Statistical learning theory provides bounds of the generalization gap, using
in particular the Vapnik-Chervonenkis dimension and the Rademacher complexity.
An alternative approach, mainly studied in the statistical physics literature,
is the study of generalization in simple synthetic-data models. Here we discuss
the connections between these approaches and focus on the link between the
Rademacher complexity in statistical learning and the theories of
generalization for typical-case synthetic models from statistical physics,
involving quantities known as Gardner capacity and ground state energy. We show
that in these models the Rademacher complexity is closely related to the ground
state energy computed by replica theories. Using this connection, one may
reinterpret many results of the literature as rigorous Rademacher bounds in a
variety of models in the high-dimensional statistics limit. Somewhat
surprisingly, we also show that statistical learning theory provides
predictions for the behavior of the ground-state energies in some...

more |
pdf
| html
None.

zdeborova:
Did you know that the Rademacher complexity is just the ground state energy of the perceptron model? If you want to know more check out our new paper: https://t.co/oWljx5urQH with @KrzakalaF

KrzakalaF:
What is the link between the theoretical work on neural nets in physics in the 90s, Statistical learning theories in the 80s and over-parametrization the 2010s? Find out in our new paper "Rademacher complexity and Spin Glasses" https://t.co/IMioEu9IOJ

arxivml:
"Rademacher complexity and spin glasses: A link between the replica and statistical theories of learning",
Alia Abb…
https://t.co/v2EF14LbzY

arxiv_cs_LG:
Rademacher complexity and spin glasses: A link between the replica and statistical theories of learning. Alia Abbara, Benjamin Aubin, Florent Krzakala, and Lenka Zdeborová https://t.co/0Ar16Elwem

TakaAomidoro:
https://t.co/pAmeMrOEYu
"Rademacher complexity and spin glasses: A link between the replica and statistical theories of learning"
Alia Abbara, Benjamin Aubin, Florent Krzakala, Lenka Zdeborová

StatsPapers:
Rademacher complexity and spin glasses: A link between the replica and statistical theories of learning. https://t.co/K4ZKAfnK4o

None.

None.

Sample Sizes : None.

Authors: 4

Total Words: 11900

Unqiue Words: 2741

This paper addresses a difficult inverse problem that involves the
reconstruction of a three-dimensional model of tetrahedral amorphous
semiconductors via inversion of diffraction data. By posing the
material-structure determination as a multi-objective optimization program, it
has been shown that the problem can be solved accurately using a few structural
constraints, but no total-energy functionals/forces, which describe the local
chemistry of amorphous networks. The approach yields highly realistic models of
amorphous silicon, with no or only a few coordination defects ($\le$ 1%), a
narrow bond-angle distribution of width 9-11.5 degree, and an electronic gap of
0.8-1.4 eV. These data-driven information-based models have been found to
produce electronic and vibrational properties of amorphous silicon that match
accurately with experimental data and rival that of the Wooten-Winer-Weaire
(W3) models. The study confirms the effectiveness of a multi-objective
optimization approach to the structural determination of complex...

more |
pdf
| html
None.

CondensedPapers:
Disorder by design: A data-driven approach to amorphous semiconductors without total-energy functionals. https://t.co/LLV4xyVjlx

mat_mol_ml:
Disorder by design: A data-driven approach to amorphous semiconductors without total-energy functionals. https://t.co/TPOYigN8eN

None.

None.

Sample Sizes : None.

Authors: 4

Total Words: 0

Unqiue Words: 0

An extensive equilibrium Monte Carlo simulation is performed on the 3D
isotropic Heisenberg SG model with the random nearest-neighbor Gaussian
coupling, with particular interest in its chiral-glass (CG) and spin-glass (SG)
orderings. For this model, the possibility of the spin-chirality decoupling,
{\it i.e.\}, the CG order setting in at a higher temperature than that of the
SG order was suggested earlier, but still remains controversial. We simulate
the model up to the maximum size (linear dimension) $L=48$ under both periodic
and open boundary conditions (BC). In locating the CG and SG transition
temperatures $T_{{\rm CG}}$ and $T_{{\rm SG}}$ by the $L\rightarrow \infty$
extrapolation, a variety of independent physical quantities under the both BC
are computed and utilized to get larger number of degrees of freedom (NDF).
Thanks to the large NDF up to NDF=43, we succeed in obtaining stable and
accurate estimates of the CG and SG transition temperatures, $T_{{\rm
CG}}=0.142\pm 0.001$ and $T_{{\rm SG}}=0.131^{+0.001}_{-0.006}$. No...

more |
pdf
| html
None.

CondensedPapers:
Monte Carlo studies of the spin-chirality decoupling in the three-dimensional Heisenberg spin glass. https://t.co/mrNNdmKe4V

None.

None.

Sample Sizes : None.

Authors: 3

Total Words: 0

Unqiue Words: 0

A spin-1/2 Ising model, defined in the body centered cubic lattice, is used
to describe some of the thermodynamic properties of Fe$_p$-Al$_q$ alloys, with
$p+q=1$. The model assumes, besides the nearest-neighbor exchange coupling, the
existence of further next-nearest-neighbor superexchange interactions, where
the latter ones depend on the aluminum atoms cluster size. The Ising system so
considered is studied by employing Monte Carlo simulations, using a hybrid
algorithm consisting of one single-spin Metropolis move together with one
single-cluster Wolff algorithm allied, in addition, with single histograms
procedures and finite-size scaling techniques. Quite good fits to the
experimental results of the ordering critical temperature, as a function of Al
concentration in the range $0\le q<0.7$, are obtained and compared to more
recent theoretical approaches done on the same alloys.

more |
pdf
| html
None.

CondensedPapers:
Cluster size effects in the magnetic properties of Fe$_p$-Al$_{q=1-p}$ alloys. https://t.co/6LSKV5cG1y

None.

None.

Sample Sizes : None.

Authors: 4

Total Words: 0

Unqiue Words: 0

We analyze predator-prey dynamics in one dimension in which a Brownian
predator adopts a chasing strategy that consists in stochastically resetting
its current position to locations previously visited by a diffusive prey. We
study three different chasing strategies, namely, active, uniform and passive
which lead to different diffusive behaviors of the predator in the absence of
capture. When capture is considered, regardless of the chasing strategy, the
mean first-encounter time is finite and decreases with the resetting rate. This
model illustrates how the use of cues significantly improves the efficiency of
random searches. We compare numerical simulations with analytical calculations
and find excellent agreement.

more |
pdf
| html
None.

CondensedPapers:
Predator-prey dynamics: Chasing by stochastic resetting. https://t.co/MxOAlyruFD

None.

None.

Sample Sizes : None.

Authors: 3

Total Words: 0

Unqiue Words: 0

The continuous development of synchrotron-based experimental techniques in
the X-ray range provides new possibilities to probe the structure and the
dynamics of bulk materials down to inter-atomic distances. However, the
interaction of intense X-ray beams with matter can also induce changes in the
structure and dynamics of materials. A reversible and non-destructive beam
induced dynamics has recently been observed in X-ray photon correlation
spectroscopy experiments in some oxide glasses at sufficiently low absorbed
doses, and is here investigated in a (Li$_2$O)$_{0.5}$(B$_2$O$_3$)$_{0.5}$
glass. The characteristic time of this induced dynamics is inversely
proportional to the intensity of the X-ray beam, with a coefficient that
depends on the chemical composition and local structure of the probed glass,
making it a potentially new tool to investigate fundamental properties of a
large class of disordered systems. While the exact mechanisms behind this
phenomenon are yet to be elucidated, we report here on the measurement of...

more |
pdf
| html
None.

CondensedPapers:
X-rays induced atomic dynamics in a lithium-borate glass. https://t.co/JduOYWNCD4

None.

None.

Sample Sizes : None.

Authors: 7

Total Words: 0

Unqiue Words: 0

Assert is a website where the best academic papers on arXiv (computer science, math, physics), bioRxiv (biology), BITSS (reproducibility), EarthArXiv (earth science), engrXiv (engineering), LawArXiv (law), PsyArXiv (psychology), SocArXiv (social science), and SportRxiv (sport research) bubble to the top each day.

Papers are scored (in real-time) based on how verifiable they are (as determined by their Github repos) and how interesting they are (based on Twitter).

To see top papers, follow us on twitter @assertpub_ (arXiv), @assert_pub (bioRxiv), and @assertpub_dev (everything else).

To see beautiful figures extracted from papers, follow us on Instagram.

*Tracking 234,443 papers.*

Sort results based on if they are interesting or reproducible.

Interesting

Reproducible