arXiv subjects

Sergey Alekseev

Publications and source records attributed to Sergey Alekseev.

8 recordsLinked to original sources

arXiv2026

Subcritical Signal Propagation at Initialization in Normalization-Free Transformers

We study signal propagation at initialization in transformers through the averaged partial Jacobian norm (APJN), a measure of gradient amplification across layers. We extend APJN analysis to transformers with bidirectional attention and permutation-symmetric input token configurations by deriving recurrence relations for activation statistics and APJNs across layers. Our theory predicts how attention modifies the asymptotic behavior of the APJN at large depth and matches APJNs measured in deep vision transformers. The criticality picture known from residual networks carries over to transformers: the pre-LayerNorm architecture exhibits power-law APJN growth, whereas transformers with LayerNorm replaced by elementwise $\tanh$-like nonlinearities have stretched-exponential APJN growth, indicating that the latter are subcritical. Applied to Dynamic Tanh (DyT) and Dynamic erf (Derf) transformers, the theory explains why these architectures can be more sensitive to initialization and optimization choices and require careful tuning for stable training.

cs.LG↗

arXiv2025

Charge density wave induced gapped nodal line

We investigate the interplay between charge density wave (CDW) order and topological nodal-line states in square-net materials. Our Ginzburg-Landau theory predicts a CDW instability that generically opens a gap at the Fermi energy while preserving the nodal line crossing. However, as the Fermi level approaches the nodal line, the density of states at the nodal line decreases, eventually disappearing as the CDW vector $\mathbf{Q}$ goes to zero. Exactly at $\mathbf{Q} = 0$, the order parameter explicitly breaks the glide symmetry protecting the nodal line, which allows a gap to open. Yet, for small but finite $\mathbf{Q}$, the nodal line may vanish within experimental resolution even when the glide symmetry is preserved. Our results provide a consistent explanation for recent experimental observations.

cond-mat.str-el↗

arXiv2024

Uncovering the Hidden Ferroaxial Density Wave as the Origin of the Axial Higgs Mode in RTe$_3$

The recent discovery of an axial amplitude (Higgs) mode in the long-studied charge density wave (CDW) systems GdTe$_3$ and LaTe$_3$ suggests a heretofore unidentified hidden order. A theoretical study proposed that the axial Higgs results from a hidden ferroaxial component of the CDW, which could arise from non-trivial orbital texture. Here, we report extensive experimental studies on ErTe$_3$ and HoTe$_3$ that possess a high-temperature CDW similar to other RTe$_3$ (R = rare earth), along with an additional low-temperature CDW with an orthogonal ordering vector. Combining Raman spectroscopy with large-angle convergent beam electron diffraction (LACBED), rotational anisotropy second-harmonic generation (RA-SHG), and muon-spin relaxation ($μ$SR), we provide unambiguous evidence that the high-temperature CDW breaks translation, rotation, and all vertical and diagonal mirror symmetries, but not time-reversal or inversion. In contrast, the low-temperature CDW only additionally breaks translation symmetry. Simultaneously, Raman scattering shows the high-temperature CDW produces an axial Higgs mode while the low-temperature mode is scalar. The weak monoclinic structural distortion and clear axial response in Raman and SHG are consistent with a ferroaxial phase in RTe$_3$ driven by coupled orbital and charge orders. Thus, our study provides a new standard for uncovering unconventional orders and confirms the power of Higgs modes to reveal them.

cond-mat.str-el↗

arXiv2024

Charge density-waves with non-trivial orbital textures in rare earth tritellurides

Motivated by recent experiments reporting unconventional collective modes in the charge density-wave (CDW) state of rare-earth tritellurides $R$Te$_3$, we derive from a multi-orbital microscopic model on the square net a CDW Ginzburg-Landau theory that allows for non-trivial orbital order. Our analysis reveals unconventional CDWs where order parameters with distinct orbital character coexist due to an approximate symmetry of the low-energy model, which becomes exact in the limit of nearest-neighbor-only hopping and decoupled $p_x$, $p_y$ orbitals. Because of this coexistence, the resulting CDW pattern displays an orbital texture that generally breaks additional symmetries of the lattice besides those explicitly broken by the CDW wave-vector. In particular, we find two competing phases that differ in whether they break or preserve inversion and vertical mirror symmetries. We explain the mechanisms that favor each outcome, and discuss experimental probes that can distinguish the different phases.

cond-mat.str-el↗

arXiv2024

Chiral life on a slab

We study chiral algebra in the reduction of 3D $\mathcal{N} = 2 $ supersymmetric gauge theories on an interval with the $\mathcal{N}=(0,2)$ Dirichlet boundary conditions on both ends. By invoking the 3D ``twisted formalism'' and the 2D $βγ$-description we explicitly find the perturbative $\overline{Q}_+$ cohomology of the reduced theory. It is shown that the vertex algebras of boundary operators are enhanced by the line operators. A full non-perturbative result is found in the abelian case, where the chiral algebra is given by the rank two Narain lattice VOA, and two more equivalent descriptions are provided. Conjectures and speculations on the nonperturbative answer in the non-abelian case are also given.

hep-th↗

arXiv2020

Problems of representation of electrocardiograms in convolutional neural networks

Using electrocardiograms as an example, we demonstrate the characteristic problems that arise when modeling one-dimensional signals containing inaccurate repeating pattern by means of standard convolutional networks. We show that these problems are systemic in nature. They are due to how convolutional networks work with composite objects, parts of which are not fixed rigidly, but have significant mobility. We also demonstrate some counterintuitive effects related to generalization in deep networks.

cs.LG↗

arXiv2019

Toward the Pole

We study the analytic structure of semiclassical conformal blocks, namely of the 1-point conformal block on the torus and of the 4-point conformal block on the sphere, as functions of the intermediate dimension. We interpret their discontinuities, which can be revealed with the use of a particular resummation procedure, holographically in terms of configurations of geodesics in AdS3. In other words, we study the behavior of the $Ω$-deformed $\mathcal N=2$ SYM theory in the Nekrasov-Shatashvili limit near those singular points, where naively the W-boson with non-vanishing angular momentum becomes massless. Upon a proper resummation of instanton contributions, these singularities disappear, which is similar to the Seiberg-Witten solution in the undeformed case where there is no massless W-boson. It is shown that the order parameter undergoes a non-analytic behavior near positions of the poles. The jump of the order parameter is interpreted holographically in terms of geodesic networks

hep-th↗

arXiv2018

ECG Segmentation by Neural Networks: Errors and Correction

In this study we examined the question of how error correction occurs in an ensemble of deep convolutional networks, trained for an important applied problem: segmentation of Electrocardiograms(ECG). We also explore the possibility of using the information about ensemble errors to evaluate a quality of data representation, built by the network. This possibility arises from the effect of distillation of outliers, which was demonstarted for the ensemble, described in this paper.

cs.LG↗