Neural sequence models are widely used to model time-series data. Equally ubiquitous is the usage of beam search (BS) as an approximate inference algorithm decode output sequences from these models. BS explores space in a greedy left-right fashion retaining only top-B candidates - resulting that differ slightly each other. Producing lists nearly identical not computationally wasteful but also typically fails capture inherent ambiguity complex AI tasks. To overcome this problem, we propose...

A single image captures the appearance and position of multiple entities in a scene as well their complex interactions. As consequence, natural language grounded visual contexts tends to be diverse---with utterances differing focus shifts specific objects, interactions, or levels detail. Recently, neural sequence models such RNNs LSTMs have been employed produce visually-grounded language. Beam Search, standard work-horse for decoding sequences from these models, is an approximate inference...

The ABCB4 gene encodes multidrug resistance protein 3(MDR3), which is a phosphatidylcholine(PC) transfer enzyme that transfers lecithin from the inner part of phospholipid bilayer to extracellular bile. occurrence intrahepatic cholestasis pregnancy(ICP) closely related variants, but there limited research on this topic in southern Anhui, China. We sequenced pregnant women with ICP and healthy explore relationship. A total 30 patients diagnosed were selected as study objects 90 control group....

We develop the first approximate inference algorithm for 1-Best (and M-Best) decoding in bidirectional neural sequence models by extending Beam Search (BS) to reason about both forward and backward time dependencies. is a widely used sequences from unidirectional models. Interestingly, remains an open problem, despite their significant advantage modeling information past future. To enable use of models, we present Bidirectional (BiBS), efficient inference. evaluate our method as interesting...

This paper studies active learning in structured probabilistic models such as Conditional Random Fields (CRFs). is a challenging problem because unlike unstructured prediction problems binary or multi-class classification, involve distribution with an exponentially-large support, for instance, over the space of all possible segmentations image. Thus, entropy typically intractable to compute. We propose crude yet surprisingly effective histogram approximation Gibbs distribution, which...

Objective To explore alterations in fecal short‐chain fatty acids (SCFA) and gut microbiota patients with diarrhea‐predominant irritable bowel disease (IBS‐D) their relationships clinical manifestations. Methods We recruited 162 IBS‐D 66 healthy controls (HC). Their manifestations psychological status were evaluated using the IBS severity scoring system Hospital Anxiety Depression Scale (HADS). Colorectal visceral sensitivity was a barostat. Systemic inflammation plasma cytokine levels....

Abstract In this paper, we propose a target detection and localization method on distributed monopulse arrays for tracking radar. An optimized mainlobe jamming (MLJ) cancellation filter was designed by maximizing the power ratio of received siackgnal to jamming-plus-noise. By exploiting different correlation characteristics between echo MLJ antennas, is able cancel maintain echo. applying identical sum-difference beams, can be cancelled, maintained simultaneously. Hence, simply detect locate...

Background: Target identification and hit can be transformed through the application of biomedical knowledge analysis, AI-driven virtual screening robotic cloud lab systems. However there are few prospective studies that evaluate efficacy such integrated approaches. Results: We synergistically integrate our in-house-developed target evaluation (SpectraView) deep-learning-driven (HydraScreen) tools with an automated designed explicitly for ultra-high-throughput screening, enabling us to...

Abstract Background Target identification and hit can be transformed through the application of biomedical knowledge analysis, AI-driven virtual screening robotic cloud lab systems. However there are few prospective studies that evaluate efficacy such integrated approaches. Results We synergistically integrate our in-house-developed target evaluation (SpectraView) deep-learning-driven (HydraScreen) tools with an automated designed explicitly for ultra-high-throughput screening, enabling us...

The electrochemical behavior of the V(IV)/V(V) and V(II)/V(Ш) redox couples has been studied at a graphite electrode by cyclic voltammetry in sulfuric acid solutions. diffusion coefficient (D) rate constant (k) for varying state charge (SOC) can be calculated voltammetry. values Da(diffusion anode) ka(the are increased couple while decreased with increasing SOC. Electrochemical impedance spectroscopy (EIS) is employed to study mechanism two medium. Characteristic EIS plots capacitive...

Aiming at this problem that the analysis of instantaneous frequency for radar emitter signal(RES), based on empirical mode decomposition(EMD) intra-pulse feature extraction method RES is studied. Using decomposition normalized energy entropy (NeEn) extracted. The experiment result displays if signal-to-noise ratio greater than 12 dB, shows only small fluctuations.

In this paper, we propose a novel architecture for DWT that can be reconfigured to adapted different kinds of filter banks and sizes inputs. High flexibility generality are achieved by using the MAC loop based filter(MLBF). Classic methods, such as polyphase structure fragment-based sample consumption, used enhance parallelism system. The 3 modes deal with 1-D or 2-D bandwidth throughput requirements.

In this paper, we propose a reconfigurable macro-pipelined systolic architecture (MAPS), which aims to accelerate multiply-accumulate based algorithms by exploiting the temporal parallelism. To illustrate performance, implement 32-PE accelerator on Xilinx ML605 experiment board for matrix multiplication and get peak performance of 51.2 GFLOPS (about 8.0 per PE GHz). demonstrate generality different algorithms, 2-dimensional convolution is also implemented MAPS. Moreover, proposed MAPS has...

Real-time emergency event processing for mega-constellation management requires an approach that can quickly obtain a satisfactory satellite range rescheduling scheme. In this article, deep reinforcement learning (DRL) is proposed rapid on the basis of established scheduling schemes. Three principles with different impacts schemes are first analyzed. A mathematical model both maximizing completion profits dynamic missions and minimizing their scheduled formulated. An efficient DRL-based to...

We develop the first approximate inference algorithm for 1-Best (and M-Best) decoding in bidirectional neural sequence models by extending Beam Search (BS) to reason about both forward and backward time dependencies. is a widely used sequences from unidirectional models. Interestingly, remains an open problem, despite their significant advantage modeling information past future. To enable use of models, we present Bidirectional (BiBS), efficient inference.To evaluate our method as...

Target signal simulation is a key link in radar system and performance evaluation. In this paper, the simulating method to liner-FM target echo put forward by analyzing its characteristics fluctuations. And aiming at monopulse radar, antenna diagram of sum/difference functional block channel are given. The methods easy implement project can be used linear-FM radar.

The signal processing elements and working course of MTI radar are analyzed in this paper, theory research is also done to the element performance double canceller. Aiming at Gaussian power-spectral clutter with differ velocity value variance, paper calculates system quantitatively corresponding improvement factor anti-passive jamming achieved, estimated synthesized.