-
Harmony in Diversity: Merging Neural Networks with Canonical Correlation Analysis
Authors:
Stefan Horoi,
Albert Manuel Orozco Camacho,
Eugene Belilovsky,
Guy Wolf
Abstract:
Combining the predictions of multiple trained models through ensembling is generally a good way to improve accuracy by leveraging the different learned features of the models, however it comes with high computational and storage costs. Model fusion, the act of merging multiple models into one by combining their parameters reduces these costs but doesn't work as well in practice. Indeed, neural net…
▽ More
Combining the predictions of multiple trained models through ensembling is generally a good way to improve accuracy by leveraging the different learned features of the models, however it comes with high computational and storage costs. Model fusion, the act of merging multiple models into one by combining their parameters reduces these costs but doesn't work as well in practice. Indeed, neural network loss landscapes are high-dimensional and non-convex and the minima found through learning are typically separated by high loss barriers. Numerous recent works have been focused on finding permutations matching one network features to the features of a second one, lowering the loss barrier on the linear path between them in parameter space. However, permutations are restrictive since they assume a one-to-one map** between the different models' neurons exists. We propose a new model merging algorithm, CCA Merge, which is based on Canonical Correlation Analysis and aims to maximize the correlations between linear combinations of the model features. We show that our alignment method leads to better performances than past methods when averaging models trained on the same, or differing data splits. We also extend this analysis into the harder setting where more than 2 models are merged, and we find that CCA Merge works significantly better than past methods. Our code is publicly available at https://github.com/shoroi/align-n-merge
△ Less
Submitted 7 July, 2024;
originally announced July 2024.
-
A fast tracking code for evaluating collective effects in linear accelerators
Authors:
F. Bosco,
O. Camacho,
M. Carillo,
E. Chiadroni,
L. Faillace,
A. Fukasawa,
A. Giribono,
L. Giuliano,
N. Najernik,
A. Mostacci,
L. Palumbo,
B. Spataro,
C. Vaccarezza,
J. B. Rosenzweig,
M. Migliorati
Abstract:
The demands on performance of advanced linear accelerator based facilities strongly depend on the quality of the particle beams produced by such machines. Indeed, state-of-the-art applications in photon production and high-energy physics colliders require to use very high brightness electron beams, implying the coexistence of high peak currents and small transverse emittances. In such systems, the…
▽ More
The demands on performance of advanced linear accelerator based facilities strongly depend on the quality of the particle beams produced by such machines. Indeed, state-of-the-art applications in photon production and high-energy physics colliders require to use very high brightness electron beams, implying the coexistence of high peak currents and small transverse emittances. In such systems, the nominal phase-space density may be diluted by the presence of self-induced electromagnetic fields, causing interaction among charged particles through space charge forces and the excitation of wakefields. The two sources of collective effects may both be present in significant levels, and be coupled by the strong externally applied transverse and longitudinal fields present in modern high gradient linear accelerators. Thus, beam dynamics studies investigating all relevant effects, applied and collective, are necessary to predict the operational limitations of a given instrument. Such modeling, involving a large number of computational particles, can require significant numerical resources. In this paper we present a fast tracking code which permits accurate evaluation of wakefield effects in rf linacs, while also including a simple, robust model for space-charge forces to streamline the computations. The features of such a tool are discussed in detail in this paper and comparisons with more time-intensive commonly used tracking codes or analytical models are utilized to validate the approach we introduce. In addition, the applications motivating the development of this code define unique and challenging scenarios from the perspective of beam physics. Specifically, the fast simulation framework developed in this paper aims to describe intense electron beams injected at low energy in high-gradient accelerating structures which introduce strong rf focusing as well as strong wakefield interactions.
△ Less
Submitted 12 August, 2022;
originally announced August 2022.
-
Versatile, high brightness, cryogenic photoinjector electron source
Authors:
River R. Robles,
Obed Camacho,
Atsushi Fukasawa,
Nathan Majernik,
James B. Rosenzweig
Abstract:
Since the introduction of the radio-frequency (rf) photoinjector electron source over thirty years ago, peak performance demands have dictated the use of high accelerating electric fields. With recent strong advances in obtainable field values, attendant increases in beam brightness are expected to be dramatic. In this article, we examine the implementation of very high gradient acceleration in a…
▽ More
Since the introduction of the radio-frequency (rf) photoinjector electron source over thirty years ago, peak performance demands have dictated the use of high accelerating electric fields. With recent strong advances in obtainable field values, attendant increases in beam brightness are expected to be dramatic. In this article, we examine the implementation of very high gradient acceleration in a high frequency, cryogenic rf photoinjector. We discuss in detail the effects of introducing, through an optimized rf cavity shape, rich spatial harmonic content in the accelerating modes in this device. Higher spatial harmonics give useful, enhanced linear focusing effects, as well as potentially deleterious nonlinear transverse forces. They also serve to strongly increase the ratio of average accelerating field to peak surface field, thus aiding in managing power and dark current-related challenges. We investigate two scenarios which are aimed at unique exploitation of the capabilities of this source. First, we investigate the obtaining of extremely high six-dimensional brightness for advanced free-electron laser applications. We also examine the use of a magnetized photocathode in the device for producing unprecedented low asymmetric emittance, high-current electron beams that reach linear collider-compatible performance. As both of the scenarios demand an advanced, compact solenoid design, we describe a novel cryogenic solenoid system. With the high field rf and magnetostatic structures introduced, we analyze the collective beam dynamics in these systems through theory and multi-particle simulations, including a particular emphasis on granularity effects associated with microscopic Coulomb interactions.
△ Less
Submitted 11 May, 2021; v1 submitted 15 March, 2021;
originally announced March 2021.
-
An Ultra-Compact X-Ray Free-Electron Laser
Authors:
J. B. Rosenzweig,
N. Majernik,
R. R. Robles,
G. Andonian,
O. Camacho,
A. Fukasawa,
A. Kogar,
G. Lawler,
Jianwei Miao,
P. Musumeci,
B. Naranjo,
Y. Sakai,
R. Candler,
B. Pound,
C. Pellegrini,
C. Emma,
A. Halavanau,
J. Hastings,
Z. Li,
M. Nasr,
S. Tantawi,
P. Anisimov,
B. Carlsten,
F. Krawczyk,
E. Simakov
, et al. (11 additional authors not shown)
Abstract:
In the field of beam physics, two frontier topics have taken center stage due to their potential to enable new approaches to discovery in a wide swath of science. These areas are: advanced, high gradient acceleration techniques, and x-ray free electron lasers (XFELs). Further, there is intense interest in the marriage of these two fields, with the goal of producing a very compact XFEL. In this con…
▽ More
In the field of beam physics, two frontier topics have taken center stage due to their potential to enable new approaches to discovery in a wide swath of science. These areas are: advanced, high gradient acceleration techniques, and x-ray free electron lasers (XFELs). Further, there is intense interest in the marriage of these two fields, with the goal of producing a very compact XFEL. In this context, recent advances in high gradient radio-frequency cryogenic copper structure research have opened the door to the use of surface electric fields between 250 and 500 MV/m. Such an approach is foreseen to enable a new generation of photoinjectors with six-dimensional beam brightness beyond the current state-of-the-art by well over an order of magnitude. This advance is an essential ingredient enabling an ultra-compact XFEL (UC-XFEL). In addition, one may accelerate these bright beams to GeV scale in less than 10 meters. Such an injector, when combined with inverse free electron laser-based bunching techniques can produce multi-kA beams with unprecedented beam quality, quantified by ~50 nm-rad normalized emittances. These beams, when injected into innovative, short-period (1-10 mm) undulators uniquely enable UC-XFELs having footprints consistent with university-scale laboratories. We describe the architecture and predicted performance of this novel light source, which promises photon production per pulse of a few percent of existing XFEL sources. We review implementation issues including collective beam effects, compact x-ray optics systems, and other relevant technical challenges. To illustrate the potential of such a light source to fundamentally change the current paradigm of XFELs with their limited access, we examine possible applications in biology, chemistry, materials, atomic physics, industry, and medicine which may profit from this new model of performing XFEL science.
△ Less
Submitted 14 August, 2020; v1 submitted 12 March, 2020;
originally announced March 2020.
-
The clustering of luminous red galaxies at z $\sim$ 0.7 from eBOSS and BOSS data
Authors:
Zhongxu Zhai,
Jeremy L. Tinker,
ChangHoon Hahn,
Hee-Jong Seo,
Michael R. Blanton,
Rita Tojeiro,
Hugo O. Camacho,
Marcos Lima,
Aurelio Carnero Rosell,
Flavia Sobreira,
Luiz N. da Costa,
Julian E. Bautista,
Joel R. Brownstein,
Johan Comparat,
Kyle Dawson,
Jeffrey A. Newman,
Alexandre Roman-Lopes,
Donald P. Schneider
Abstract:
We present the first scientific results from the luminous red galaxy sample (LRG) of the extended Baryon Oscillation Spectroscopic Survey (eBOSS). We measure the small and intermediate scale clustering from a sample of more than 61,000 galaxies in the redshift range $0.6 < z < 0.9$. We interpret these measurements in the framework of the Halo Occupation Distribution. The bias of eBOSS LRGs is…
▽ More
We present the first scientific results from the luminous red galaxy sample (LRG) of the extended Baryon Oscillation Spectroscopic Survey (eBOSS). We measure the small and intermediate scale clustering from a sample of more than 61,000 galaxies in the redshift range $0.6 < z < 0.9$. We interpret these measurements in the framework of the Halo Occupation Distribution. The bias of eBOSS LRGs is $2.30 \pm 0.03$, with a satellite fraction of $13\pm3$\% and a mean halo mass of $2.5\times10^{13}h^{-1}M_{\odot}$. These results are consistent with expectations, demonstrating that eBOSS galaxies will be reliable tracers of large scale structure at $z\sim 0.7$. The eBOSS galaxy bias implies a scatter of luminosity at fixed halo mass, $σ_{\log L}$, of 0.19 dex. Using the clustering of massive galaxies from BOSS-CMASS, BOSS-LOWZ, and SDSS, we find that $σ_{\log L}=0.19$ is consistent with observations over the full redshift range that these samples cover. The addition of eBOSS to previous surveys allows investigation of the evolution of massive galaxies over the past $\sim 7$ Gyr.
△ Less
Submitted 13 October, 2017; v1 submitted 18 July, 2016;
originally announced July 2016.
-
Large-scale analysis of the SDSS-III DR8 photometric luminous galaxies angular correlation function
Authors:
Fernando de Simoni,
Flavia Sobreira,
Aurelio Carnero,
Ashley J. Ross,
Hugo O. Camacho,
Rogerio Rosenfeld,
Marcos Lima,
Luiz A. N. da Costa,
Marcio A. G. Maia
Abstract:
We analyse the large-scale angular correlation function (ACF) of the CMASS luminous galaxies (LGs), a photometric-redshift catalogue based on the Data Release 8 (DR8) of the Sloan Digital Sky Survey-III. This catalogue contains over $600 \, \, 000$ LGs in the range $0.45 \leq z \leq 0.65$, which was split into four redshift shells of constant width. First, we estimate the constraints on the redshi…
▽ More
We analyse the large-scale angular correlation function (ACF) of the CMASS luminous galaxies (LGs), a photometric-redshift catalogue based on the Data Release 8 (DR8) of the Sloan Digital Sky Survey-III. This catalogue contains over $600 \, \, 000$ LGs in the range $0.45 \leq z \leq 0.65$, which was split into four redshift shells of constant width. First, we estimate the constraints on the redshift-space distortion (RSD) parameters $bσ_8$ and $fσ_8$, where $b$ is the galaxy bias, $f$ the growth rate and $σ_8$ is the normalization of the perturbations, finding that they vary appreciably among different redshift shells, in agreement with previous results using DR7 data. When assuming constant RSD parameters over the survey redshift range, we obtain $fσ_8 = 0.69 \pm 0.21$, which agrees at the $1.5σ$ level with Baryon Oscillation Spectroscopic Survey DR9 spectroscopic results. Next, we performed two cosmological analyses, where relevant parameters not fitted were kept fixed at their fiducial values. In the first analysis, we extracted the baryon acoustic oscillation peak position for the four redshift shells, and combined with the sound horizon scale from 7-year \textit{Wilkinson Microwave Anisotropy Probe} $(WMAP7)$ to produce the constraints $Ω_{m}=0.249 \pm 0.031$ and $w=-0.885 \pm 0.145$. In the second analysis, we used the ACF full shape information to constrain cosmology using real data for the first time, finding $Ω_{m} = 0.280 \pm 0.022$ and $f_b = Ω_b/Ω_m = 0.211 \pm 0.026$. These results are in good agreement with $WMAP7$ findings, showing that the ACF can be efficiently applied to constrain cosmology in future photometric galaxy surveys.
△ Less
Submitted 2 September, 2013; v1 submitted 2 August, 2013;
originally announced August 2013.
