-
Hamiltonian quantization of solitons in the $φ^4_{1+1}$ quantum field theory. I. The semiclassical mass shift
Authors:
David M. A. Stuart
Abstract:
We study of the interaction of a scalar quantum field $φ$ with a fixed (external) electromagnetic field $A^{ext}_μdx^μ$ in two dimensional space-time. The dynamics is determined by the action functional \[ S\,=\,\int\,\Bigl(\frac{1}{2}\partial_μφ\partial^μφ- \frac{1}{2}\,g^2\bigl(φ^2-\frac{m^2}{g^2}\bigr)^2 +λε_{μν}\partial_μA^{ext}_νφ\,\Bigr)\,dxdt\,. \] The quartic interaction, a double well pot…
▽ More
We study of the interaction of a scalar quantum field $φ$ with a fixed (external) electromagnetic field $A^{ext}_μdx^μ$ in two dimensional space-time. The dynamics is determined by the action functional \[ S\,=\,\int\,\Bigl(\frac{1}{2}\partial_μφ\partial^μφ- \frac{1}{2}\,g^2\bigl(φ^2-\frac{m^2}{g^2}\bigr)^2 +λε_{μν}\partial_μA^{ext}_νφ\,\Bigr)\,dxdt\,. \] The quartic interaction, a double well potential, supports the existence of solitons in the classical theory. The existence of the corresponding quantum theory can be proved by the methods of constructive quantum field theory. The aim is to analyze the dynamics of the soliton in this quantized theory as $g\to 0 $, which corresponds to a nonrelativistic limit for the soliton, which has mass which diverges as $g^{-2} $ in this limit. In this first paper attention is focused on the case $λ=0$. We develop the analytical framework for quantizing the theory, identify the appropriate degrees of freedom to describe the soliton and the $g\to 0 $ limiting dynamics and provide an interpretation and proof of the Dashen-Hasslacher-Neveu semiclassical mass correction formula.
△ Less
Submitted 4 April, 2019;
originally announced April 2019.
-
Bit-Tactical: Exploiting Ineffectual Computations in Convolutional Neural Networks: Which, Why, and How
Authors:
Alberto Delmas,
Patrick Judd,
Dylan Malone Stuart,
Zissis Poulos,
Mostafa Mahmoud,
Sayeh Sharify,
Milos Nikolic,
Andreas Moshovos
Abstract:
We show that, during inference with Convolutional Neural Networks (CNNs), more than 2x to $8x ineffectual work can be exposed if instead of targeting those weights and activations that are zero, we target different combinations of value stream properties. We demonstrate a practical application with Bit-Tactical (TCL), a hardware accelerator which exploits weight sparsity, per layer precision varia…
▽ More
We show that, during inference with Convolutional Neural Networks (CNNs), more than 2x to $8x ineffectual work can be exposed if instead of targeting those weights and activations that are zero, we target different combinations of value stream properties. We demonstrate a practical application with Bit-Tactical (TCL), a hardware accelerator which exploits weight sparsity, per layer precision variability and dynamic fine-grain precision reduction for activations, and optionally the naturally occurring sparse effectual bit content of activations to improve performance and energy efficiency. TCL benefits both sparse and dense CNNs, natively supports both convolutional and fully-connected layers, and exploits properties of all activations to reduce storage, communication, and computation demands. While TCL does not require changes to the CNN to deliver benefits, it does reward any technique that would amplify any of the aforementioned weight and activation value properties. Compared to an equivalent data-parallel accelerator for dense CNNs, TCLp, a variant of TCL improves performance by 5.05x and is 2.98x more energy efficient while requiring 22% more area.
△ Less
Submitted 9 March, 2018;
originally announced March 2018.
-
Quantization Error as a Metric for Dynamic Precision Scaling in Neural Net Training
Authors:
Ian Taras,
Dylan Malone Stuart
Abstract:
Recent work has explored reduced numerical precision for parameters, activations, and gradients during neural network training as a way to reduce the computational cost of training (Na & Mukhopadhyay, 2016) (Courbariaux et al., 2014). We present a novel dynamic precision scaling (DPS) scheme. Using stochastic fixed-point rounding, a quantization-error based scaling scheme, and dynamic bit-widths d…
▽ More
Recent work has explored reduced numerical precision for parameters, activations, and gradients during neural network training as a way to reduce the computational cost of training (Na & Mukhopadhyay, 2016) (Courbariaux et al., 2014). We present a novel dynamic precision scaling (DPS) scheme. Using stochastic fixed-point rounding, a quantization-error based scaling scheme, and dynamic bit-widths during training, we achieve 98.8% test accuracy on the MNIST dataset using an average bit-width of just 16 bits for weights and 14 bits for activations, compared to the standard 32-bit floating point values used in deep learning frameworks.
△ Less
Submitted 24 January, 2019; v1 submitted 25 January, 2018;
originally announced January 2018.
-
The Schwinger model on $S^1$: Hamiltonian formulation, vacuum and anomaly
Authors:
David M. A. Stuart
Abstract:
We present a Hamiltonian formulation of the Schwinger model on the circle in Coulomb gauge as a semi-bounded self-adjoint operator which is invariant under the modular group ${\cal M}=\Z$ of large gauge transformations. There is a nontrivial action of $\cm$ on fermionic Fock space $\mcH_0$ and its vacuum which plays a role analogous to that of the spectral flow in the formalism involving the infin…
▽ More
We present a Hamiltonian formulation of the Schwinger model on the circle in Coulomb gauge as a semi-bounded self-adjoint operator which is invariant under the modular group ${\cal M}=\Z$ of large gauge transformations. There is a nontrivial action of $\cm$ on fermionic Fock space $\mcH_0$ and its vacuum which plays a role analogous to that of the spectral flow in the formalism involving the infinite Dirac sea. The formulation allows (i) a description of the anomaly and its relation to this group action, and (ii) an explicit identification of the interacting vacuum which arises after the destabilization of the non-interacting vacuum in $\mcH_0$.
△ Less
Submitted 18 June, 2012;
originally announced June 2012.
-
Notes on the Schwinger model: regularization and gauge invariance
Authors:
David M. A. Stuart
Abstract:
The point-splitting computation of the gauge invariant Hamiltonian for the Schwinger model on the circle in a positive energy representation is presented.
The point-splitting computation of the gauge invariant Hamiltonian for the Schwinger model on the circle in a positive energy representation is presented.
△ Less
Submitted 5 June, 2012;
originally announced June 2012.
