-
AutoML for Large Capacity Modeling of Meta's Ranking Systems
Authors:
Hang Yin,
Kuang-Hung Liu,
Mengying Sun,
Yuxin Chen,
Buyun Zhang,
Jiang Liu,
Vivek Sehgal,
Rudresh Rajnikant Panchal,
Eugen Hotaj,
Xi Liu,
Daifeng Guo,
Jamey Zhang,
Zhou Wang,
Shali Jiang,
Huayu Li,
Zhengxing Chen,
Wen-Yen Chen,
Jiyan Yang,
Wei Wen
Abstract:
Web-scale ranking systems at Meta serving billions of users is complex. Improving ranking models is essential but engineering heavy. Automated Machine Learning (AutoML) can release engineers from labor intensive work of tuning ranking models; however, it is unknown if AutoML is efficient enough to meet tight production timeline in real-world and, at the same time, bring additional improvements to…
▽ More
Web-scale ranking systems at Meta serving billions of users is complex. Improving ranking models is essential but engineering heavy. Automated Machine Learning (AutoML) can release engineers from labor intensive work of tuning ranking models; however, it is unknown if AutoML is efficient enough to meet tight production timeline in real-world and, at the same time, bring additional improvements to the strong baselines. Moreover, to achieve higher ranking performance, there is an ever-increasing demand to scale up ranking models to even larger capacity, which imposes more challenges on the efficiency. The large scale of models and tight production schedule requires AutoML to outperform human baselines by only using a small number of model evaluation trials (around 100). We presents a sampling-based AutoML method, focusing on neural architecture search and hyperparameter optimization, addressing these challenges in Meta-scale production when building large capacity models. Our approach efficiently handles large-scale data demands. It leverages a lightweight predictor-based searcher and reinforcement learning to explore vast search spaces, significantly reducing the number of model evaluations. Through experiments in large capacity modeling for CTR and CVR applications, we show that our method achieves outstanding Return on Investment (ROI) versus human tuned baselines, with up to 0.09% Normalized Entropy (NE) loss reduction or $25\%$ Query per Second (QPS) increase by only sampling one hundred models on average from a curated search space. The proposed AutoML method has already made real-world impact where a discovered Instagram CTR model with up to -0.36% NE gain (over existing production baseline) was selected for large-scale online A/B test and show statistically significant gain. These production results proved AutoML efficacy and accelerated its adoption in ranking systems at Meta.
△ Less
Submitted 16 November, 2023; v1 submitted 13 November, 2023;
originally announced November 2023.
-
Electromagnetic Multipoles for Morris-Thorne Wormhole
Authors:
A. H. Hasmani,
Sagar V. Soni,
Ravi Panchal
Abstract:
Wormholes are interesting space-time structures connecting two asymptotic regions found in a universe or multiverse and are solutions to Einstein's field equations. These objects have many interesting features as far as physics is concerned. Morris and Thorne introduced traversable wormholes, which increases the possibility of space-time travel. In this work, the wave equation of the Morris-Thorne…
▽ More
Wormholes are interesting space-time structures connecting two asymptotic regions found in a universe or multiverse and are solutions to Einstein's field equations. These objects have many interesting features as far as physics is concerned. Morris and Thorne introduced traversable wormholes, which increases the possibility of space-time travel. In this work, the wave equation of the Morris-Thorne wormhole has been derived by the technique of differential forms. The solution of the wave equation for a particular choice of red-shift function and shape function is obtained. The potential has also been computed in order to analyze electromagnetic fields. The behavior of electromagnetic multipoles is expressed and investigated in their behavior at the wormhole's throat.
△ Less
Submitted 25 August, 2023;
originally announced August 2023.
-
Rubio de Francia Extrapolation Theorems for Quasi-Monotone Functions
Authors:
Arun Pal Singh,
Ragul Panchal,
Pankaj Jain,
Monika Singh
Abstract:
We prove Rubio de Francia extrapolation results in Lebesgue and grand Lebesgue spaces for quasi monotone functions with $QB_{β,p}$ weights. The extrapolation in Lebesgue spaces with the weight class $QB_{β,\infty}$ has also been investigated. As an application, we characterize the boundedness of the Hardy averaging operator for quasi monotone functions in the grand Lebesgue spaces.
We prove Rubio de Francia extrapolation results in Lebesgue and grand Lebesgue spaces for quasi monotone functions with $QB_{β,p}$ weights. The extrapolation in Lebesgue spaces with the weight class $QB_{β,\infty}$ has also been investigated. As an application, we characterize the boundedness of the Hardy averaging operator for quasi monotone functions in the grand Lebesgue spaces.
△ Less
Submitted 5 February, 2022;
originally announced February 2022.
-
How do galaxy properties affect void statistics?
Authors:
Rushy R. Panchal,
Alice Pisani,
David N. Spergel
Abstract:
Using a map** from dark matter halos to galaxy properties based on hydrodynamical simulations, we explore the impact of galaxy properties on the void size function and the void-galaxy correlation function. We replicate the properties of galaxies from Illustris on MassiveNus halos, to perform both luminosity and star formation rate cuts on MassiveNus halos. We compare the impact of such cuts on v…
▽ More
Using a map** from dark matter halos to galaxy properties based on hydrodynamical simulations, we explore the impact of galaxy properties on the void size function and the void-galaxy correlation function. We replicate the properties of galaxies from Illustris on MassiveNus halos, to perform both luminosity and star formation rate cuts on MassiveNus halos. We compare the impact of such cuts on voids properties with respect to cuts on halo mass (as usually performed on halo catalogs driven from N-body simulations). We find that void catalogs built from luminosity-selected galaxies and halos are consistent within errors, while void catalogs built from star formation rate selected galaxies differ from void catalogs built on halos. We investigate the reason for this difference. Our work suggests that voids built on galaxy catalogs (selected through luminosity cut) can be reliably studied by using halos in dark matter simulations.
△ Less
Submitted 30 September, 2020;
originally announced September 2020.
-
Prediction of helium vapor quality in steady state Two-phase operation for SST-1 Toroidal field magnets
Authors:
G. K. Singh,
R. Panchal,
V. L. Tanna,
S. Pradhan
Abstract:
Steady State Superconducting Tokamak (SST-1) at the Institute for Plasma Research (IPR) is an operational device and is the first superconducting Tokamak in India. Superconducting Magnets System (SCMS) in SST-1 comprises of sixteen Toroidal field (TF) magnets and nine Poloidal Field (PF) magnets manufactured using NbTi/Cu based cable-in-conduit-conductor (CICC) concept. SST-1, superconducting TF m…
▽ More
Steady State Superconducting Tokamak (SST-1) at the Institute for Plasma Research (IPR) is an operational device and is the first superconducting Tokamak in India. Superconducting Magnets System (SCMS) in SST-1 comprises of sixteen Toroidal field (TF) magnets and nine Poloidal Field (PF) magnets manufactured using NbTi/Cu based cable-in-conduit-conductor (CICC) concept. SST-1, superconducting TF magnets are operated in a Cryo-stable manner being cooled with two-phase (TP) flow helium. The typical operating pressure of the TP helium is 1.6 bar (a) at corresponding saturation temperature. The SCMS has a typical cool-down time of about 14 days from 300 K down to 4.5 K using Helium plant of equivalent cooling capacity of 1350 W at 4.5 K. Using the onset of experimental data from the HRL, we estimated the vapor quality for the input heat load on to the TF magnets system. In this paper, we report the characteristics of two-phase flow for given thermo-hydraulic conditions during long steady state operation of the SST-1 TF magnets. Finally, the experimentally obtained results have been compared with the well-known correlations of two-phase flow.
△ Less
Submitted 9 February, 2018; v1 submitted 11 April, 2017;
originally announced April 2017.
