71 Hits in 7.6 sec

Compressing Language Models using Doped Kronecker Products [article]

Urmish Thakker, Paul N. Whatmough, Zhi-Gang Liu, Matthew Mattina, Jesse Beu
2020 arXiv   pre-print
Kronecker Products (KP) have been used to compress IoT RNN Applications by 15-38x compression factors, achieving better results than traditional compression methods.  ...  More formally, we propose doping, a process of adding an extremely sparse overlay matrix on top of the pre-defined KP structure. We call this compression method doped kronecker product compression.  ...  This methodology of compression is called Doped Kronecker Product (DKP) in this paper.  ... 
arXiv:2001.08896v5 fatcat:pwyear3j3vepphqrh6zwp7eapm

Doping: A technique for Extreme Compression of LSTM Models using Sparse Structured Additive Matrices

Urmish Thakker, Paul N. Whatmough, Zhi Gang Liu, Matthew Mattina, Jesse G. Beu
2021 Conference on Machine Learning and Systems  
Additionally, results with doped kronecker product matrices demonstrate state-of-the-art accuracy at large compression factors (10 − 25×) across 4 natural language processing applications with minor loss  ...  Structured matrices, such as those derived from Kronecker products (KP), are effective at compressing neural networks, but can lead to unacceptable accuracy loss when applied to large models.  ...  DOPED KRONECKER PRODUCT In this paper we will focus on applying the doping technique to Kronecker product structured matrices.  ... 
dblp:conf/mlsys/ThakkerWLMB21 fatcat:xcxa4buiq5hlfedzh337xpxxhy

Closed-form solutions for the piezoresistivity properties of short-fiber reinforced composites with percolation-type behavior

Federico C. Buroni, Enrique García-Macías
2021 Carbon  
model.  ...  To illustrate the potentials of the proposed formulation, numerical results and discussion are presented on its application to the modeling of the piezoresistive composites doped with carbon nanotubes  ...  The derived formulas are available in MATLAB and Python languages as part of the supplementary material as a tool for free use of the community.  ... 
doi:10.1016/j.carbon.2021.08.083 fatcat:3x4xwsynxrhlvo4kgakmxx42um

Machine Learning for Microcontroller-Class Hardware – A Review [article]

Swapnil Sayan Saha, Sandeep Singh Sandha, Mani Srivastava
2022 arXiv   pre-print
We present both qualitative and numerical insights into different stages of model development by showcasing several use cases.  ...  Researchers use a specialized model development workflow for resource-limited applications to ensure the compute and latency budget is within the device limits while still maintaining the desired performance  ...  Kronecker recurrent units [90] [106] use Kronecker products to stabilize RNN training and decompose large RNN matrices into rank-preserving smaller matrices with fewer parameters, compressing RNN by  ... 
arXiv:2205.14550v3 fatcat:y272riitirhwfgfiotlwv5i7nu

Molecular-Level Study of the Effect of Prior Axial Compression/Torsion on the Axial-Tensile Strength of PPTA Fibers

M. Grujicic, R. Yavari, S. Ramaswami, J. S. Snipes, C.-F. Yen, B. A. Cheeseman
2013 Journal of materials engineering and performance (Print)  
Experimental and computational results available in the relevant open literature were utilized to construct various defects within the molecular-level model and to assign the concentration to these defects  ...  A comprehensive all-atom molecular-level computational investigation is carried out in order to identify and quantify: (i) the effect of prior longitudinal-compressive or axial-torsional loading on the  ...  Acknowledgments The material presented in this paper is based on work supported by the Army Research Office (ARO) research contract entitled ''Multi-length Scale Material Model Development for Armor-grade  ... 
doi:10.1007/s11665-013-0648-2 fatcat:7o72xvyadfd6fj54bzsj32wsua

Interacting fermions in one dimension: from weak to strong correlation [article]

H.J. Schulz
1993 arXiv   pre-print
Using the Kubo formula, one straightforwardly nds (!) = 2K u (!) ; (3.13) i.e. the product K u determines the weight of the dc peak in the conductivity.  ...  The spin susceptibility and the compressibility are equally easy to obtain.  ... 
arXiv:cond-mat/9302006v1 fatcat:r7ehianowzfznfxd7w4jyswb24

Model of phase fluctuations in a latticed-wave superconductor: Application to the Cooper-pair charge-density wave in underdoped cuprates

Ashot Melikyan, Zlatko Tešanović
2005 Physical Review B  
A 4x4 checkerboard-type modulation pattern naturally arises as an energetically favored ground state at and near the x = 1/8 doping and produces good agreement with experimental observations.  ...  The model is geared toward describing not only the long distance but also the intermediate lengthscale physics of underdoped cuprates.  ...  Such a model indeed generically leads to a phase diagram with compressible superfluid and incompressible Wigner crystal states just as our long-distance argument has suggested.  ... 
doi:10.1103/physrevb.71.214511 fatcat:r5kgv4ujjzb23gprf75q7gprzm

Paired chiral spin liquid with a Fermi surface inS=1model on the triangular lattice

Samuel Bieri, Maksym Serbyn, T. Senthil, Patrick A. Lee
2012 Physical Review B  
We use Variational Monte Carlo techniques to calculate the energies of microscopic spin liquid wave functions where spin is represented by three flavors of fermionic spinon operators.  ...  Motivated by recent experiments on Ba3NiSb2O9, we investigate possible quantum spin liquid ground states for spin S=1 Heisenberg models on the triangular lattice.  ...  (D2), in spin language using the Gell-Mann basis, let us mention an equivalent set of generators.  ... 
doi:10.1103/physrevb.86.224409 fatcat:oacbai63ineoxfijma7yfyjaie

Applications and Techniques for Fast Machine Learning in Science

Allison McCarn Deiana, Nhan Tran, Joshua Agar, Michaela Blott, Giuseppe Di Guglielmo, Javier Duarte, Philip Harris, Scott Hauck, Mia Liu, Mark S. Neubauer, Jennifer Ngadiuba, Seda Ogrenci-Memik (+35 others)
2022 Frontiers in Big Data  
A given NN model can be compressed by using low precision quantization instead of single precision.  ...  A notable work here is Deep Compression (Han et al., 2016), which used quantization to compress the model footprint of the SqueezeNet model discussed above, bringing its size to 500x smaller than AlexNet  ... 
doi:10.3389/fdata.2022.787421 pmid:35496379 pmcid:PMC9041419 fatcat:5w2exf7vvrfvnhln7nj5uppjga

Food engineering and technology [chapter]

2017 Advanced Engineering and Technology III  
It is one of the three most widely used cross-platform languages, namelyTcl and Perl (Zhong et al. (2006) ).  ...  It is one of the three most widely used cross-platform languages, namelyTcl and Perl (Zhong et al. (2006) ).  ...  Currently, most of these seed shells are dis carded, although some are made into handicraft products or used as fuel.  ... 
doi:10.4324/9781315387222-8 fatcat:7x3p2xawojexflyeu3erexehoy

Variational Description of Mott Insulators

Manuela Capello, Federico Becca, Michele Fabrizio, Sandro Sorella, Erio Tosatti
2005 Physical Review Letters  
The Gutzwiller wave function for a strongly correlated model can, if supplemented with a long-range Jastrow factor, provide a proper variational description of Mott insulators, so far unavailable.  ...  We demonstrate this concept in the prototypical one-dimensional t-t^' Hubbard model, where at half filling we reproduce all known phases, namely the ordinary Mott undimerized insulator with power-law spin  ...  Let us now consider the behavior of the Jastrow parameters when the doping δ is varied.  ... 
doi:10.1103/physrevlett.94.026406 pmid:15698205 fatcat:knvah2r4c5a6rpg6elg4krmpzq

New Logic Synthesis as Nanotechnology Enabler

Luca Amaru, Pierre-Emmanuel Gaillardon, Subhasish Mitra, Giovanni De Micheli
2015 Proceedings of the IEEE  
We describe models and data-structures for logic design using emerging technologies and we show results of applying new synthesis algorithms and tools.  ...  SiNWFETs can be doped and used to implement standard complementary logic [47] .  ...  His X-Compact technique for test compression has been key to cost-effective manufacturing and high-quality testing of a vast majority of electronic systems, including numerous Intel products.  ... 
doi:10.1109/jproc.2015.2460377 fatcat:kakxk5w5q5bjlfgwirfyjnglwq

Towards the Co-Simulation of Charge Qubits: A Methodology Grounding on an Equivalent Circuit Representation

Elena Blokhina, Andrii Sokolov, Panagiotis Giounanlis, Xutong Wu, Imran Bashir, Dirk Leipold, Robert Bogdan Staszewski, Angelo Brambilla, Federico Bizzarri
2021 IEEE Open Journal of Circuits and Systems  
Results obtained in this study using a circuit simulator correspond to quantum dot arrays fabricated in a CMOS technology and recently published in the literature.  ...  This paper presents a methodology to describe quantum mechanical states of charge qubits, realized as coupled quantum dots occupied by single electrons, using equivalent electrical circuits.  ...  The preferred language for this modeling is Verilog-AMS, which can be compiled and executed by a time-domain simulator such as Cadence Spectre.  ... 
doi:10.1109/ojcas.2021.3105005 fatcat:2defxk2xpjf73gvbezots4tkp4

Non-Abelian Reciprocal Braiding of Weyl Nodes and its Manifestation in ZrTe [article]

Adrien Bouhon, QuanSheng Wu, Robert-Jan Slager, Hongming Weng, Oleg V. Yazyev, Tomáš Bzdušek
2020 arXiv   pre-print
predictive power of our framework is firmly underpinned by tangible material identification, namely zirconium telluride (ZrTe), which we predict to exhibit the aforementioned conversion effect under uniaxial compression  ...  product.  ...  The WPs are located only 50 meV above the Fermi level, which could possibly be further lowered by doping.  ... 
arXiv:1907.10611v3 fatcat:m6kkmqlfsne6jfjtqs7phwx45y

15th International Congress on Instrumentation in Aerospace Simulation Facilities (ICIASF)

1993 Computers & Fluids  
Bacher for their help and useful suggestions throughout this work. Acknowledgments We would like to thank DASA for providing the VoMo Model.  ...  Ron Dykhuizen of Sandia National Laboratories first informed us of the highly dependent nature of acoustic speed in water in a pipe on the elasticity of the pipe walls.  ...  Thus it appears reasonable to "compress" the model by setting the eigenvalues past the first two to zero.  ... 
doi:10.1016/0045-7930(93)90012-x fatcat:5un7rizd3jcblmj2fgh5rciy3a
« Previous Showing results 1 — 15 out of 71 results