This paper investigates continual learning for semantic parsing. In this setting, a neural semantic parser learns tasks sequentially without accessing full training data from previous tasks. Direct application of the SOTA continual learning algorithm
s to this problem fails to achieve comparable performance with re-training models with all seen tasks because they have not considered the special properties of structured outputs yielded by semantic parsers. Therefore, we propose TotalRecall, a continual learning method designed for neural semantic parsers from two aspects: i) a sampling method for memory replay that diversifies logical form templates and balances distributions of parse actions in a memory; ii) a two-stage training method that significantly improves generalization capability of the parsers across tasks. We conduct extensive experiments to study the research problems involved in continual semantic parsing and demonstrate that a neural semantic parser trained with TotalRecall achieves superior performance than the one trained directly with the SOTA continual learning algorithms and achieve a 3-6 times speedup compared to re-training from scratch. Code and datasets are available at: https://github.com/zhuang-li/cl_nsp.
We consider the photo-excitation of charm and bottom pentaquarks with the holographic assignments $[frac 12frac 12^-]_{S=0,1}$ and $[frac 12frac 32^-]_{S=1}$, in the photo-production of heavy vector mesons such as charmonia and bottomonia near thresh
old. We use a Witten diagram to combine the s-channel photo-excitation of holographic pentaquarks with a massive t-channel graviton or tensor glueball exchange, to extract the scattering amplitude for heavy meson photo-production in the threshold region. The pentaquark signal is too weak to be detected at current electron facilities.
We analyze the decay modes of the three $[frac 12frac 12^-]_{S=0,1}$ and $[frac 12frac 32^-]_{S=1}$ non-strange pentaquarks with hidden charm and bottom, predicted by holographic QCD in the heavy quark limit. In leading order, the pentaquarks %are co
mposed of heavy-light mesons in bulk bound to an instanton core. They are degenerate and stable by heavy quark symmetry. At next to leading order, the spin interactions lift the degeneracy and cause the pentaquarks to decay. We show that the open charm (bottom) decay modes dwarf the hidden charm (bottom) ones, with total widths that are consistent with those recently reported by LHCb for charm pentaquarks. Predictions for bottom pentaquarks are given.
We revisit the three non-strange pentaquarks $[frac 12frac 12^-]_{S=0,1}$ and $[frac 12frac 32^-]_{S=1}$ predicted using the holographic dual description, where chiral and heavy quark symmetry are manifest in the triple limit of a large number of c
olors, large quark mass and strong $^prime$t Hooft gauge coupling. In the heavy quark limit, the pentaquarks with internal heavy quark spin $S$ are all degenerate. The holographic pentaquarks are dual to an instanton bound to heavy mesons in bulk, without the shortcomings related to the nature of the interaction and the choice of the hard core inherent to the molecular constructions. We explicitly derive the spin-spin and spin-orbit couplings arising from next to leading order in the heavy quark mass, and lift totally the internal spin degeneray, in fair agreement with the newly reported charmed pentaquarks from LHCb. New charm and bottom pentaquark states are predicted.
Light-front wave functions play a fundamental role in the light-front quantization approach to QCD and hadron structure. However, a naive implementation of the light-front quantization suffers from various subtleties including the well-known zero-mod
e problem, the associated rapidity divergences which mixes ultra-violet divergences with infrared physics, as well as breaking of spatial rotational symmetry. We advocate that the light-front quantization should be viewed as an effective theory in which small $k^+$ modes have been effectively ``integrated out, with an infinite number of renormalization constants. Instead of solving light-front quantized field theories directly, we make the large momentum expansion of the equal-time Euclidean correlation functions in instant quantization as an effective way to systematically calculate light-front correlations, including the light-front wave function amplitudes. This large-momentum effective theory accomplishes an effective light-front quantization through lattice QCD calculations. We demonstrate our approach using an example of a pseudo-scalar meson wave function.
Gluino-SUGRA ($tilde{g}$SUGRA), which is an economical extension of mSUGRA, adopts much heavier gluino mass parameter than other gauginos mass parameters and universal scalar mass parameter at the unification scale. It can elegantly reconcile the exp
erimental results on the Higgs boson mass, the muon $g-2$, the null results in search for supersymmetry at the LHC and the results from B-physics. In this work, we propose several new ways to generate large gaugino hierarchy (i.e. $M_3gg M_1,M_2$) for $tilde{g}$SUGRA model building and then discuss in detail the implications of the new muon $g-2$ results with the updated LHC constraints on such $tilde{g}$SUGRA scenarios. We obtain the following observations: (i) For the most interesting $M_1=M_2$ case at the GUT scale with a viable bino-like dark matter, the $tilde{g}$SUGRA can explain the muon $g-2$ anomaly at $1sigma$ level and be consistent with the updated LHC constraints for $6geq M_3/M_1 geq 9$ at the GUT scale; (ii) For $M_1:M_2=5:1$ at the GUT scale with wino-like dark matter, the $tilde{g}$SUGRA model can explain the muon $g-2$ anomaly at $2sigma$ level and be consistent with the updated LHC constraints for $3geq M_3/M_1 geq 4$ at the GUT scale; (iii) For $M_1:M_2=3:2$ at the GUT scale with mixed bino-wino dark matter, the $tilde{g}$SUGRA model can explain the muon $g-2$ anomaly at $2sigma$ level and be consistent with the updated LHC constraints for $6.7geq M_3/M_1 geq 7.8$ at the GUT scale.
Channel pruning and tensor decomposition have received extensive attention in convolutional neural network compression. However, these two techniques are traditionally deployed in an isolated manner, leading to significant accuracy drop when pursuing
high compression rates. In this paper, we propose a Collaborative Compression (CC) scheme, which joints channel pruning and tensor decomposition to compress CNN models by simultaneously learning the model sparsity and low-rankness. Specifically, we first investigate the compression sensitivity of each layer in the network, and then propose a Global Compression Rate Optimization that transforms the decision problem of compression rate into an optimization problem. After that, we propose multi-step heuristic compression to remove redundant compression units step-by-step, which fully considers the effect of the remaining compression space (i.e., unremoved compression units). Our method demonstrates superior performance gains over previous ones on various datasets and backbone architectures. For example, we achieve 52.9% FLOPs reduction by removing 48.4% parameters on ResNet-50 with only a Top-1 accuracy drop of 0.56% on ImageNet 2012.
We study the anomalous scale symmetry breaking effects on the proton mass in QCD due to quantum fluctuations at ultraviolet scales. We confirm that a novel contribution naturally arises as a part of the proton mass, which we call the quantum anomalou
s energy (QAE). We discuss the QAE origins in both lattice and dimensional regularizations and demonstrate its role as a scheme-and-scale independent component in the mass decomposition. We further argue that QAE role in the proton mass resembles a dynamical Higgs mechanism, in which the anomalous scale symmetry breaking field generates mass scales through its vacuum condensate, as well as its static and dynamical responses to the valence quarks. We demonstrate some of our points in two simpler but closely related quantum field theories, namely the 1+1 dimensional non-linear sigma model in which QAE is non-perturbative and scheme-independent, and QED where the anomalous energy effect is perturbative calculable.
Binary neural networks (BNNs) have received increasing attention due to their superior reductions of computation and memory. Most existing works focus on either lessening the quantization error by minimizing the gap between the full-precision weights
and their binarization or designing a gradient approximation to mitigate the gradient mismatch, while leaving the dead weights untouched. This leads to slow convergence when training BNNs. In this paper, for the first time, we explore the influence of dead weights which refer to a group of weights that are barely updated during the training of BNNs, and then introduce rectified clamp unit (ReCU) to revive the dead weights for updating. We prove that reviving the dead weights by ReCU can result in a smaller quantization error. Besides, we also take into account the information entropy of the weights, and then mathematically analyze why the weight standardization can benefit BNNs. We demonstrate the inherent contradiction between minimizing the quantization error and maximizing the information entropy, and then propose an adaptive exponential scheduler to identify the range of the dead weights. By considering the dead weights, our method offers not only faster BNN training, but also state-of-the-art performance on CIFAR-10 and ImageNet, compared with recent methods. Code can be available at https://github.com/z-hXu/ReCU.
The near threshold photo or electroproduction of heavy vector quarkonium off the proton is studied in quantum chromodynamics. Similar to the high-energy limit, the production amplitude can be factorized in terms of gluonic Generalized Parton Distribu
tions and the quarkonium distribution amplitude. At the threshold, the threshold kinematics has a large skewness parameter $xi$, leading to the dominance of the spin-2 contribution over higher-spin twist-2 operators. Thus threshold production data are useful to extract the gluonic gravitational form factors, allowing studying the gluonic contributions to the quantum anomalous energy, mass radius, spin and mechanical pressure in the proton. We use the recent GlueX data on the $J/psi$ photoproduction to illustrate the potential physics impact from the high-precision data from future JLab 12 GeV and EIC physics program.
