Reducing repetition in convolutional abstractive summarization

Convolutional sequence to sequence (CNN seq2seq) models have met success in abstractive summarization. However, their outputs often contain repetitive word sequences and logical inconsistencies, limiting the practicality of their application. In this paper, we find the reasons behind the repetition problem in CNN-based abstractive summarization through observing the attention map between the summaries with repetition and their corresponding source documents and mitigate the repetition problem. We propose to reduce the repetition in summaries by attention filter mechanism (ATTF) and sentence-level backtracking decoder (SBD), which dynamically redistributes attention over the input sequence as the output sentences are generated. The ATTF can record previously attended locations in the source document directly and prevent the decoder from attending to these locations. The SBD prevents the decoder from generating similar sentences more than once via backtracking at test. The proposed model outperforms the baselines in terms of ROUGE score, repeatedness, and readability. The results show that this approach generates high-quality summaries with minimal repetition and makes the reading experience better.

Attention-Based LSTM with Filter Mechanism for Entity Relation Classification

Relation classification is an important research area in the field of natural language processing (NLP), which aims to recognize the relationship between two tagged entities in a sentence. The noise caused by irrelevant words and the word distance between the tagged entities may affect the relation classification accuracy. In this paper, we present a novel model multi-head attention long short term memory (LSTM) network with filter mechanism (MALNet) to extract the text features and classify the relation of two entities in a sentence. In particular, we combine LSTM with attention mechanism to obtain the shallow local information and introduce a filter layer based on attention mechanism to strength the available information. Besides, we design a semantic rule for marking the key word between the target words and construct a key word layer to extract its semantic information. We evaluated the performance of our model on SemEval-2010 Task8 dataset and KBP-37 dataset. We achieved an F1-score of 86.3% on SemEval-2010 Task8 dataset and F1-score of 61.4% on KBP-37 dataset, which shows that our method is superior to the previous state-of-the-art methods.

Orienting auditory attention in time: Lateralized alpha power reflects spatio-temporal filtering

The deployment of neural alpha (8-12 Hz) lateralization in service of spatial attention is well-established: Alpha power increases in the cortical hemisphere ipsilateral to the attended hemifield, and decreases in the contralateral hemisphere, respectively. Much less is known about humans’ ability to deploy such alpha lateralization in time, and to thus exploit alpha power as a spatio-temporal filter. Here we show that spatially lateralized alpha power does signify - beyond the direction of spatial attention - the distribution of attention in time and thereby qualifies as a spatio-temporal attentional filter. Participants (N = 20) selectively listened to spoken numbers presented on one side (left vs right), while competing numbers were presented on the other side. Key to our hypothesis, temporal foreknowledge was manipulated via a visual cue, which was either instructive and indicated the to-be-probed number position (70% valid) or neutral. Temporal foreknowledge did guide participants’ attention, as they recognized numbers from the to-be-attended side more accurately following valid cues. In the magnetoencephalogram (MEG), spatial attention to the left versus right side induced lateralization of alpha power in all temporal cueing conditions. Modulation of alpha lateralization at the 0.8-Hz presentation rate of spoken numbers was stronger following instructive compared to neutral temporal cues. Critically, we found stronger modulation of lateralized alpha power specifically at the onsets of temporally cued numbers. These results suggest that the precisely timed hemispheric lateralization of alpha power qualifies as a spatio-temporal attentional filter mechanism susceptible to top-down behavioural goals.

KiMoSys 2.0: an upgraded database for submitting, storing and accessing experimental data for kinetic modeling

The KiMoSys (https://kimosys.org), launched in 2014, is a public repository of published experimental data, which contains concentration data of metabolites, protein abundances and flux data. It offers a web-based interface and upload facility to share data, making it accessible in structured formats, while also integrating associated kinetic models related to the data. In addition, it also supplies tools to simplify the construction process of ODE (Ordinary Differential Equations)-based models of metabolic networks. In this release, we present an update of KiMoSys with new data and several new features, including (i) an improved web interface, (ii) a new multi-filter mechanism, (iii) introduction of data visualization tools, (iv) the addition of downloadable data in machine-readable formats, (v) an improved data submission tool, (vi) the integration of a kinetic model simulation environment and (vii) the introduction of a unique persistent identifier system. We believe that this new version will improve its role as a valuable resource for the systems biology community. Database URL: www.kimosys.org

Lyot-based Multi-wavelength Fiber Laser

A multi-wavelength fiber laser which is based on a Lyot filter is experimentally demonstrated. A combination of four-wave mixing in a highly nonlinear fiber and Lyot filter mechanism in the laser cavity is able to generate multi-wavelength with relatively high extinction ratio (ER). At the input current of 100mA, six laser lines with ER more than 5 dB are successfully generated. The wavelength spacing for the multi-wavelength is 0.15nm, corresponding to the characteristics of the Lyot filter used.

The Cellular Expression and Genetics of Purple Body (Pb) in the Ocular Media of the Guppy Poecilia reticulata

Our study revealed the presence of all major classes of chromatophores (melanophores, xanthophores, erythrophores, violet-blue iridophores, xantho-erythrophores) and crystalline platelets in various combinations in the iris and ocular media (cornea, aqueous humor, vitreous humor, outer lens membrane) of Poecilia reticulata. This novel ocular media study of P. reticulata takes into account the distinct interactions of Purple Body (Pb) based on results of previous Bias and Squire Purple Body (Pb) publications. Taken in conjunction with other researcher’s published results (regarding UV reflected color and pattern, vision, mate choice, individual preferences, and opsin capabilities) this indicates that these ocular chromatophore populations together create a complex ocular filter mechanism. This mechanism in turn provides spectral capabilities into the UV and Near-UV wavelengths in both Pb and non-Pb individuals. The chromatophores in the cornea, aqueous humor, covering membranes of the lens, and the vitreous humor comprise an ocular filter system that could reduce UV damage to the internal structures of the eye. The guppy’s ability to use UVA as a visual component provides a “private signally system” that cannot be detected by some predators. While non-Pb guppies should derive benefit in the near-UV from violet-blue iridophore units, greater benefit will be derived by Pb individuals with more violet iridophores functioning in the lower UV and near-UV wavelengths. To our knowledge little has been published for P. reticulata concerning pigmentation within the guppy eye. Macroscopic and microscopic imagery is presented.