Retrieval Augmentation for Deep Neural Networks

Deep neural networks have achieved state-of-the-art results in various vision and/or language tasks. Despite the use of large training datasets, most models are trained by iterating over single input-output pairs, discarding the remaining examples for the current prediction. In this work, we actively exploit the training data, using the information from nearest training examples to aid the prediction both during training and testing. Specifically, our approach uses the target of the most similar training example to initialize the memory state of an LSTM model, or to guide attention mechanisms. We apply this approach to image captioning and sentiment analysis, respectively through image and text retrieval. Results confirm the effectiveness of the proposed approach for the two tasks, on the widely used Flickr8 and IMDB datasets. Our code is publicly available at http://github.com/RitaRamo/retrieval-augmentation-nn.

Carbon stars as standard candles: II. The median J magnitude as a distance indicator

We introduce a new distance determination method using carbon-rich asymptotic giant branch stars (CS) as standard candles and the Large and Small Magellanic Clouds (LMC and SMC) as the fundamental calibrators. We select the samples of CS from the ($(J-K_{s})_0$, $J_0$) colour-magnitude diagrams, as, in this combination of filters, CS are bright and easy to identify. We fit the CS $J$-band luminosity functions using a Lorentzian distribution modified to allow the distribution to be asymmetric. We use the parameters of the best-fit distribution to determine if the CS luminosity function of a given galaxy resembles that of the LMC or SMC. Based on this resemblance, we use either the LMC or SMC as the calibrator and estimate the distance to the given galaxy using the median $J$ magnitude ($overline{J}$) of the CS samples. We apply this new method to the two Local Group galaxies NGC 6822 and IC 1613. We find that NGC 6822 has an LMC-like CS luminosity function while IC 1613 is more SMC-like. Using the values for the median absolute $J$ magnitude for the LMC and SMC found in Paper I we find a distance modulus of $mu_{0}=23.54pm0.03$ (stat) for NGC 6822 and $mu_{0}=24.34pm0.05$ (stat) for IC 1613.

New limits on neutrino magnetic moments from low energy neutrino data

Here we give a brief review on the current bounds on the general Majorana transition neutrino magnetic moments (TNMM) which cover also the conventional neutrino magnetic moments (NMM). Leptonic CP phases play a key role in constraining TNMMs. While the Borexino experiment is the most sensitive to the TNMM magnitudes, one needs complementary information from reactor and accelerator experiments in order to probe the complex CP phases.