• We propose a new method for learning the structure of convolutional neural networks (CNNs) that is more efficient than recent state-of-the-art methods based on reinforcement learning and evolutionary algorithms. Our approach uses a sequential model-based optimization (SMBO) strategy, in which we search for structures in order of increasing complexity, while simultaneously learning a surrogate model to guide the search through structure space. Direct comparison under the same search space shows that our method is up to 5 times more efficient than the RL method of Zoph et al. (2018) in terms of number of models evaluated, and 8 times faster in terms of total compute. The structures we discover in this way achieve state of the art classification accuracies on CIFAR-10 and ImageNet.
  • A large and high-quality single crystal (Li0.84Fe0.16)OHFe0.98Se, the optimal superconductor of newly reported (Li1-xFex)OHFe1-ySe system, has been successfully synthesized via a hydrothermal ion-exchange technique. The superconducting transition temperature (Tc) of 42 K is determined by magnetic susceptibility and electric resistivity measurements, and the zero-temperature upper critical magnetic fields are evaluated as 79 and 313 Tesla for the field along the c-axis and the ab-plane, respectively. The ratio of out-of-plane to in-plane electric resistivity,\r{ho}c/\r{ho}ab, is found to increases with decreasing temperature and to reach a high value of 2500 at 50 K, with an evident kink occurring at a characteristic temperature T*=120 K. The negative in-plane Hall coefficient indicates that electron carriers dominate in the charge transport, and the hole contribution is significantly reduced as the temperature is lowered to approach T*. From T* down to Tc, we observe the linear temperature dependences of the in-plane electric resistivity and the magnetic susceptibility for the FeSe layers. Our findings thus reveal that the normal state of (Li0.84Fe0.16)OHFe0.98Se becomes highly two-dimensional and anomalous prior to the superconducting transition, providing a new insight into the mechanism of high-Tc superconductivity.
  • In the history of condensed matter physics, reinvestigation of a well-studied material with enhanced quality sometimes led to important scientific discoveries. A well-known example is the discovery of fractional quantum Hall effect in high quality GaAs/AlGaAs heterojunctions. Here we report the first single crystal growth and magnetoresistance (MR) measurements of the silver chalcogenide $\beta $-Ag$_2$Se (Naumannite), a compound has been known for the unusual, linear-field-dependent MR in its polycrystalline form for over a decade. With the quantum limit (QL) as low as 3 Tesla, a moderate field produced by a superconductor magnet available in many laboratories can easily drive the electrons in Ag$_2$Se to an unprecedented state. We observed significant negative longitudinal MR beyond the QL, which was understood as a `charge-pumping' effect between the novel fermions with opposite chiralities. Characterization of the single-crystalline Ag$_2$Se and the fabrication of electric devices working above the QL, will represent a new direction for the study of these exotic electrons.
  • We investigate the resonant production of the lighter sbottom at the proposed Large Hadron Electron Collider and its subsequent decay into $e^- + jet$ final state, under the single coupling dominance hypothesis in the framework of the $R$-parity-violating minimal supersymmetric standard model. It is concluded that the LHeC running at the high electron beam energy configuration of $E_e = 150 {\rm GeV}$ would provide excellent opportunities to search for the sbottom resonance and probe the lepton-number-violating $\hat{L}\hat{Q}\hat{D}$ $\lambda^{\prime}_{113}$-type Yukawa interaction. With about $1 fb^{-1}$ integrated luminosity expected in $E_e = 150 {\rm GeV}$ $e^-p$ collision at the LHeC, either the lighter sbottom RPV resonance could be directly detected up to $1 {\rm TeV}$, or the $L$-violating coupling $\lambda^{\prime}_{113}$ could be constrained at an unprecedented level of precision compared with all the knowledge derived from indirect measurements.