Search

Recent work on enhancing BERT-based language representation models with knowledge graphs (KGs) and knowledge bases (KBs) has yielded promising results on multiple NLP tasks. State-of-the-art approaches typically integrate the original input sentences with KG triples and feed the combined representation into a BERT model. However, as the sequence length of a BERT model is limited, such a framework supports little knowledge other than the original input sentences and is thus forced to discard some knowledge. This problem is especially severe for downstream tasks for which the input is a long paragraph or even a document, such as QA or reading comprehension tasks. We address this problem with Roof-Transformer, a model with two underlying BERTs and a fusion layer on top. One underlying BERT encodes the knowledge resources and the other one encodes the original input sentences, and the fusion layer integrates the two resultant encodings. Experimental results on a QA task and the GLUE benchmark attest the effectiveness of the proposed model.

We present a field-based method of toolpath generation for 3D printing continuous fibre reinforced thermoplastic composites. Our method employs the strong anisotropic material property of continuous fibres by generating toolpaths along the directions of tensile stresses in the critical regions. Moreover, the density of toolpath distribution is controlled in an adaptive way proportionally to the values of stresses. Specifically, a vector field is generated from the stress tensors under given loads and processed to have better compatibility between neighboring vectors. An optimal scalar field is computed later by making its gradients approximate the vector field. After that, isocurves of the scalar field are extracted to generate the toolpaths for continuous fibre reinforcement, which are also integrated with the boundary conformal toolpaths in user selected regions. The performance of our method has been verified on a variety of models in different loading conditions. Experimental tests are conducted on specimens by 3D printing continuous carbon fibres (CCF) in a polylactic acid (PLA) matrix. Compared to reinforcement by load-independent toolpaths, the specimens fabricated by our method show up to 71.4% improvement on the mechanical strength in physical tests when using the same (or even slightly smaller) amount of continuous fibres.

This study investigates the ductile-to-brittle transition behavior in elliptical vibration cutting (EVC) of silicon and identifies the practical process window for ductile-regime cutting. EVC has been reported to increase the critical depth of ductile-regime cutting of silicon. This study demonstrates that the enhanced ductile cutting performance, however, is only optimal in a carefully-determined process window. The vibration amplitudes and nominal cutting velocity have significant impacts on the ductile-to-brittle transition behaviors. Systematic experiments covering a wide span of vibration amplitudes and cutting velocity have been conducted to investigate their effects. Two quantitative performance indices, the critical depth and ductile ratio, are utilized to analyze cutting performance by considering two unique characteristics of elliptical vibration cutting, i.e., the time-varying undeformed chip thickness and effective cutting direction angle. The results show that there exists a lower bound for the nominal cutting velocity to ensure the ductile-regime material removal, besides the well-known upper bound. Besides, the increases of vibration amplitudes in both the cutting and depth-of-cut (DOC) directions first enhance but then deteriorate the cutting performance. Based on the theoretical analysis and experimental results, the optimal process parameters have been recommended for the elliptical vibration cutting of silicon.

A central task of a Disc Jockey (DJ) is to create a mixset of mu-sic with seamless transitions between adjacent tracks. In this paper, we explore a data-driven approach that uses a generative adversarial network to create the song transition by learning from real-world DJ mixes. In particular, the generator of the model uses two differentiable digital signal processing components, an equalizer (EQ) and a fader, to mix two tracks selected by a data generation pipeline. The generator has to set the parameters of the EQs and fader in such away that the resulting mix resembles real mixes created by humanDJ, as judged by the discriminator counterpart. Result of a listening test shows that the model can achieve competitive results compared with a number of baselines., Comment: To be published at ICASSP 2022

Additive manufacturing has been adopted to process nickel-titanium shape memory alloys due to its advantages of flexibility and minimal defects. The current layer-by-layer method is accompanied by a complex temperature history, which is not beneficial to the final characteristics of shape memory alloys. In this study, a continuous directed energy deposition method has been proposed to improve microstructure uniformity. The spatial characterization of nickel-titanium shape memory alloy fabricated by continuous directed energy deposition is investigated to study the temperature history, phase constituent, microstructure, and mechanical properties. The results indicate that the fabricated specimen has a monotonic temperature history, relatively uniform phase distribution and microstructure morphology, as well as high compressive strength (2982 MPa~3105 MPa) and strain (37.7%~41.1%). The reported method is expected to lay the foundation for spatial control during the printing of functional structures.

Fabrication of nickel-titanium shape memory alloy through additive manufacturing has attracted increasing interest due to its advantages of flexible manufacturing capability, low-cost customization, and minimal defects. The process parameters in directed energy deposition (DED) have a crucial impact on its molten pool characteristics (geometry, microstructure, etc.), thus influencing the final properties of shape memory effect and pseudoelasticity. In this paper, a three-dimensional numerical model considering heat transfer, phase change, and fluid flow has been developed to simulate the cladding geometry, melt pool depth, and deposition rate. The experimental and simulated results indicated that laser power plays a critical role in determining the melt pool width and deposition rate while scan speed and powder feed rate have less effect on cladding geometry and deposition rate. The fluid velocity has a huge influence on the distribution of elements in the molten pool. The temperature gradient G, solidification rate R, as well as shape factor G/R were calculated to illustrate the underlying mechanisms of grain structure evolution. The grain morphology distribution of cross-section from the experimental samples agreed well with the simulation results. The model reported in this paper is expected to shed light on the optimization of the deposition process and grain structure prediction.

We explore the possibility that the dark matter relic density is not produced by thermal mechanism directly, but by the decay of other heavier dark sector particles which on the other hand can be produced by the thermal freeze-out mechanism. Using a concrete model with a light dark matter from dark sector decay, we study the collider signature of the dark sector particles in association with Higgs production processes. We find that the future lepton colliders can be a better place to probe the signature of this kind of light dark matter model than the hadron collider such as LHC. Meanwhile, it is found that a Higgs factory with center of mass energy 250 GeV has a better potential to resolve the signature of this kind of light dark matter model than the Higgs factory with center of mass energy 350 GeV., Comment: 15 pages, 14 figures

Federated learning (FL) for medical image segmentation becomes more challenging in multi-task settings where clients might have different categories of labels represented in their data. For example, one client might have patient data with "healthy'' pancreases only while datasets from other clients may contain cases with pancreatic tumors. The vanilla federated averaging algorithm makes it possible to obtain more generalizable deep learning-based segmentation models representing the training data from multiple institutions without centralizing datasets. However, it might be sub-optimal for the aforementioned multi-task scenarios. In this paper, we investigate heterogeneous optimization methods that show improvements for the automated segmentation of pancreas and pancreatic tumors in abdominal CT images with FL settings., Comment: Accepted by MICCAI DCL Workshop 2021

The probabilistic reachability problems of nondeterministic systems are studied. Based on the existing studies, the definition of probabilistic reachable sets is generalized by taking into account time-varying target set and obstacle. A numerical method is proposed to compute probabilistic reachable sets. First, a scalar function in the state space is constructed by backward recursion and grid interpolation, and then the probability reachable set is represented as a nonzero level set of this scalar function. In addition, based on the constructed scalar function, the optimal control policy can be designed. At the end of this paper, some examples are taken to illustrate the validity and accuracy of the proposed method., Comment: 12 pages, 5 figures