We propose associative domain adaptation, a novel technique for end-to-end domain adaptation with neural networks, the task of inferring class labels for an unlabeled target domain based on the statistical properties of a labeled source domain. Our training scheme follows the paradigm that in order to effectively derive class labels for the target domain, a network should produce statistically domain invariant embeddings, while minimizing the classification error on the labeled source domain. We accomplish this by reinforcing associations between source and target data directly in embedding space. Our method can easily be added to any existing classification network with no structural and almost no computational overhead. We demonstrate the effectiveness of our approach on various benchmarks and achieve state-of-the-art results across the board with a generic convolutional neural network architecture not specifically tuned to the respective tasks. Finally, we show that the proposed association loss produces embeddings that are more effective for domain adaptation compared to methods employing maximum mean discrepancy as a similarity measure in embedding space.

In many real-world scenarios, labeled data for a specific machine learning task is costly to obtain. Semi-supervised training methods make use of abundantly available unlabeled data and a smaller number of labeled examples. We propose a new framework for semi-supervised training of deep neural networks inspired by learning in humans. "Associations" are made from embeddings of labeled samples to those of unlabeled ones and back. The optimization schedule encourages correct association cycles that end up at the same class from which the association was started and penalizes wrong associations ending at a different class. The implementation is easy to use and can be added to any existing end-to-end training setup. We demonstrate the capabilities of learning by association on several data sets and show that it can improve performance on classification tasks tremendously by making use of additionally available unlabeled data. In particular, for cases with few labeled data, our training scheme outperforms the current state of the art on SVHN.

Optical microcavities are a powerful tool to enhance spontaneous emission of individual quantum emitters. However, the broad emission spectra encountered in the solid state at room temperature limit the influence of a cavity, and call for ultra-small mode volume. We demonstrate Purcellenhanced single photon emission from nitrogen-vacancy (NV) centers in nanodiamonds coupled to a tunable fiber-based microcavity with a mode volume down to 1.0 λ 3 . We record cavity-enhanced fluorescence images and study several single emitters with one cavity. The Purcell effect is evidenced by enhanced fluorescence collection, as well as tunable fluorescence lifetime modification, and we infer an effective Purcell factor of up to 2.0. With numerical simulations, we furthermore show that a novel regime for light confinement can be achieved, where a Fabry-Perot mode is combined with additional mode confinement by the nanocrystal itself. In this regime, effective Purcell factors of up to 11 for NV centers and 63 for silicon vacancy centers are feasible, holding promise for bright single photon sources and efficient spin readout under ambient conditions. II. ULTRA-SMALL MODE VOLUME TUNABLE CAVITYThe microcavity is assembled of a planar mirror onto which the sample is applied, and a concave mirror on the tip of an optical fiber [26], see Fig. 1(a). The tip of the cavity fiber is shaped by advanced CO 2 laser machining [34,35]. In a first step, the extent of the endfacet is tapered to enable sub-micron mirror separations, arXiv:1606.00167v2 [quant-ph]

No abstract

Ablation strategies remain poorly defined for persistent atrial fibrillation (AF) patients with recurrence despite intact pulmonary vein isolation (PVI). As the ability to perform durable PVI improves, the need for advanced mapping to identify extra-PV sources of AF becomes increasingly evident. Multiple mapping technologies attempt to localize these self-sustained triggers and/or drivers responsible for initiating and/or maintaining AF; however, current approaches suffer from technical limitations. Electrographic flow (EGF) mapping is a novel mapping method based on well-established principles of optical flow and fluid dynamics. It enables the full spatiotemporal reconstruction of organized wavefront propagation within the otherwise chaotic and disorganized electrical conduction of AF. Given the novelty of EGF mapping and relative unfamiliarity of most clinical electrophysiologists with the mathematical principles powering the EGF algorithm, this paper provides an in-depth explanation of the technical/mathematical foundations of EGF mapping and demonstrates clinical applications of EGF mapping data and analyses.