action recognition. Recently, modeling skeletons as spatio-temporal graphs with Graph Convolutional Networks (GCNs) has been reported to solidly advance the state-of-the-art performance. However, GCN-based approaches exclusively learn from raw skeleton data, and are expected to extract the inherent structural information on their own. This paper describes REGINA, introducing a novel way to REasoning Graph convolutional networks IN Human Action recognition. The rationale is to provide to the GCNs additional knowledge about the skeleton data, obtained by handcrafted features, in order to facilitate the learning process, while guaranteeing that it remains fully trainable in an end-to-end manner. The challenge is to capture complementary information over the dynamics between consecutive frames, which is the key information extracted by state-of-the-art GCN techniques. Moreover, the proposed strategy can be easily integrated in the existing GCN-based methods, which we also regard positively. Our experiments were carried out in well known action recognition datasets and enabled to conclude that REGINA contributes for solid improvements in performance when incorporated to other GCN-based approaches, without any other adjustment regarding the original method. For reproducibility, the REGINA code and all the experiments carried out will be publicly available at this https URL.

Ovarian cancer peritoneal metastases (OCPMs) are a pathophysiologically heterogeneous group of tumors that are rarely curable. αvβ3 integrin (αvβ3) is overexpressed on tumoral neovessels and frequently on ovarian cancer cells. Here, using two clinically relevant αvβ3-positive OCPM mouse models, we studied the theranostic potential of an αvβ3-specific radiopeptide, 64Cu-cyclam-RAFT-c(-RGDfK-)4 (64Cu-RaftRGD), and its intra- and intertumoral distribution in relation to the tumor microenvironment.

International audience; Cables are ubiquitous in modern vehicles and the rate of failures has grown with the complexity of the vehicular communication network. The severity of the faults in cables begins in a mild manner, known as soft faults, where the network is still able to perform its required functions. Such faults if not caught in early stages, may develop into hard/severe faults which can lead to serious breakdowns of the whole system. It is thus of prime importance to design and implement an efficient soft fault detection system. In this paper, a bus Ethernet network composed of one source and multiple receivers is studied. A soft fault, which may be caused by a localized degradation of the cable quality, is equivalent to an additional series resistor. Since all the receivers are linked to the same line (the backbone), a fault in the network may affect all the receivers which make it difficult to pinpoint the faulty branch. The transmission transfer function measured between the source and each end of line is monitored. A new health indicator inspired from the classical chain matrix model of the cables is proposed. This indicator is computed at each receiver to detect a soft fault and to locate the faulty branch. Based on the studied topology, a signature matrix is generated, which will be used for fault detection and localization. Simulation results are presented to illustrate our approach.

Remote, precisely controlled activation of the brain is a fundamental challenge in the development of brain-machine interfaces providing feasible rehabilitation strategies for neurological disorders. Low-frequency ultrasound stimulation can be used to modulate neuronal activity deep in the brain1–7, but this approach lacks spatial resolution and cellular selectivity and loads the brain with high levels of acoustic energy. The combination of the expression of ultrasound-sensitive proteins with ultrasound stimulation (‘sonogenetic stimulation’) can provide cellular selectivity and higher sensitivity, but such strategies have been subject to severe limitations in terms of spatiotemporal resolution in vivo8–10, precluding their use for real-life applications. We used the expression of large-conductance mechanosensitive ion channels (MscL) with high-frequency ultrasonic stimulation for a duration of milliseconds to activate neurons selectively at a relatively high spatiotemporal resolution in the rat retina ex vivo and the primary visual cortex of rodents in vivo. This spatiotemporal resolution was achieved at low energy levels associated with negligible tissue heating and far below those leading to complications in ultrasound neuromodulation6,11. We showed, in an associative learning test, that sonogenetic stimulation of the visual cortex generated light perception. Our findings demonstrate that sonogenetic stimulation is compatible with millisecond pattern presentation for visual restoration at the cortical level. They represent a step towards the precise transfer of information over large distances to the cortical and subcortical regions of the brain via an approach less invasive than that associated with current brain-machine interfaces and with a wide range of applications in neurological disorders.

Neural Architecture Search (NAS) methods have been successfully applied to image tasks with excellent results. However, NAS methods are often complex and tend to converge to local minima as soon as generated architectures seem to yield good results. In this paper, we propose G-EA, a novel approach for guided evolutionary NAS. The rationale behind G-EA, is to explore the search space by generating and evaluating several architectures in each generation at initialization stage using a zero-proxy estimator, where only the highest-scoring network is trained and kept for the next generation. This evaluation at initialization stage allows continuous extraction of knowledge from the search space without increasing computation, thus allowing the search to be efficiently guided. Moreover, G-EA forces exploitation of the most performant networks by descendant generation while at the same time forcing exploration by parent mutation and by favouring younger architectures to the detriment of older ones. Experimental results demonstrate the effectiveness of the proposed method, showing that G-EA achieves state-of-the-art results in NAS-Bench-201 search space in CIFAR-10, CIFAR-100 and ImageNet16-120, with mean accuracies of 93.98%, 72.12% and 45.94% respectively., Paper accepted at 35th Conference on Neural Information Processing Systems (NeurIPS) - New In ML. 9 pages, 2 figures, 1 table