
For each positive integer $n$ we introduce the notion of $n$exangulated
categories as higher dimensional analogues of extriangulated categories defined
by NakaokaPalu. We characterize which $n$exangulated categories are $n$exact
in the sense of Jasso and which are $(n+2)$angulated in the sense of
GeissKellerOppermann. For extriangulated categories with enough projectives
and injectives we introduce the notion of $n$cluster tilting subcategories and
show that under certain conditions such $n$cluster tilting subcategories are
$n$exangulated.

Singlesource and top$k$ SimRank queries are two important types of
similarity search in graphs with numerous applications in web mining, social
network analysis, spam detection, etc. A plethora of techniques have been
proposed for these two types of queries, but very few can efficiently support
similarity search over large dynamic graphs, due to either significant
preprocessing time or large space overheads.
This paper presents ProbeSim, an indexfree algorithm for singlesource and
top$k$ SimRank queries that provides a nontrivial theoretical guarantee in
the absolute error of query results. ProbeSim estimates SimRank similarities
without precomputing any indexing structures, and thus can naturally support
realtime SimRank queries on dynamic graphs. Besides the theoretical guarantee,
ProbeSim also offers satisfying practical efficiency and effectiveness due to
several nontrivial optimizations. We conduct extensive experiments on a number
of benchmark datasets, which demonstrate that our solutions significantly
outperform the existing methods in terms of efficiency and effectiveness.
Notably, our experiments include the first empirical study that evaluates the
effectiveness of SimRank algorithms on graphs with billion edges, using the
idea of pooling.

In this paper, we study the stochastic combinatorial multiarmed bandit
(CMAB) framework that allows a general nonlinear reward function, whose
expected value may not depend only on the means of the input random variables
but possibly on the entire distributions of these variables. Our framework
enables a much larger class of reward functions such as the $\max()$ function
and nonlinear utility functions. Existing techniques relying on accurate
estimations of the means of random variables, such as the upper confidence
bound (UCB) technique, do not work directly on these functions. We propose a
new algorithm called stochastically dominant confidence bound (SDCB), which
estimates the distributions of underlying random variables and their
stochastically dominant confidence bounds. We prove that SDCB can achieve
$O(\log{T})$ distributiondependent regret and $\tilde{O}(\sqrt{T})$
distributionindependent regret, where $T$ is the time horizon. We apply our
results to the $K$MAX problem and expected utility maximization problems. In
particular, for $K$MAX, we provide the first polynomialtime approximation
scheme (PTAS) for its offline problem, and give the first $\tilde{O}(\sqrt T)$
bound on the $(1\epsilon)$approximation regret of its online problem, for any
$\epsilon>0$.

In this paper, we propose a zoomoutandin network for generating object
proposals. A key observation is that it is difficult to classify anchors of
different sizes with the same set of features. Anchors of different sizes
should be placed accordingly based on different depth within a network: smaller
boxes on highresolution layers with a smaller stride while larger boxes on
lowresolution counterparts with a larger stride. Inspired by the conv/deconv
structure, we fully leverage the lowlevel local details and highlevel
regional semantics from two feature map streams, which are complimentary to
each other, to identify the objectness in an image. A map attention decision
(MAD) unit is further proposed to aggressively search for neuron activations
among two streams and attend the most contributive ones on the feature learning
of the final loss. The unit serves as a decisionmaker to adaptively activate
maps along certain channels with the solely purpose of optimizing the overall
training loss. One advantage of MAD is that the learned weights enforced on
each feature channel is predicted onthefly based on the input context, which
is more suitable than the fixed enforcement of a convolutional kernel.
Experimental results on three datasets, including PASCAL VOC 2007, ImageNet
DET, MS COCO, demonstrate the effectiveness of our proposed algorithm over
other stateofthearts, in terms of average recall (AR) for region proposal
and average precision (AP) for object detection.

In this article, we study localizations of hearts of cotorsion pairs (U,V)
where U is rigid on an extriangulated category B. The hearts of such cotorsion
pairs are equivalent to the functor categories over the stable category of U.
Inspired by Marsh and Palu, we consider the mutation of U that induces a
cotorsion pair (U',V'). Generally speaking, the hearts of (U,V) and (U',V') are
not equivalent to each other, but we will give a generalized pseudoMorita
equivalence between certain localizations of their hearts.

The performance of deep learning based semantic segmentation models heavily
depends on sufficient data with careful annotations. However, even the largest
public datasets only provide samples with pixellevel annotations for rather
limited semantic categories. Such data scarcity critically limits scalability
and applicability of semantic segmentation models in real applications. In this
paper, we propose a novel transferable semisupervised semantic segmentation
model that can transfer the learned segmentation knowledge from a few strong
categories with pixellevel annotations to unseen weak categories with only
imagelevel annotations, significantly broadening the applicable territory of
deep segmentation models. In particular, the proposed model consists of two
complementary and learnable components: a Label transfer Network (LNet) and a
Prediction transfer Network (PNet). The LNet learns to transfer the
segmentation knowledge from strong categories to the images in the weak
categories and produces coarse pixellevel semantic maps, by effectively
exploiting the similar appearance shared across categories. Meanwhile, the
PNet tailors the transferred knowledge through a carefully designed
adversarial learning strategy and produces refined segmentation results with
better details. Integrating the LNet and PNet achieves 96.5% and 89.4%
performance of the fullysupervised baseline using 50% and 0% categories with
pixellevel annotations respectively on PASCAL VOC 2012. With such a novel
transfer mechanism, our proposed model is easily generalizable to a variety of
new categories, only requiring imagelevel annotations, and offers appealing
scalability in real applications.

In this work, we investigate the possibility of replacing the $\ell_2$ loss
with perceptually derived loss functions (SSIM, MSSSIM, etc.) in training an
endtoend dehazing neural network. Objective experimental results suggest that
by merely changing the loss function we can obtain significantly higher PSNR
and SSIM scores on the SOTS set in the RESIDE dataset, compared with a
stateoftheart endtoend dehazing neural network (AODNet) that uses the
$\ell_2$ loss. The best PSNR we obtained was 23.50 (4.2% relative improvement),
and the best SSIM we obtained was 0.8747 (2.3% relative improvement.)

Millimeter wave (mmWave) is an attractive candidate for highspeed wireless
communication in the future. However, due to the propagation characteristics of
mmWave, beam alignment becomes a key challenge in mobile environments. In this
paper, we develop a joint 3D recursive beam and channel tracking algorithm that
can achieve low tracking error. We discuss the special challenges in optimazing
the beamforming matrix of 3D tracking. A general asymptotically optimal
beamforming matrix is given to obtain the minimum CramerRao lower bound (CRLB)
of beam and channel tracking. The beamforming matrix can be used in different
channel environments when antenna number is large enough. In static scenarios,
three theorems are developed to prove that the algorithm converges to the
minimum CRLB. Simulation results show that our algorithm outperforms several
existing algorithms.

We experimentally observe Floquet Raman transitions in the weakly driven
solid state spin system of nitrogenvacancy center in diamond. The periodically
driven spin system simulates a twoband WannierStark ladder model, and allows
us to observe coherent spin state transfer arising from Raman transition
mediated by Floquet synthetic levels. It also leads to the prediction of analog
photonassisted Floquet Raman transition and dynamical localisation in a driven
twolevel quantum system. The demonstrated rich Floquet dynamics offers new
capabilities to achieve effective Floquet coherent control of a quantum system
with potential applications in various types of quantum technologies based on
driven quantum dynamics. In particular, the FloquetRaman system may be used as
a quantum simulator for the physics of periodically driven systems.

We report a study of Codoped berthierite Fe$_{1x}$Co$_x$Sb$_2$S$_4$ ($x$ =
0, 0.1, and 0.2). The alloy series of Fe$_{1x}$Co$_x$Sb$_2$S$_4$ crystallize
in an orthorhombic structure with the Pnma space group, similar to FeSb$_2$,
and show semiconducting behavior. The large discrepancy between activation
energy for conductivity, $E_\rho$ (146 $\sim$ 270 meV), and thermopower, $E_S$
(47 $\sim$ 108 meV), indicates the polaronic transport mechanism. Bulk
magnetization and heat capacity measurements of pure FeSb$_2$S$_4$ ($x$ = 0)
exhibit a broad antiferromagnetic (AFM) transition ($T_N$ = 46 K) followed by
an additional weak transition ($T^*$ = 50 K). Transition temperatures ($T_N$
and $T^*$) slightly decrease with increasing Co content $x$. This is also
reflected in the thermal conductivity measurement, indicating strong
spinlattice coupling. Fe$_{1x}$Co$_x$Sb$_2$S$_4$ shows relatively high value
of thermopower (up to $\sim$ 624 $\mu$V K$^{1}$ at 300 K) and thermal
conductivity much lower when compared to FeSb$_{2}$, a feature desired for
potential applications based on FeSb$_{2}$ materials.

We report anomalous Hall effect (AHE) in single crystals of
quasitwodimensional Fe$_{3x}$GeTe$_2$ ($x \approx 0.36$) ferromagnet grown
by the flux method which induces defects on Fe site and bad metallic
resistivity. Fe Kedge xray absorption spectroscopy was measured to provide
information on local atomic environment in such crystals. The dc and ac
magnetic susceptibility measurements indicate a secondstage transition below
119 K in addition to the paramagnetic to ferromagnetic transition at 153 K. A
linear scaling behavior between the modified anomalous Hall resistivity
$\rho_{xy}/\mu_0H_{eff}$ and longitudinal resistivity
$\rho_{xx}^2M/\mu_0H_{eff}$ implies that the AHE in Fe$_{3x}$GeTe$_2$ should
be dominated by the intrinsic KarplusLuttinger mechanism rather than the
extrinsic skewscattering and sidejump mechanisms. The observed deviation in
the linearM Hall conductivity $\sigma_{xy}^A$ below 30 K is in line with its
transport characteristic at low temperatures, implying the scattering of
conduction electrons due to magnetic disorder and the evolution of the Fermi
surface induced by possible spinreorientation transition.

Fully convolutional neural network (FCN) has been dominating the game of face
detection task for a few years with its congenital capability of
slidingwindowsearching with shared kernels, which boiled down all the
redundant calculation, and most recent stateoftheart methods such as
FasterRCNN, SSD, YOLO and FPN use FCN as their backbone. So here comes one
question: Can we find a universal strategy to further accelerate FCN with
higher accuracy, so could accelerate all the recent FCNbased methods? To
analyze this, we decompose the face searching space into two orthogonal
directions, `scale' and `spatial'. Only a few coordinates in the space expanded
by the two base vectors indicate foreground. So if FCN could ignore most of the
other points, the searching space and false alarm should be significantly
boiled down. Based on this philosophy, a novel method named scale estimation
and spatial attention proposal ($S^2AP$) is proposed to pay attention to some
specific scales and valid locations in the image pyramid. Furthermore, we adopt
a maskedconvolution operation based on the attention result to accelerate FCN
calculation. Experiments show that FCNbased method RPN can be accelerated by
about $4\times$ with the help of $S^2AP$ and maskedFCN and at the same time it
can also achieve the stateoftheart on FDDB, AFW and MALF face detection
benchmarks as well.

This paper proposes learning disentangled but complementary face features
with minimal supervision by face identification. Specifically, we construct an
identity Distilling and Dispelling Autoencoder (D2AE) framework that
adversarially learns the identitydistilled features for identity verification
and the identitydispelled features to fool the verification system. Thanks to
the design of twostream cues, the learned disentangled features represent not
only the identity or attribute but the complete input image. Comprehensive
evaluations further demonstrate that the proposed features not only maintain
stateoftheart identity verification performance on LFW, but also acquire
competitive discriminative power for face attribute recognition on CelebA and
LFWA. Moreover, the proposed system is ready to semantically control the face
generation/editing based on various identities and attributes in an
unsupervised manner.

IIIV semiconductor nanowires have shown great potential in various quantum
transport experiments. However, realizing a scalable highquality
nanowirebased platform that could lead to quantum information applications has
been challenging. Here, we study the potential of selective area growth by
molecular beam epitaxy of InAs nanowire networks grown on GaAsbased buffer
layers. The buffered geometry allows for substantial elastic strain relaxation
and a strong enhancement of field effect mobility. We show that the networks
possess strong spinorbit interaction and long phase coherence lengths with a
temperature dependence indicating ballistic transport. With these findings, and
the compatibility of the growth method with hybrid epitaxy, we conclude that
the material platform fulfills the requirements for a wide range of quantum
experiments and applications.

A multiscale model is presented for predicting the magnitude and rate of
powder blending in a rotating drum blender. The model combines particle
diffusion coefficient correlations from the literature with advective flow
field information from blender finite element method simulations. The
multiscale model predictions for overall mixing and local concentration
variance closely match results from discrete element method (DEM) simulations
for a rotating drum, but take only hours to compute as opposed to taking days
of computation time for the DEM simulations. Parametric studies were performed
using the multiscale model to investigate the influence of various parameters
on mixing behavior. The multiscale model is expected to be more amenable to
predicting mixing in complex geometries and scale more efficiently to
industrialscale blenders than DEM simulations or analytical solutions.

The critical properties of the singlecrystalline van der Waals bonded
ferromagnet Fe$_{3x}$GeTe$_2$ were investigated by bulk dc magnetization
around the paramagnetic (PM) to ferromagnetic (FM) phase transition. The
Fe$_{3x}$GeTe$_2$ single crystals grown by selfflux method with Fe deficiency
$x \approx 0.36$ exhibit bulk FM ordering below $T_c = 152$ K. The
M\"{o}ssbauer spectroscopy was used to provide information on defects and local
atomic environment in such crystals. Critical exponents $\beta = 0.372(4)$ with
a critical temperature $T_c = 151.25(5)$ K and $\gamma = 1.265(15)$ with $T_c =
151.17(12)$ K are obtained by the KouvelFisher method whereas $\delta =
4.50(1)$ is obtained by a critical isotherm analysis at $T_c = 151$ K. These
critical exponents obey the Widom scaling relation $\delta = 1+\gamma/\beta$,
indicating selfconsistency of the obtained values. With these critical
exponents the isotherm $M(H)$ curves below and above the critical temperatures
collapse into two independent universal branches, obeying the single scaling
equation $m = f_\pm(h)$, where $m$ and $h$ are renormalized magnetization and
field, respectively. The exponents determined in this study are close to those
calculated from the results of the renormalization group approach for a
heuristic model of threedimensional Heisenberg ($d = 3, n = 3$) spins coupled
with the attractive longrange interactions between spins that decay as
$J(r)\approx r^{(3+\sigma)}$ with $\sigma=1.89$.

The critical properties of fluxgrown singlecrystalline
quasitwodimensional weak itinerant ferromagnet Cr$_{0.62}$Te were
investigated by bulk dc magnetization around the paramagnetic (PM) to
ferromagnetic (FM) phase transition. Critical exponents $\beta = 0.315(7)$ with
a critical temperature $T_c = 230.6(3)$ K and $\gamma = 1.81(2)$ with $T_c =
229.1(1)$ K are obtained by the KouvelFisher method whereas $\delta = 6.35(4)$
is obtained by a critical isotherm analysis at $T_c = 230$ K. With these
obtained exponents, the magnetizationfieldtemperature curves collapse into
two independent curves following a single scaling equation
$M\frac{TT_c}{T_c}^{\beta} = f_\pm(H\frac{TT_c}{T_c}^{\beta\delta})$
around $T_c$, suggesting the reliability of the obtained exponents.
Additionally, the determined exponents of Cr$_{0.62}$Te exhibit an Isinglike
behavior with a change from shortrange order to longrange order in the nature
of magnetic interaction and with an extension from 2D to 3D on cooling through
$T_c$.

CrI$_3$ is a promising candidate for the van der Waals bonded ferromagnetic
devices since its ferromagnetism can be maintained upon exfoliating of bulk
crystals down to single layer. In this work we studied critical properties of
bulk CrI$_3$ single crystals around the paramagnetic to ferromagnetic phase
transition. Critical exponents $\beta$ = 0.260(4) with a critical temperature
$T_c$ = 60.05(13) K and $\gamma$ = 1.136(6) with $T_c$ = 60.43(4) K are
obtained by the KouvelFisher method, whereas $\delta$ = 5.32(2) is obtained by
a critical isotherm analysis at $T_c$ = 60 K. The critical exponents determined
in bulk CrI$_3$ single crystals suggest a threedimensional longrange magnetic
coupling with the exchange distance decaying as $J(r)\approx r^{4.69}$.

Extensive efforts have been undertaken on the photoelectric physics of hybrid
organolead halide perovskites to unveil the reason for the attractive
photovoltaic performance. Yet, the resulting evidences are far from being fully
conclusive. Herein, we provide another direct support for this issue. In
addition to the observation on the conventional band edge at 1.58 eV that
presents a blueshift toward temperature increase, interestingly, we also
observe an unusual optical band edge at 1.48 eV in CH3NH3PbI3 perovskite single
crystals. Contrary to the conventional band edge, this one shows an obvious
redshift toward the enhancement in temperature, in agreement with the Varshni
relation. More interestingly, the unusual band edge exhibits a series of
obvious absorption and photocurrent signals, but the according
photoluminescence signals are not observable. This indicates that this band
edge is particularly beneficial for the photovoltaic effect due to the
inhibited radiative recombination. The kinetics on photoinvolved charge
transition and transfer are investigated using the pumpprobe photoconductivity
technique, and a changeable band structure model was proposed.

The observational analysis is performed to study the excitation mechanism and
the propagation properties of a quasiperiodic fastpropagating (QFP)
magnetosonic wave. The QFP wave was associated with the eruption of a nearby
minifilament and a small B4 GOES flare, which may indicate that the generation
of a QFP wave do not need too much flare energy. The propagation of the QFP
wave was along a bundle of funnelshaped open loops with a speed of about
1100+/78, and an acceleration of 2.2+/1.1. Periodicity analysis indicates
that the periods of the QFP wave are 43+/6, 79+/18 second. For the first
time, we find that the periods of the QFP wave and the accompanying flare are
inconsistent, which is different from the findings as reported in previous
studies. We propose that the present QFP wave was possibly caused by the
mechanism of dispersive evolution of an initially broadband disturbance
resulted from the nearby minifilament eruption.

We report magnetotransport properties of BaZnBi$_{2}$ single crystals.
Whereas electronic structure features Dirac states, such states are removed
from the Fermi level by spinorbit coupling (SOC) and consequently electronic
transport is dominated by the small hole and electron pockets. Our results are
consistent with three dimensional (3D) but also with quasi two dimensional (2D)
portions of the Fermi surface. The spinorbit couplinginduced gap in Dirac
states is much larger when compared to isostructural SrMnBi$_{2}$. This
suggests that not only long range magnetic order but also mass of the alkaline
earth atoms A in ABX$_{2}$ (A = alkaine earth, B = transition metal and
X=Bi/Sb) are important for the presence of lowenergy states obeying the
relativistic Dirac equation at the Fermi surface

Last decade witnesses significant methodological and theoretical advances in
estimating large precision matrices. In particular, there are scientific
applications such as longitudinal data, meteorology and spectroscopy in which
the ordering of the variables can be interpreted through a bandable structure
on the Cholesky factor of the precision matrix. However, the minimax theory has
still been largely unknown, as opposed to the well established minimax results
over the corresponding bandable covariance matrices. In this paper, we focus on
two commonly used types of parameter spaces, and develop the optimal rates of
convergence under both the operator norm and the Frobenius norm. A striking
phenomenon is found: two types of parameter spaces are fundamentally different
under the operator norm but enjoy the same rate optimality under the Frobenius
norm, which is in sharp contrast to the equivalence of corresponding two types
of bandable covariance matrices under both norms. This fundamental difference
is established by carefully constructing the corresponding minimax lower
bounds. Two new estimation procedures are developed: for the operator norm, our
optimal procedure is based on a novel local cropping estimator targeting on all
principle submatrices of the precision matrix while for the Frobenius norm, our
optimal procedure relies on a delicate regressionbased blockthresholding
rule. We further establish rate optimality in the nonparanormal model.
Numerical studies are carried out to confirm our theoretical findings.

Since convolutional neural network (CNN) lacks an inherent mechanism to
handle large scale variations, we always need to compute feature maps multiple
times for multiscale object detection, which has the bottleneck of
computational cost in practice. To address this, we devise a recurrent scale
approximation (RSA) to compute feature map once only, and only through this map
can we approximate the rest maps on other levels. At the core of RSA is the
recursive rolling out mechanism: given an initial map at a particular scale, it
generates the prediction at a smaller scale that is half the size of input. To
further increase efficiency and accuracy, we (a): design a scaleforecast
network to globally predict potential scales in the image since there is no
need to compute maps on all levels of the pyramid. (b): propose a landmark
retracing network (LRN) to trace back locations of the regressed landmarks and
generate a confidence score for each landmark; LRN can effectively alleviate
false positives caused by the accumulated error in RSA. The whole system can be
trained endtoend in a unified CNN framework. Experiments demonstrate that our
proposed algorithm is superior against stateoftheart methods on face
detection benchmarks and achieves comparable results for generic proposal
generation. The source code of RSA is available at
github.com/sciencefans/RSAforobjectdetection.

Explaining the origin of life requires us to explain how selfreplication
arises. To be specific, how can a selfreplicating entity develop spontaneously
from a chemical reaction system in which no reaction is selfreplicating?
Previously proposed mathematical models either supply an explicit framework for
a minimal living system or only consider catalyzed reactions, and thus fail to
provide a comprehensive theory. We set up a general model for chemical reaction
systems that properly accounts for energetics, kinetics and the conservation
law. We find that (1) some systems are collectivelycatalytic where reactants
are transformed into end products with the assistance of intermediates (as in
the citric acid cycle), while some others are selfreplicating where different
parts replicate each other and the system selfreplicates as a whole (as in the
formose reaction); (2) many alternative chemical universes often contain one or
more such systems; (3) it is possible to construct a selfreplicating system
where the entropy of some parts spontaneously decreases, in a manner similar to
that discussed by Schr\"odinger; (4) complex selfreplicating molecules can
emerge spontaneously and relatively easily from simple chemical reaction
systems through a sequence of transitions. Together these results start to
explain the origins of prebiotic evolution.

This paper is a followup work about the artificial ecosystem model: number
soup (Liu and Sumpter, J. Royal Soc. Interface, 2017). It elaborates more
details about this model and points out future directions.