arXiv subjects

Renjun Xu

Publications and source records attributed to Renjun Xu.

30 records · Page 2Linked to original sources

arXiv2021

AdaRNN: Adaptive Learning and Forecasting of Time Series

Time series has wide applications in the real world and is known to be difficult to forecast. Since its statistical properties change over time, its distribution also changes temporally, which will cause severe distribution shift problem to existing methods. However, it remains unexplored to model the time series in the distribution perspective. In this paper, we term this as Temporal Covariate Shift (TCS). This paper proposes Adaptive RNNs (AdaRNN) to tackle the TCS problem by building an adaptive model that generalizes well on the unseen test data. AdaRNN is sequentially composed of two novel algorithms. First, we propose Temporal Distribution Characterization to better characterize the distribution information in the TS. Second, we propose Temporal Distribution Matching to reduce the distribution mismatch in TS to learn the adaptive TS model. AdaRNN is a general framework with flexible distribution distances integrated. Experiments on human activity recognition, air quality prediction, and financial analysis show that AdaRNN outperforms the latest methods by a classification accuracy of 2.6% and significantly reduces the RMSE by 9.0%. We also show that the temporal distribution matching algorithm can be extended in Transformer structure to boost its performance.

cs.LG↗

arXiv2021

Learning Invariant Representations across Domains and Tasks

Being expensive and time-consuming to collect massive COVID-19 image samples to train deep classification models, transfer learning is a promising approach by transferring knowledge from the abundant typical pneumonia datasets for COVID-19 image classification. However, negative transfer may deteriorate the performance due to the feature distribution divergence between two datasets and task semantic difference in diagnosing pneumonia and COVID-19 that rely on different characteristics. It is even more challenging when the target dataset has no labels available, i.e., unsupervised task transfer learning. In this paper, we propose a novel Task Adaptation Network (TAN) to solve this unsupervised task transfer problem. In addition to learning transferable features via domain-adversarial training, we propose a novel task semantic adaptor that uses the learning-to-learn strategy to adapt the task semantics. Experiments on three public COVID-19 datasets demonstrate that our proposed method achieves superior performance. Especially on COVID-DA dataset, TAN significantly increases the recall and F1 score by 5.0% and 7.8% compared to recently strong baselines. Moreover, we show that TAN also achieves superior performance on several public domain adaptation benchmarks.

eess.IV↗

arXiv2020

Learning to Match Distributions for Domain Adaptation

When the training and test data are from different distributions, domain adaptation is needed to reduce dataset bias to improve the model's generalization ability. Since it is difficult to directly match the cross-domain joint distributions, existing methods tend to reduce the marginal or conditional distribution divergence using predefined distances such as MMD and adversarial-based discrepancies. However, it remains challenging to determine which method is suitable for a given application since they are built with certain priors or bias. Thus they may fail to uncover the underlying relationship between transferable features and joint distributions. This paper proposes Learning to Match (L2M) to automatically learn the cross-domain distribution matching without relying on hand-crafted priors on the matching loss. Instead, L2M reduces the inductive bias by using a meta-network to learn the distribution matching loss in a data-driven way. L2M is a general framework that unifies task-independent and human-designed matching features. We design a novel optimization algorithm for this challenging objective with self-supervised label propagation. Experiments on public datasets substantiate the superiority of L2M over SOTA methods. Moreover, we apply L2M to transfer from pneumonia to COVID-19 chest X-ray images with remarkable performance. L2M can also be extended in other distribution matching applications where we show in a trial experiment that L2M generates more realistic and sharper MNIST samples.

cs.LG↗

arXiv2015

Weierstrass cycles and tautological rings in various moduli spaces of algebraic curves

We analyze Weierstrass cycles and tautological rings in moduli space of smooth algebraic curves and in moduli spaces of integral algebraic curves with embedded disks with special attention to moduli spaces of curves having genus $\leq 6$. In particular, we show that our general formula gives a good estimate for the dimension of Weierstrass cycles for lower genera.

math.AG↗

arXiv2014

Integral invariants in flat superspace

We are solving for the case of flat superspace some homological problems that were formulated by Berkovits and Howe. (Our considerations can be applied also to the case of supertorus.) These problems arise in the attempt to construct integrals invariant with respect to supersymmetry. They appear also in other situations, in particular, in the pure spinor formalism in supergravity.

hep-th↗

arXiv2013

Cohomology ring of the BRST operator associated to the sum of two pure spinors

In the study of the Type II superstring, it is useful to consider the BRST complex associated to the sum of two pure spinors. The cohomology of this complex is an infinite-dimensional vector space. It is also a finite-dimensional algebra over the algebra of functions of a single pure spinor. In this paper we study the multiplicative structure.

hep-th↗

arXiv2013

BRST cohomology of the sum of two pure spinors

We study the zero mode cohomology of the sum of two pure spinors. The knowledge of this cohomology allows us to better understand the structure of the massless vertex operator of the Type IIB pure spinor superstring.

hep-th↗

arXiv2011

Homology of Lie algebra of supersymmetries and of super Poincare Lie algebra

We study the homology and cohomology groups of super Lie algebra of supersymmetries and of super Poincare Lie algebra in various dimensions. We give complete answers for (non-extended) supersymmetry in all dimensions $\leq 11$. For dimensions $D=10,11$ we describe also the cohomology of reduction of supersymmetry Lie algebra to lower dimensions. Our methods can be applied to extended supersymmetry algebra.

hep-th↗

arXiv2010

Homology of Lie algebra of supersymmetries

We study the homology and cohomology groups of super Lie algebra of supersymmetries and of super Poincare algebra. We discuss in detail the calculation in dimensions D=10 and D=6. Our methods can be applied to extended supersymmetry algebra and to other dimensions.

hep-th↗

arXiv2007

Ab initio investigation of optical properties of high-pressure phases of ice

We report a detailed ab initio investigation on the optical properties of ice under a wide high pressure range. The ice X phase (up to 380GPa), the theoretical proposed higher pressure phase ice XV (300GPa), as well as the ambient pressure low-temperature phase ice XI are involved. Our results show that the dispersion relations of optical properties of ice under high pressure are quite different from those under ambient pressure. Under higher pressure, there is whole tendency of blue shift in all optical properties of ice, and the energy region for optical response become broader, such as the absorption band and reflection band. In addition to the augmented absorption edge, all absorption peaks are found to be rising, and the reflection peaks are also enhanced a bit; hence the transmissivity of ice is inferred to be dropping. The photoconductivity is enhanced, and we explain this behavior from the increase of earner density. The static optical properties are found to be pressure independent, and principally due to the network topology of hydrogen bonding.

cond-mat.mtrl-sci↗

arXiv2007

Ab initio investigation of hydrogen bonding and electronic structure of high-pressure phases of ice

We report a detailed ab initio investigation on hydrogen bonding, geometry, electronic structure, and lattice dynamics of ice under a large high pressure range, including the ice X phase (55-380GPa), the previous theoretically proposed higher-pressure phase ice XIIIM (Refs. 1-2) (380GPa), ice XV (a new structure we derived from ice XIIIM) (300-380GPa), as well as the ambient pressure low-temperature phase ice XI. Different from many other materials, the band gap of ice X is found to be increasing linearly with pressure from 55GPa up to 290GPa, the electronic density of states (DOS) shows that the valence bands have a tendency of red shift (move to lower energies) referring to the Fermi energy while the conduction bands have a blue shift (move to higher energies). This behavior is interpreted as the high pressure induced change of s-p charge transfers between hydrogen and oxygen. It is found that ice X exists in the pressure range from 75GPa to about 290GPa. Beyond 300GPa, a new hydrogen-bonding structure with 50% hydrogen atoms in symmetric positions in O-H-O bonds and the other half being asymmetric, ice XV, is identified. The physical mechanism for this broken symmetry in hydrogen bonding is revealed.

cond-mat.mtrl-sci↗

arXiv2005

Alternative Mathematical Technique to Determine LS Spectral Terms

We presented an alternative computational method for determining the permitted LS spectral terms arising from $l^N$ electronic configurations. This method makes the direct calculation of LS terms possible. Using only basic algebra, we derived our theory from LS-coupling scheme and Pauli exclusion principle. As an application, we have performed the most complete set of calculations to date of the spectral terms arising from $l^N$ electronic configurations, and the representative results were shown. As another application on deducing LS-coupling rules, for two equivalent electrons, we deduced the famous Even Rule; for three equivalent electrons, we derived a new simple rule.

physics.atom-ph↗