arXiv subjects

Mehdi Ahmadi

Publications and source records attributed to Mehdi Ahmadi.

At least 19 recordsLinked to original sources

arXiv2023

SwiftLearn: A Data-Efficient Training Method of Deep Learning Models using Importance Sampling

In this paper, we present SwiftLearn, a data-efficient approach to accelerate training of deep learning models using a subset of data samples selected during the warm-up stages of training. This subset is selected based on an importance criteria measured over the entire dataset during warm-up stages, aiming to preserve the model performance with fewer examples during the rest of training. The importance measure we propose could be updated during training every once in a while, to make sure that all of the data samples have a chance to return to the training loop if they show a higher importance. The model architecture is unchanged but since the number of data samples controls the number of forward and backward passes during training, we can reduce the training time by reducing the number of training samples used in each epoch of training. Experimental results on a variety of CV and NLP models during both pretraining and finetuning show that the model performance could be preserved while achieving a significant speed-up during training. More specifically, BERT finetuning on GLUE benchmark shows that almost 90% of the data can be dropped achieving an end-to-end average speedup of 3.36x while keeping the average accuracy drop less than 0.92%.

cs.LG↗

arXiv2021

Quantum clocks observe classical and quantum time dilation

At the intersection of quantum theory and relativity lies the possibility of a clock experiencing a superposition of proper times. We consider quantum clocks constructed from the internal degrees of relativistic particles that move through curved spacetime. The probability that one clock reads a given proper time conditioned on another clock reading a different proper time is derived. From this conditional probability distribution, it is shown that when the center-of-mass of these clocks move in localized momentum wave packets they observe classical time dilation. We then illustrate a quantum correction to the time dilation observed by a clock moving in a superposition of localized momentum wave packets that has the potential to be observed in experiment. The Helstrom-Holevo lower bound is used to derive a proper time-energy/mass uncertainty relation.

quant-ph↗

arXiv2020

CARLA: A Convolution Accelerator with a Reconfigurable and Low-Energy Architecture

Convolutional Neural Networks (CNNs) have proven to be extremely accurate for image recognition, even outperforming human recognition capability. When deployed on battery-powered mobile devices, efficient computer architectures are required to enable fast and energy-efficient computation of costly convolution operations. Despite recent advances in hardware accelerator design for CNNs, two major problems have not yet been addressed effectively, particularly when the convolution layers have highly diverse structures: (1) minimizing energy-hungry off-chip DRAM data movements; (2) maximizing the utilization factor of processing resources to perform convolutions. This work thus proposes an energy-efficient architecture equipped with several optimized dataflows to support the structural diversity of modern CNNs. The proposed approach is evaluated by implementing convolutional layers of VGGNet-16 and ResNet-50. Results show that the architecture achieves a Processing Element (PE) utilization factor of 98% for the majority of 3x3 and 1x1 convolutional layers, while limiting latency to 396.9 ms and 92.7 ms when performing convolutional layers of VGGNet-16 and ResNet-50, respectively. In addition, the proposed architecture benefits from the structured sparsity in ResNet-50 to reduce the latency to 42.5 ms when half of the channels are pruned.

cs.AR↗

arXiv2020

An Energy-Efficient Accelerator Architecture with Serial Accumulation Dataflow for Deep CNNs

Convolutional Neural Networks (CNNs) have shown outstanding accuracy for many vision tasks during recent years. When deploying CNNs on portable devices and embedded systems, however, the large number of parameters and computations result in long processing time and low battery life. An important factor in designing CNN hardware accelerators is to efficiently map the convolution computation onto hardware resources. In addition, to save battery life and reduce energy consumption, it is essential to reduce the number of DRAM accesses since DRAM consumes orders of magnitude more energy compared to other operations in hardware. In this paper, we propose an energy-efficient architecture which maximally utilizes its computational units for convolution operations while requiring a low number of DRAM accesses. The implementation results show that the proposed architecture performs one image recognition task using the VGGNet model with a latency of 393 ms and only 251.5 MB of DRAM accesses.

eess.SP↗

arXiv2019

Quantizing time: Interacting clocks and systems

This article generalizes the conditional probability interpretation of time in which time evolution is realized through entanglement between a clock and a system of interest. This formalism is based upon conditioning a solution to the Wheeler-DeWitt equation on a subsystem of the Universe, serving as a clock, being in a state corresponding to a time $t$. Doing so assigns a conditional state to the rest of the Universe $|ψ_S(t)\rangle$, referred to as the system. We demonstrate that when the total Hamiltonian appearing in the Wheeler-DeWitt equation contains an interaction term coupling the clock and system, the conditional state $|ψ_S(t)\rangle$ satisfies a time-nonlocal Schrödinger equation in which the system Hamiltonian is replaced with a self-adjoint integral operator. This time-nonlocal Schrödinger equation is solved perturbatively and three examples of clock-system interactions are examined. One example considered supposes that the clock and system interact via Newtonian gravity, which leads to the system's Hamiltonian developing corrections on the order of $G/c^4$ and inversely proportional to the distance between the clock and system.

quant-ph↗

arXiv2018

Reproducing AmbientGAN: Generative models from lossy measurements

In recent years, Generative Adversarial Networks (GANs) have shown substantial progress in modeling complex distributions of data. These networks have received tremendous attention since they can generate implicit probabilistic models that produce realistic data using a stochastic procedure. While such models have proven highly effective in diverse scenarios, they require a large set of fully-observed training samples. In many applications access to such samples are difficult or even impractical and only noisy or partial observations of the desired distribution is available. Recent research has tried to address the problem of incompletely observed samples to recover the distribution of the data. \citep{zhu2017unpaired} and \citep{yeh2016semantic} proposed methods to solve ill-posed inverse problem using cycle-consistency and latent-space mappings in adversarial networks, respectively. \citep{bora2017compressed} and \citep{kabkab2018task} have applied similar adversarial approaches to the problem of compressed sensing. In this work, we focus on a new variant of GAN models called AmbientGAN, which incorporates a measurement process (e.g. adding noise, data removal and projection) into the GAN training. While in the standard GAN, the discriminator distinguishes a generated image from a real image, in AmbientGAN model the discriminator has to separate a real measurement from a simulated measurement of a generated image. The results shown by \citep{bora2018ambientgan} are quite promising for the problem of incomplete data, and have potentially important implications for generative approaches to compressed sensing and ill-posed problems.

eess.SP↗

arXiv2018

Brain Tumor Segmentation Using Deep Learning by Type Specific Sorting of Images

Recently deep learning has been playing a major role in the field of computer vision. One of its applications is the reduction of human judgment in the diagnosis of diseases. Especially, brain tumor diagnosis requires high accuracy, where minute errors in judgment may lead to disaster. For this reason, brain tumor segmentation is an important challenge for medical purposes. Currently several methods exist for tumor segmentation but they all lack high accuracy. Here we present a solution for brain tumor segmenting by using deep learning. In this work, we studied different angles of brain MR images and applied different networks for segmentation. The effect of using separate networks for segmentation of MR images is evaluated by comparing the results with a single network. Experimental evaluations of the networks show that Dice score of 0.73 is achieved for a single network and 0.79 in obtained for multiple networks.

cs.CV↗

arXiv2018

Quantification and manipulation of magic states

Magic states can be used as a resource to circumvent the restrictions due to stabilizer-preserving operations, and magic-state conversion has not been studied in the single-copy regime thus far. Here we solve the question of whether a stabilizer-preserving quantum operation exists that can convert between two given magic states in the single-shot regime. We first phrase this question as a feasibility problem for a semi-definite program (SDP), which provides a procedure for constructing a stabilizer-preserving quantum operation (free channel) if it exists. Then we employ a variant of the Farkas Lemma to derive necessary and sufficient conditions for existence, and this method is used to construct a complete set of magic monotones.

quant-ph↗

arXiv2017

Massive Unruh particles cannot be directly observed

We show that massive particles created in a relativistically accelerated reference frame, as predicted by the Unruh effect, can only be found in a tiny layer above the event horizon, whose thickness corresponds to a single Compton wavelength. This is beyond the reach of any detector and suggests that the Unruh effect may not ever be directly observed for massive fields. The case of massless particles is also examined, for which qualitatively different behaviour is observed in a low-acceleration regime, suggesting that an observation of the Unruh effect for massless particles is more promising.

quant-ph↗

arXiv2017

Relativistic (2,3)-threshold quantum secret sharing

In quantum secret sharing protocols, the usual presumption is that the distribution of quantum shares and players' collaboration are both performed inertially. Here we develop a quantum secret sharing protocol that relaxes these assumptions wherein we consider the effects due to the accelerating motion of the shares. Specifically, we solve the (2,3)-threshold continuous-variable quantum secret sharing in non-inertial frames. To this aim, we formulate the effect of relativistic motion on the quantum field inside a cavity as a bosonic quantum Gaussian channel. We investigate how the fidelity of quantum secret sharing is affected by non-uniform motion of the quantum shares. Furthermore, we fully characterize the canonical form of the Gaussian channel which can be utilized in quantum information processing protocols to include relativistic effects.

quant-ph↗

arXiv2017

Spatial entanglement of nonvacuum Gaussian states

The vacuum state of a relativistic quantum field contains entanglement between regions separated by spacelike intervals. Such spatial entanglement can be revealed using an operational method introduced in Ann. Phys. 351, 112 (2014), Phys. Rev. D 91, 016005 (2014). In this approach, a cavity is instantaneously divided into halves by an introduction of an extra perfect mirror. Causal separation of the two regions of the cavity reveals nonlocal spatial correlations present in the field, which can be quantified by measuring particles generated in the process. We use this method to study spatial entanglement properties of nonvacuum Gaussian field states. In particular we show how to enhance the amount of harvested spatial entanglement by an appropriate choice of the initial state of the field in the cavity. We find a counterintuitive influence of the initial entanglement between cavity modes on the spatial entanglement which is revealed by dividing the cavity in half.

quant-ph↗

arXiv2016

Communication between inertial observers with partially correlated reference frames

In quantum communication protocols the existence of a shared reference frame between two spatially separated parties is normally presumed. However, in many practical situations we are faced with the problem of misaligned reference frames. In this paper, we study communication between two inertial observers who have partial knowledge about the Lorentz transformation that relates their frames of reference. Since every Lorentz transformation can be decomposed into a pure boost followed by a rotation, we begin by analysing the effects on communication when the parties have partial knowledge about the transformation relating their frames, when the transformation is either a rotation or pure boost. This then enables us to investigate how the efficiency of communication is affected due to partially correlated inertial reference frames related by an arbitrary Lorentz transformation. Furthermore, we show how the results of previous studies where reference frames are completely uncorrelated are recovered from our results in appropriate limits.

quant-ph↗

arXiv2016

Effect of relativistic acceleration on localized two-mode Gaussian quantum states

We study how an arbitrary Gaussian state of two localized wave packets, prepared in an inertial frame of reference, is described by a pair of uniformly accelerated observers. We explicitly compute the resulting state for arbitrarily chosen proper accelerations of the observers and independently tuned distance between them. To do so, we introduce a generalized Rindler frame of reference and analytically derive the corresponding state transformation as a Gaussian channel. Our approach provides several new insights into the phenomenon of vacuum entanglement such as the highly non-trivial effect of spatial separation between the observers including sudden death of entanglement. We also calculate the fidelity of the two-mode channel for non-vacuum Gaussian states and obtain bounds on classical and quantum capacities of a single-mode channel. Our framework can be directly applied to any continuous variable quantum information protocol in which the effects of acceleration or gravity cannot be neglected.

quant-ph↗

arXiv2014

Phonon creation by gravitational waves

We show that gravitational waves create phonons in a Bose-Einstein condensate (BEC). A traveling spacetime distortion produces particle creation resonances that correspond to the dynamical Casimir effect in a BEC phononic field contained in a cavity-type trap. We propose to use this effect to detect gravitational waves. The amplitude of the wave can be estimated applying recently developed relativistic quantum metrology techniques. We provide the optimal precision bound on the estimation of the wave's amplitude. Finally, we show that the parameter regime required to detect gravitational waves with this technique could be, in principle, within experimental reach in a medium-term timescale.

quant-ph↗

arXiv2014

Relativistic Quantum Metrology: Exploiting relativity to improve quantum measurement technologies

We present a framework for relativistic quantum metrology that is useful for both Earth-based and space-based technologies. Quantum metrology has been so far successfully applied to design precision instruments such as clocks and sensors which outperform classical devices by exploiting quantum properties. There are advanced plans to implement these and other quantum technologies in space, for instance Space-QUEST and Space Optical Clock projects intend to implement quantum communications and quantum clocks at regimes where relativity starts to kick in. However, typical setups do not take into account the effects of relativity on quantum properties. To include and exploit these effects, we introduce techniques for the application of metrology to quantum field theory. Quantum field theory properly incorporates quantum theory and relativity, in particular, at regimes where space-based experiments take place. This framework allows for high precision estimation of parameters that appear in quantum field theory including proper times and accelerations. Indeed, the techniques can be applied to develop a novel generation of relativistic quantum technologies for gravimeters, clocks and sensors. As an example, we present a high precision device which in principle improves the state-of-the-art in quantum accelerometers by exploiting relativistic effects.

quant-ph↗

arXiv2014

Quantum parameter estimation with imperfect reference frames

Quantum metrology studies quantum strategies which enable us to outperform their classical counterparts. In this framework, the existence of perfect classical reference frames is usually assumed. However, such ideal reference frames might not always be available. The reference frames required in metrology strategies can either degrade or become misaligned during the estimation process. We investigate how the imperfectness of reference frames leads to noise which in general affects the ultimate precision limits in measurement of physical parameters. Moreover, since quantum parameter estimation can be phrased as a quantum communication protocol between two parties, our results provide deeper insight into quantum communication protocols with misaligned reference frames. Our framework allows for the study of general noise on the efficiency of such schemes.

quant-ph↗

arXiv2013

Quantum metrology for relativistic quantum fields

In quantum metrology quantum properties such as squeezing and entanglement are exploited in the design of a new generation of clocks, sensors and other measurement devices that can outperform their classical counterparts. Applications of great technological relevance lie in the precise measurement of parameters which play a central role in relativity, such as proper accelerations, relative distances, time and gravitational field strengths. In this paper we generalise recently introduced techniques to estimate physical quantities within quantum field theory in flat and curved space-time. We consider a bosonic quantum field that undergoes a generic transformation, which encodes the parameter to be estimated. We present analytical formulas for optimal precision bounds on the estimation of small parameters in terms of Bogoliubov coefficients for single mode and two-mode Gaussian channels.

quant-ph↗

arXiv2012

The WAY theorem and the quantum resource theory of asymmetry

The WAY theorem establishes an important constraint that conservation laws impose on quantum mechanical measurements. We formulate the WAY theorem in the broader context of resource theories, where one is constrained to a subset of quantum mechanical operations described by a symmetry group. Establishing connections with the theory of quantum state discrimination we obtain optimal unitaries describing the measurement of arbitrary observables, explain how prior information can permit perfect measurements that circumvent the WAY constraint, and provide a framework that establishes a natural ordering on measurement apparatuses through a decomposition into asymmetry and charge subsystems.

quant-ph↗