arXiv subjects

Andrew Ross

Publications and source records attributed to Andrew Ross.

24 records · Page 2Linked to original sources

arXiv2019

Learning Key-Value Store Design

We introduce the concept of design continuums for the data layout of key-value stores. A design continuum unifies major distinct data structure designs under the same model. The critical insight and potential long-term impact is that such unifying models 1) render what we consider up to now as fundamentally different data structures to be seen as views of the very same overall design space, and 2) allow seeing new data structure designs with performance properties that are not feasible by existing designs. The core intuition behind the construction of design continuums is that all data structures arise from the very same set of fundamental design principles, i.e., a small set of data layout design concepts out of which we can synthesize any design that exists in the literature as well as new ones. We show how to construct, evaluate, and expand, design continuums and we also present the first continuum that unifies major data structure designs, i.e., B+tree, B-epsilon-tree, LSM-tree, and LSH-table. The practical benefit of a design continuum is that it creates a fast inference engine for the design of data structures. For example, we can predict near instantly how a specific design change in the underlying storage of a data system would affect performance, or reversely what would be the optimal data structure (from a given set of designs) given workload characteristics and a memory budget. In turn, these properties allow us to envision a new class of self-designing key-value stores with a substantially improved ability to adapt to workload and hardware changes by transitioning between drastically different data structure designs to assume a diverse set of performance properties at will.

cs.DB↗

arXiv2019

Improving Sepsis Treatment Strategies by Combining Deep and Kernel-Based Reinforcement Learning

Sepsis is the leading cause of mortality in the ICU. It is challenging to manage because individual patients respond differently to treatment. Thus, tailoring treatment to the individual patient is essential for the best outcomes. In this paper, we take steps toward this goal by applying a mixture-of-experts framework to personalize sepsis treatment. The mixture model selectively alternates between neighbor-based (kernel) and deep reinforcement learning (DRL) experts depending on patient's current history. On a large retrospective cohort, this mixture-based approach outperforms physician, kernel only, and DRL-only experts.

cs.LG↗

arXiv2019

Spin-dependent (inverse) spin Hall effect in Co$_{60}$Fe$_{20}$B$_{20}$

In ferromagnetic metals, the interconversion of spin and charge currents via the spin Hall effect and its inverse can depend on the angle between the ferromagnets magnetization and the spin current polarization direction. Here, such a spin-dependent (inverse) spin Hall effect is found in the ferromagnetic alloy Co$_{60}$Fe$_{20}$B$_{20}$. In a nonlocal magnon transport experiment, Co$_{60}$Fe$_{20}$B$_{20}$ is used to both excite and detect magnonic spin currents flowing in the ferrimagnetic insulator Y$_{3}$Fe$_{5}$O$_{12}$. We find that the signal amplitude is significantly modulated by tuning the direction of the Co$_{60}$Fe$_{20}$B$_{20}$ magnetization. We design a sample structure that completely prevents direct magnonic coupling between the ferromagnets. Thus, we can identify unambiguously an intrinsic electronic origin of the observed effect.

cond-mat.mtrl-sci↗

arXiv2019

Anisotropies and magnetic phase transitions in insulating antiferromagnets determined by a Spin-Hall magnetoresistance probe

We demonstrate that we can determine the antiferromagnetic anisotropies and the bulk Dzyaloshinskii-Moriya fields of the insulating iron oxide hematite, α-Fe2O3, using a surface sensitive spin-Hall magnetoresistance (SMR) technique. We develop an analytical model that in combination with SMR measurements, allow for the identification of the material parameters of this prototypical antiferromagnet over a wide range of temperatures and magnetic field values. Using devices with different orientations, we demonstrate that the SMR response strongly depends on the direction of the charge current with respect to the magneto-crystalline anisotropies axis. We show that we can extract the anisotropies over a wide temperature range including across the Morin phase transition. We observe that the electrical response is dominated by the orientation of the antiferromagnetic Néel order parameter, rather than by the emergent weak magnetic moment. Our results highlight that the surface sensitivity of the SMR allows accessing the magnetic anisotropies of antiferromagnetic crystals and in particular thin films where other methods to determine anisotropies such as bulk-sensitive magnetic susceptibility measurements do not provide sufficient sensitivity.

cond-mat.mes-hall↗

arXiv2018

3D sub-nanoscale imaging of unit cell doubling due to octahedral tilting and cation modulation in strained perovskite thin films

Determining the 3-dimensional crystallography of a material with sub-nanometre resolution is essential to understanding strain effects in epitaxial thin films. A new scanning transmission electron microscopy imaging technique is demonstrated that visualises the presence and strength of atomic movements leading to a period doubling of the unit cell along the beam direction, using the intensity in an extra Laue zone ring in the back focal plane recorded using a pixelated detector method. This method is used together with conventional atomic resolution imaging in the plane perpendicular to the beam direction to gain information about the 3D crystal structure in an epitaxial thin film of LaFeO3 sandwiched between a substrate of (111) SrTiO3 and a top layer of La0.7Sr0.3MnO3. It is found that a hitherto unreported structure of LaFeO3 is formed under the unusual combination of compressive strain and (111) growth, which is triclinic with a periodicity doubling from primitive perovskite along one of the three <110> directions lying in the growth plane. This results from a combination of La-site modulation along the beam direction, and modulation of oxygen positions resulting from octahedral tilting. This transition to the period-doubled cell is suppressed near both the substrate and near the La0.7Sr0.3MnO3 top layer due to the clamping of the octahedral tilting by the absence of tilting in the substrate and due to an incompatible tilt pattern being present in the La0.7Sr0.3MnO3 layer. This work shows a rapid and easy way of scanning for such transitions in thin films or other systems where disorder-order transitions or domain structures may be present and does not require the use of atomic resolution imaging, and could be done on any scanning TEM instrument equipped with a suitable camera.

cond-mat.mtrl-sci↗

arXiv2018

Spin transport in multilayer systems with fully epitaxial NiO thin films

We report on the generation and transport of thermal spin currents in fully epitaxial γ-Fe$_2$O$_3$/NiO(001)/Pt and Fe$_3$O$_4$/NiO(001)/Pt trilayers. A thermal gradient, perpendicular to the plane of the sample, generates a magnonic spin current in the ferrimagnetic maghemite (γ-Fe$_2$O$_3$) and magnetite (Fe$_3$O$_4$) thin films by means of the spin Seebeck effect. The spin current propagates across the epitaxial, antiferromagnetic insulating NiO layer, before being detected in the Pt layer by the inverse spin Hall effect. The transport of the spin signal is studied as a function of the NiO thickness, temperature and ferrimagnetic material where the spin current is generated. In epitaxial NiO grown on maghemite, the spin Seebeck signal decays exponentially as a function of the NiO thickness, with a spin-diffusion length for thermally-generated magnons of λ$_{MSDL}$ = $1.6 \pm 0.2$ nm, largely independent on temperature. We see no enhancement of the spin current signal as previously reported for certain temperatures and thicknesses of the NiO. In epitaxial NiO grown on magnetite, the temperature-averaged spin diffusion length is λ$_{MSDL}$ = $3.8 \pm 0.3$ nm, and we observe an enhancement of the spin signal when the NiO thickness is 0.8 nm, demonstrating that the growth conditions dramatically affect the spin transport properties of the NiO even for full epitaxial growth. In contrast to theoretical predictions for coherent spin transport, we do not see vastly different spin diffusion lengths between epitaxial and polycrystalline NiO layers.

cond-mat.mtrl-sci↗