Searcharxiv⌕ Search

arXiv subjects

Stavros Theodorakis

Publications and source records attributed to Stavros Theodorakis.

10 recordsLinked to original sources
arXiv2022

An Incremental Learning framework for Large-scale CTR Prediction

In this work we introduce an incremental learning framework for Click-Through-Rate (CTR) prediction and demonstrate its effectiveness for Taboola's massive-scale recommendation service. Our approach enables rapid capture of emerging trends through warm-starting from previously deployed models and fine tuning on "fresh" data only. Past knowledge is maintained via a teacher-student paradigm, where the teacher acts as a distillation technique, mitigating the catastrophic forgetting phenomenon. Our incremental learning framework enables significantly faster training and deployment cycles (x12 speedup). We demonstrate a consistent Revenue Per Mille (RPM) lift over multiple traffic segments and a significant CTR increase on newly introduced items.

cs.IR↗
arXiv2020

A variational model for the delayed collapse of Bose Einstein condensates

We present an action that can be used to study variationally the collapse of Bose Einstein condensates. This action is real, even though it includes dissipative terms. It adopts long range interactions between the atoms, so that there is always a stable minimum of the energy, even if the remaining number of atoms is above the number that in the case of local interactions is the critical one. The proposed action incorporates the time needed for the abrupt and delayed onset of collapse, yielding in fact its dependence on the scattering length. We show that the evolution of the condensate is equivalent to the motion of a particle in an effective potential. The particle begins its motion far from the point of stable equilibrium and it then proceeds to oscillate about that point. We prove that the resulting large oscillations in the shape of the wavefunction after the collapse have frequencies equal to twice the frequencies of the traps. Our results agree with the experimental observations.

cond-mat.quant-gas↗
arXiv2019

Nonlocal interactions in collapsing condensates in a box

The collapse of attractive Bose-Einstein condensates in a box with tunable interatomic interactions was studied experimentally recently. Not only were remarkably stable remnant condensates observed, but furthermore they often seem to involve two stable plateaus. We suggest that these plateaus correspond in fact to two minima of the energy, the attractive atomic interactions being nonlocal. We show in detail that all the experimental data for these remnant condensates can be accounted for by minimising a variational energy involving a nonlocal interatomic potential that is attractive everywhere and that has a range of the order of the magnetic length.

cond-mat.quant-gas↗
arXiv2018

Deep density networks and uncertainty in recommender systems

Building robust online content recommendation systems requires learning complex interactions between user preferences and content features. The field has evolved rapidly in recent years from traditional multi-arm bandit and collaborative filtering techniques, with new methods employing Deep Learning models to capture non-linearities. Despite progress, the dynamic nature of online recommendations still poses great challenges, such as finding the delicate balance between exploration and exploitation. In this paper we show how uncertainty estimations can be incorporated by employing them in an optimistic exploitation/exploration strategy for more efficient exploration of new recommendations. We provide a novel hybrid deep neural network model, Deep Density Networks (DDN), which integrates content-based deep learning models with a collaborative scheme that is able to robustly model and estimate uncertainty. Finally, we present online and offline results after incorporating DNN into a real world content recommendation system that serves billions of recommendations per day, and show the benefit of using DDN in practice.

cs.IR↗
arXiv2017

The Wavefunction of the Collapsing Bose-Einstein Condensate

Bose-Einstein condensates with tunable interatomic interactions have been studied intensely in recent experiments. The investigation of the collapse of a condensate following a sudden change in the nature of the interaction from repulsive to attractive has led to the observation of a remnant condensate that did not undergo further collapse. We suggest that this high-density remnant is in fact the absolute minimum of the energy, if the attractive atomic interactions are nonlocal, and is therefore inherently stable. We show that a variational trial function consisting of a superposition of two distinct gaussians is an accurate representation of the wavefunction of the ground state of the conventional local Gross-Pitaevskii field equation for an attractive condensate and gives correctly the points of emergence of instability. We then use such a superposition of two gaussians as a variational trial function in order to calculate the minima of the energy when it includes a nonlocal interaction term. We use experimental data in order to study the long range of the nonlocal interaction, showing that they agree very well with a dimensionally derived expression for this range.

cond-mat.quant-gas↗
arXiv2001

Speed Selection Mechanism for Propagating Fronts in Reaction-Diffusion Systems with Multiple Fields

We introduce a speed selection mechanism for front propagation in reaction-diffusion systems with multiple fields. This mechanism applies to pulled and pushed fronts alike, and operates by restricting the fields to large "finite" intervals in the comoving frames of reference. The unique velocity for which the center of a monotonic solution for a particular field is insensitive to the location of the ends of the finite interval is the velocity that is physically selected for that field, making thus the solution approximately translation invariant. The fronts for the various fields may propagate at different speeds, all of them being determined through this mechanism. We present analytic results for the case of piecewise parabolic potentials, and numerical results for other cases.

nlin.PS↗
arXiv2000

Emergence of Approximate Translation Invariance in Finite Intervals as a Speed Selection Mechanism for Propagating Fronts

We introduce a new velocity selection criterion for fronts propagating into unstable and metastable states. We restrict these fronts to large finite intervals in the comoving frame of reference and require their centers be insensitive to the locations of the ends of the finite intervals, exhibiting thus effectively an approximate translation invariance. Only one monotonic front has this behavior and its velocity is the one that is physically selected. We present analytic results in the case of piecewise parabolic potentials, and numerical results in other cases.

nlin.PS↗
arXiv1997

Magnetic Field Inversion in Vortices in Multilayers

We present a description of very dense vortex lattices in highly anisotropic multilayers, for high fields parallel to the layers. We show that a magnetic field inversion can occur away from the center of a vortex, provided the layers are sufficiently far apart.

cond-mat↗