Search

We advance the thesis that the simulation of quantum circuits is fundamentally about the efficient management of a large (potentially exponential) number of delimited continuations. The family of Scheme languages, with its efficient implementations of first-class continuations and with its imperative constructs, provides an elegant host for modeling and simulating quantum circuits., Comment: Appeared at Scheme Workshop 2022

Imaginary-time path integral (PI) is a rigorous tool to treat nuclear quantum effects in static properties. However, with its high computational demand, it is crucial to devise precise estimators. We introduce generalized PI estimators for the energy and heat capacity that utilize coordinate mapping. While it can reduce to the standard thermodynamic and centroid virial (CVir) estimators, the formulation can also take advantage of harmonic character of quantum oscillators and crystals to construct a coordinate mapping. This yields harmonically mapped averaging (HMA) estimators, with mappings that decouple (HMAc) or couple (HMAq) the centroid and internal modes. The HMAq is constructed with normal mode coordinates (HMAq-NM) with quadratic scaling of cost or harmonic oscillator staging (HMAq-SG) coordinates with linear scaling. The estimator performance is examined for a 1D anharmonic oscillator and a 3D Lennard-Jones crystal using path integral molecular dynamics (PIMD) simulation. The HMA estimators consistently provide more precise estimates compared to CVir, with the best performance obtained by HMAq-NM, followed by HMAq-SG, and then HMAc. We also examine the effect of anharmonicity (for AO), intrinsic quantumness, and Trotter number. The HMA formulation introduced assumes the availability of forces and Hessian matrix; however, an equally efficient finite difference alternative is possible when these derivatives are inaccessible. The remarkable improvement in precision offered by HMAq estimators provides a framework for efficient PI simulation of more challenging systems, such as those based on \textit{ab initio} calculations.

From SAE Level 3 of automation onwards, drivers are allowed to engage in activities that are not directly related to driving during their travel. However, in level 3, a misunderstanding of the capabilities of the system might lead drivers to engage in secondary tasks, which could impair their ability to react to challenging traffic situations. Anticipating driver activity allows for early detection of risky behaviors, to prevent accidents. To be able to predict the driver activity, a Deep Learning network needs to be trained on a dataset. However, the use of datasets based on simulation for training and the migration to real-world data for prediction has proven to be suboptimal. Hence, this paper presents a real-world driver activity dataset, openly accessible on IEEE Dataport, which encompasses various activities that occur in autonomous driving scenarios under various illumination and weather conditions. Results from the training process showed that the dataset provides an excellent benchmark for implementing models for driver activity recognition., Comment: 6 pages

Classical Arabic represents a significant era, encompassing the golden age of Arab culture, philosophy, and scientific literature. With a broad consensus on the importance of translating these literatures to enrich knowledge dissemination across communities, the advent of large language models (LLMs) and translation systems offers promising tools to facilitate this goal. However, we have identified a scarcity of translation datasets in Classical Arabic, which are often limited in scope and topics, hindering the development of high-quality translation systems. In response, we present the ATHAR dataset, comprising 66,000 high-quality Classical Arabic to English translation samples that cover a wide array of subjects including science, culture, and philosophy. Furthermore, we assess the performance of current state-of-the-art LLMs under various settings, concluding that there is a need for such datasets in current systems. Our findings highlight how models can benefit from fine-tuning or incorporating this dataset into their pretraining pipelines. The dataset is publicly available on the HuggingFace Data Hub at \url{https://huggingface.co/datasets/mohamed-khalil/ATHAR}.

Reversible computing is motivated by both pragmatic and foundational considerations arising from a variety of disciplines. We take a particular path through the development of reversible computation, emphasizing compositional reversible computation. We start from a historical perspective, by reviewing those approaches that developed reversible extensions of lambda-calculi, Turing machines, and communicating process calculi. These approaches share a common challenge: computations made reversible in this way do not naturally compose locally. We then turn our attention to computational models that eschew the detour via existing irreversible models. Building on an original analysis by Landauer, the insights of Bennett, Fredkin, and Toffoli introduced a fresh approach to reversible computing in which reversibility is elevated to the status of the main design principle. These initial models are expressed using low-level bit manipulations, however. Abstracting from the low-level of the Bennett-Fredkin-Toffoli models and pursuing more intrinsic, typed, and algebraic models, naturally leads to rig categories as the canonical model for compositional reversible programming. The categorical model reveals connections to type isomorphisms, symmetries, permutations, groups, and univalent universes. This, in turn, paves the way for extensions to reversible programming based on monads and arrows. These extensions are shown to recover conventional irreversible programming, a variety of reversible computational effects, and more interestingly both pure (measurement-free) and measurement-based quantum programming., Comment: 18 pages

Convergence of path integral simulations requires a substantial number of beads when quantum effects are significant. Traditional Trotter scaling approaches estimate the continuum limit through extrapolation, however they are restricted to the asymptotic behavior near this limit. We introduce an efficient extrapolation approach for thermodynamic properties of quantum oscillators and crystals from primitive path integral simulations. The method utilizes a fitting function inspired by the analytic solution of the harmonic oscillator (HO), or the Einstein crystal for solids. The formulation is for first derivative properties, such as energy and pressure; however, extension to second derivative properties, such as elastic constants, is straightforward. We apply the method to a one-dimensional HO and anharmonic oscillator (AO), as well as a three-dimensional Lennard-Jones crystal. Configurations are sampled using path integral molecular dynamics simulations in the canonical ensemble, at a low temperature of $T=0.1$ (simulation units). Compared to Trotter extrapolation approaches, the new method demonstrates substantial accuracy in estimating the continuum limit, using only a few simulations of relatively smaller system sizes. This capability significantly reduces computational cost, providing a powerful tool to facilitate computations for more complex and challenging systems, such as molecules and real crystals.

Deep neural networks (DNNs) have been widely used in many artificial intelligence (AI) tasks. However, deploying them brings significant challenges due to the huge cost of memory, energy, and computation. To address these challenges, researchers have developed various model compression techniques such as model quantization and model pruning. Recently, there has been a surge in research of compression methods to achieve model efficiency while retaining the performance. Furthermore, more and more works focus on customizing the DNN hardware accelerators to better leverage the model compression techniques. In addition to efficiency, preserving security and privacy is critical for deploying DNNs. However, the vast and diverse body of related works can be overwhelming. This inspires us to conduct a comprehensive survey on recent research toward the goal of high-performance, cost-efficient, and safe deployment of DNNs. Our survey first covers the mainstream model compression techniques such as model quantization, model pruning, knowledge distillation, and optimizations of non-linear operations. We then introduce recent advances in designing hardware accelerators that can adapt to efficient model compression approaches. Additionally, we discuss how homomorphic encryption can be integrated to secure DNN deployment. Finally, we discuss several issues, such as hardware evaluation, generalization, and integration of various compression approaches. Overall, we aim to provide a big picture of efficient DNNs, from algorithm to hardware accelerators and security perspectives., Comment: This manuscript is the accepted version for TNNLS(IEEE Transactions on Neural Networks and Learning Systems)

Imaginary-time path integral (PI) is a rigorous tool to compute static properties at finite temperatures. However, the stiff PI internal modes poses a sampling challenge. This is commonly tackled using staging coordinates, in which the free particle (FP) term of the PI action is diagonalized. We introduce novel and simple staging coordinates that diagonalize the entire action of the harmonic oscillator (HO) model, rendering it efficiently applicable to systems with harmonic character, such as quantum oscillators and crystals. The method is not applicable to fluids or systems with imaginary modes. Unlike FP staging, the HO staging provides a unique treatment of the centroid mode. We provide implementation schemes for PIMC and PIMD in NVT ensemble. Sampling efficiency is assessed in terms of precision and accuracy of estimating the energy and heat capacity of a one-dimensional HO and an asymmetric anharmonic oscillator (AO). In PIMC, the HO coordinates propose collective moves that perfectly sample the HO contribution, then (for AO) the anharmonic term is sampled using standard Metropolis method. This results in a high acceptance rate and, hence, high precision, in comparison to the FP staging. In PIMD, the HO coordinates prescribe definitions for the fictitious masses, yielding equal frequencies when applied to HO model. This allows for larger time step sizes relative to standard staging, without affecting accuracy or integrator stability. We also present results using normal mode (NM) coordinates, based on both HO and FP models. While staging and NM coordinates show similar performance (for FP or HO), staging is computationally preferable due to its cheaper scaling with Trotter number. The enhanced sampling of HO coordinates open avenues for efficient estimation of nuclear quantum effects in more complex systems with harmonic character, such as real molecular bonds and quantum crystals.

As the cost of training ever larger language models has grown, so has the interest in reusing previously learnt knowledge. Transfer learning methods have shown how reusing non-task-specific knowledge can help in subsequent task-specific learning. In this paper, we investigate the inverse: porting whole functional modules that encode task-specific knowledge from one model to another. We designed a study comprising 1,440 training/testing runs to test the portability of modules trained by parameter-efficient finetuning (PEFT) techniques, using sentiment analysis as an example task. We test portability in a wide range of scenarios, involving different PEFT techniques and different pretrained host models, among other dimensions. We compare the performance of ported modules with that of equivalent modules trained (i) from scratch, and (ii) from parameters sampled from the same distribution as the ported module. We find that the ported modules far outperform the two alternatives tested, but that there are interesting performance differences between the four PEFT techniques. We conclude that task-specific knowledge in the form of structurally modular sets of parameters as produced by PEFT techniques is highly portable, but that degree of success depends on type of PEFT and on differences between originating and receiving pretrained models., Comment: Accepted to Findings of EACL 2024. Camera ready version

Rig groupoids provide a semantic model of \PiLang, a universal classical reversible programming language over finite types. We prove that extending rig groupoids with just two maps and three equations about them results in a model of quantum computing that is computationally universal and equationally sound and complete for a variety of gate sets. The first map corresponds to an $8^{\text{th}}$ root of the identity morphism on the unit $1$. The second map corresponds to a square root of the symmetry on $1+1$. As square roots are generally not unique and can sometimes even be trivial, the maps are constrained to satisfy a nondegeneracy axiom, which we relate to the Euler decomposition of the Hadamard gate. The semantic construction is turned into an extension of \PiLang, called \SPiLang, that is a computationally universal quantum programming language equipped with an equational theory that is sound and complete with respect to the Clifford gate set, the standard gate set of Clifford+T restricted to $\le 2$ qubits, and the computationally universal Gaussian Clifford+T gate set.