arXiv subjects

Minghui Liu

Publications and source records attributed to Minghui Liu.

45 records · Page 3Linked to original sources

arXiv2021

Feature Mining: A Novel Training Strategy for Convolutional Neural Network

In this paper, we propose a novel training strategy for convolutional neural network(CNN) named Feature Mining, that aims to strengthen the network's learning of the local feature. Through experiments, we find that semantic contained in different parts of the feature is different, while the network will inevitably lose the local information during feedforward propagation. In order to enhance the learning of local feature, Feature Mining divides the complete feature into two complementary parts and reuse these divided feature to make the network learn more local information, we call the two steps as feature segmentation and feature reusing. Feature Mining is a parameter-free method and has plug-and-play nature, and can be applied to any CNN models. Extensive experiments demonstrate the wide applicability, versatility, and compatibility of our method.

cs.CV↗

arXiv2021

Learning from Miscellaneous Other-Class Words for Few-shot Named Entity Recognition

Few-shot Named Entity Recognition (NER) exploits only a handful of annotations to identify and classify named entity mentions. Prototypical network shows superior performance on few-shot NER. However, existing prototypical methods fail to differentiate rich semantics in other-class words, which will aggravate overfitting under few shot scenario. To address the issue, we propose a novel model, Mining Undefined Classes from Other-class (MUCO), that can automatically induce different undefined classes from the other class to improve few-shot NER. With these extra-labeled undefined classes, our method will improve the discriminative ability of NER classifier and enhance the understanding of predefined classes with stand-by semantic knowledge. Experimental results demonstrate that our model outperforms five state-of-the-art models in both 1-shot and 5-shots settings on four NER benchmarks. We will release the code upon acceptance. The source code is released on https: //github.com/shuaiwa16/OtherClassNER.git.

cs.CL↗

arXiv2021

Go Small and Similar: A Simple Output Decay Brings Better Performance

Regularization and data augmentation methods have been widely used and become increasingly indispensable in deep learning training. Researchers who devote themselves to this have considered various possibilities. But so far, there has been little discussion about regularizing outputs of the model. This paper begins with empirical observations that better performances are significantly associated with output distributions, that have smaller average values and variances. By audaciously assuming there is causality involved, we propose a novel regularization term, called Output Decay, that enforces the model to assign smaller and similar output values on each class. Though being counter-intuitive, such a small modification result in a remarkable improvement on performance. Extensive experiments demonstrate the wide applicability, versatility, and compatibility of Output Decay.

cs.CV↗

arXiv2021

FocusedDropout for Convolutional Neural Network

In convolutional neural network (CNN), dropout cannot work well because dropped information is not entirely obscured in convolutional layers where features are correlated spatially. Except randomly discarding regions or channels, many approaches try to overcome this defect by dropping influential units. In this paper, we propose a non-random dropout method named FocusedDropout, aiming to make the network focus more on the target. In FocusedDropout, we use a simple but effective way to search for the target-related features, retain these features and discard others, which is contrary to the existing methods. We found that this novel method can improve network performance by making the network more target-focused. Besides, increasing the weight decay while using FocusedDropout can avoid the overfitting and increase accuracy. Experimental results show that even a slight cost, 10\% of batches employing FocusedDropout, can produce a nice performance boost over the baselines on multiple datasets of classification, including CIFAR10, CIFAR100, Tiny Imagenet, and has a good versatility for different CNN models.

cs.CV↗

arXiv2020

Reduction of PDF uncertainty in the measurement of the weak mixing angle at the ATLAS experiment

We investigate the parton distribution function (PDF) uncertainty in the measurement of the effective weak mixing angle $\sin^2θ_{\text{eff}}^{\ell}$ at the CERN Large Hadron Collider (LHC). The PDF-induced uncertainty is large in the proton-proton collisions at the LHC due to the dilution effect. The measurement of the Drell-Yan forward-backward asymmetry ($A_{FB}$) at the LHC can be used to reduce the PDF uncertainty in the $\sin^2θ_{\text{eff}}^{\ell}$ measurement. However, when including the full mass range of lepton pairs in the $A_{FB}$ data analysis, the correlation between the PDF updating procedure and the $\sin^2θ_{\text{eff}}^{\ell}$ extraction leads to a sizable bias in the obtained $\sin^2θ_{\text{eff}}^{\ell}$ value. From our studies, we find that the bias can be significantly reduced by removing Drell-Yan events with invariant mass around the $Z$ pole region, while most of the sensitivity in reducing the PDF uncertainty remains. Furthermore, the lepton charge asymmetry in the $W$ boson events as a function of the rapidity of the charged leptons, $A_\pm(η_\ell)$, is known to be another observable which can be used to reduce the PDF uncertainty in the $\sin^2θ_{\text{eff}}^{\ell}$ measurement. The constraint from $A_\pm(η_\ell)$ is complementary to that from the $A_{FB}$, thus no bias affects the $\sin^2θ_{\text{eff}}^{\ell}$ extraction. The studies are performed using the Error PDF Updating Method Package ({\sc ePump}), which is based on the Hessian updating methods. In this article, the CT14HERA2 PDF set is used as an example.

hep-ex↗

arXiv2018

Quasi-trivial Quandles and Biquandles, Cocycle Enhancements and Link-Homotopy of Pretzel links

We investigate some algebraic structures called quasi-trivial quandles and we use them to study link-homotopy of pretzel links. Precisely, a necessary and sufficient condition for a pretzel link with at least two components being trivial under link-homotopy is given. We also generalize the quasi-trivial quandle idea to the case of biquandles and consider enhancement of the quasi-trivial biquandle cocycle counting invariant by quasi-trivial biquandle cocycles, obtaining invariants of link-homotopy type of links analogous to the quasi-trivial quandle cocycle invariants in Ayumu Inoue's article arXiv:1205.5891.

math.GT↗

arXiv2018

A novel method for lepton energy calibration at Hadron Collider Experiments

This report is to provide a novel method for the lepton energy calibration at Hadron Collider Experiments. The method improves the classic lepton energy calibration procedure widely used at hadron collider experiments. The classic method parameterizes the potential bias in the lepton en- ergy calibration, and determines the value of the parameter by the invariant mass of $Z/γ^*\rightarrow \ell^+\ell^-$ events. The precision of the calibration is dominated by the number of parameters or terms consid- ered in the parameterization, for example, a polynomial extension. With one physics constraint of the reconstructed Z boson mass, the classic procedure can use and determine one parameter. The novel method improves the precision of lepton calibration by introducing more terms in the parameterization. To precisely determine the values of multiple parameters, the method first ac- quires more constraints by separating the $Z/γ^*\rightarrow \ell^+\ell^-$ samples according to the decay kinematics, and then reduces the correlation between multiple parameters. Since the new method is still using the reconstructed Z boson masses as the only constraints, it is much faster and easier than detailed study of detector simulations.

hep-ex↗

arXiv2017

Exact duals and short certificates of infeasibility and weak infeasibility in conic linear programming

In conic linear programming -- in contrast to linear programming -- the Lagrange dual is not an exact dual: it may not attain its optimal value, or there may be a positive duality gap. The corresponding Farkas' lemma is also not exact (it does not always prove infeasibility). We describe exact duals, and certificates of infeasibility and weak infeasibility for conic LPs which are nearly as simple as the Lagrange dual, but do not rely on any constraint qualification. Some of our exact duals generalize the SDP duals of Ramana, and Klep and Schweighofer to the context of general conic LPs. Some of our infeasibility certificates generalize the row echelon form of a linear system of equations: they consist of a small, trivially infeasible subsystem obtained by elementary row operations. We prove analogous results for weakly infeasible systems. We obtain some fundamental geometric corollaries: an exact characterization of when the linear image of a closed convex cone is closed, and an exact characterization of nice cones. Our infeasibility certificates provide algorithms to generate {\em all} infeasible conic LPs over several important classes of cones; and {\em all} weakly infeasible SDPs in a natural class. Using these algorithms we generate a public domain library of infeasible and weakly infeasible SDPs. The status of our instances can be verified by inspection in exact arithmetic, but they turn out to be challenging for commercial and research codes.

math.OC↗

arXiv2015

Exact duality in semidefinite programming based on elementary reformulations

In semidefinite programming (SDP), unlike in linear programming, Farkas' lemma may fail to prove infeasibility. Here we obtain an exact, short certificate of infeasibility in SDP by an elementary approach: we reformulate any semidefinite system of the form Ai*X = bi (i=1,...,m) (P) X >= 0 using only elementary row operations, and rotations. When (P) is infeasible, the reformulated system is trivially infeasible. When (P) is feasible, the reformulated system has strong duality with its Lagrange dual for all objective functions. As a corollary, we obtain algorithms to generate the constraints of {\em all} infeasible SDPs and the constraints of {\em all} feasible SDPs with a fixed rank maximal solution.

math.OC↗