-
What are Neural Networks made of?
Authors:
Rene Schaub
Abstract:
The success of Deep Learning methods is not well understood, though various attempts at explaining it have been made, typically centered on properties of stochastic gradient descent. Even less clear is why certain neural network architectures perform better than others. We provide a potential opening with the hypothesis that neural network training is a form of Genetic Programming.
The success of Deep Learning methods is not well understood, though various attempts at explaining it have been made, typically centered on properties of stochastic gradient descent. Even less clear is why certain neural network architectures perform better than others. We provide a potential opening with the hypothesis that neural network training is a form of Genetic Programming.
△ Less
Submitted 25 August, 2019;
originally announced September 2019.
-
Adsorption energies of benzene on close packed transition metal surfaces using the random phase approximation
Authors:
J. A. Garrido Torres,
B. Ramberger,
H. Früchtl,
R. Schaub,
G. Kresse
Abstract:
The adsorption energy of benzene on various metal substrates is predicted using the random phase approximation (RPA) for the correlation energy. Agreement with available experimental data is systematically better than 10% for both coinage and reactive metals. The results are also compared with more approximate methods, including vdW-density functional theory (DFT), as well as dispersion corrected…
▽ More
The adsorption energy of benzene on various metal substrates is predicted using the random phase approximation (RPA) for the correlation energy. Agreement with available experimental data is systematically better than 10% for both coinage and reactive metals. The results are also compared with more approximate methods, including vdW-density functional theory (DFT), as well as dispersion corrected DFT functionals. Although dispersion corrected DFT can yield accurate results, for instance, on coinage metals, the adsorption energies are clearly overestimated on more reactive transition metals. Furthermore, coverage dependent adsorption energies are well described by the RPA. This shows that for the description of aromatic molecules on metal surfaces further improvements in density functionals are necessary, or more involved many body methods such as the RPA are required.
△ Less
Submitted 30 August, 2017;
originally announced August 2017.
-
Channel saturation and conductance quantization in single-atom gold constrictions
Authors:
Jason N. Armstrong,
R. M. Schaub,
Susan Z. Hua,
Harsh Deep Chopra
Abstract:
Notwithstanding the discreteness of metallic constrictions, it is shown that the finite elasticity of stable, single-atom gold constrictions allows for a continuous and reversible change in conductance, thereby enabling observation of channel saturation and conductance quantization. The observed channel saturation and signature for conductance quantization is achieved by superposition of atomic/su…
▽ More
Notwithstanding the discreteness of metallic constrictions, it is shown that the finite elasticity of stable, single-atom gold constrictions allows for a continuous and reversible change in conductance, thereby enabling observation of channel saturation and conductance quantization. The observed channel saturation and signature for conductance quantization is achieved by superposition of atomic/subatomic-scale oscillations on a retracting/approaching gold tip against a gold substrate of a scanning probe. Results also show that conductance histograms are neither suitable for evaluating the stability of atomic configurations through peak positions or peak height nor appropriate for assessing conductance quantization. A large number of atomic configurations with similar conductance values give rise to peaks in the conductance histogram. The positions of the peaks and counts at each peak can be varied by changing the conditions under which the histograms are made. Histogram counts below 1Go cannot necessarily be assumed to arise from single-atom constrictions.
△ Less
Submitted 4 August, 2010;
originally announced August 2010.
-
Strong electron correlations in the normal state of FeSe0.42Te0.58
Authors:
A. Tamai,
A. Y. Ganin,
E. Rozbicki,
J. Bacsa,
W. Meevasana,
P. D. C. King,
M. Caffio,
R. Schaub,
S. Margadonna,
K. Prassides,
M. J. Rosseinsky,
F. Baumberger
Abstract:
We investigate the normal state of the '11' iron-based superconductor FeSe0.42Te0.58 by angle resolved photoemission. Our data reveal a highly renormalized quasiparticle dispersion characteristic of a strongly correlated metal. We find sheet dependent effective carrier masses between ~ 3 - 16 m_e corresponding to a mass enhancement over band structure values of m*/m_band ~ 6 - 20. This is nearly…
▽ More
We investigate the normal state of the '11' iron-based superconductor FeSe0.42Te0.58 by angle resolved photoemission. Our data reveal a highly renormalized quasiparticle dispersion characteristic of a strongly correlated metal. We find sheet dependent effective carrier masses between ~ 3 - 16 m_e corresponding to a mass enhancement over band structure values of m*/m_band ~ 6 - 20. This is nearly an order of magnitude higher than the renormalization reported previously for iron-arsenide superconductors of the '1111' and '122' families but fully consistent with the bulk specific heat.
△ Less
Submitted 10 February, 2010; v1 submitted 16 December, 2009;
originally announced December 2009.
