Minimal Code Research Articles

From GPU to CPU (and Beyond): Extending Hardware Support in GPUSPH Through a SYCL‐Inspired Interface

ABSTRACTWhile most software is originally designed for serial or parallel execution on CPU, and porting to GPU comes later in its development, GPUSPH was designed from the ground up to run on GPUs using CUDA. Making it accessible to a wider audience by introducing support for other computational hardware, and in particular CPUs, poses challenges that are complementary to the ones normally faced when porting CPU code to GPU. We present the approach we have adopted to support CPUs as computational devices in GPUSPH with minimal code changes and low developer effort. Detailed benchmarks illustrating the performance of the implementation and its scalability across multiple cores in both single‐socket and NUMA configurations show good strong and weak scaling.

Live Migration of Video Analytics Applications in Edge Computing

In order to schedule resources efficiently or maintain applications' continuity for mobile customers, edge platforms often need to adaptively migrate the applications on them. However, our measurement shows that existing migration solutions cannot solve the issue of migrating video analytics applications in edge computing because the memory states of video analytics applications have different characteristics from other applications. We conduct a breakdown analysis of the memory states of video analytics applications, and propose to treat three types of states separately with three different techniques, <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">i.e.</i> , warm-up, sync, and replay, to minimize the negative influence of migrations on application performance. Based on this idea, we implement a prototype system in which two new components, <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">i.e.</i> , <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">state store</i> and <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">sidecar</i> , are designed to achieve near-transparent live migration with minimal application code modifications. Evaluation experiments demonstrate that the time of application interruption caused by migrating a video analytics application with our solution is less than 405ms, and our solution does not consume much resources.

POPs identification using simple low-code machine learning

Effectively identifying persistent organic pollutants (POPs) with extensive organic chemical datasets poses a formidable challenge but is of utmost importance. Leveraging machine learning techniques can enhance this process, but previous models often demanded advanced programming skills and high-end computing resources. In this study, we harnessed the simplicity of PyCaret, a Python-based package, to construct machine-learning models for POP screening based on 2D molecular descriptors. We compared the performance of these models against a deep convolutional neural network (DCNN) model. Utilising minimal Python code, we generated several models that exhibited superior or comparable performance to the DCNN. The most outstanding performer, the Light Gradient Boosting Machine (LGBM), achieved an accuracy of 96.20 %, an AUC of 97.70 %, and an F1 score of 82.58 %. This model outshone the DCNN model. Furthermore, it excelled in identifying POPs within the REACH PBT and compiled industrial chemical lists. Our findings highlight the accessibility and simplicity of PyCaret, requiring only a few lines of code, rendering it suitable for non-computing professionals in environmental sciences. The ability of low code machine learning tools (e.g. PyCaret) to facilitate model comparison and interpretation holds promise, encouraging prompt assessment and management of chemical substances.

Spatial bagging to integrate spatial correlation into ensemble machine learning

We propose a novel spatial bagging workflow for predictive ensemble machine learning that improves on standard bagging models. Our proposed method integrates spatial bootstrap for bagging with the number of effective sample size, neff, for integration of the spatial context of the dataset. We benchmark the improved performance over standard machine learning bagging models with a large number of two-dimensional synthetic datasets with varying degrees of Gaussian noise. For noise free datasets, both methods demonstrate equivalent accuracy; however, spatial bagging achieves this with a significantly smaller sample size, showcasing its improved efficiency. As data noise increases, spatial bagging consistently outperforms standard bagging, displaying an improved Mean Squared Error (MSE) and robustness against overfitting. Our proposed spatial bagging method computes the optimal effective sample size for spatial data, reducing model overfitting. Furthermore, our proposed method requires only the additional step of variogram calculation and modeling, and can be implemented with any predictive machine learning bagging model with minimal code modification. i.e., specification of the number of bootstrap samples as the number of effective data. We recommend using spatial bagging for improved predictions for any spatial data setting across diverse scientific fields, e.g., atmospheric, agricultural, subsurface resources etc.

DOMAS: a data management software framework for advanced light sources.

In recent years, China's advanced light sources have entered a period of rapid construction and development. As modern X-ray detectors and data acquisition technologies advance, these facilities are expected to generate massive volumes of data annually, presenting significant challenges in data management and utilization. These challenges encompass data storage, metadata handling, data transfer and user data access. In response, the Data Organization Management Access Software (DOMAS) has been designed as a framework to address these issues. DOMAS encapsulates four fundamental modules of data management software, including metadata catalogue, metadata acquisition, data transfer and data service. For light source facilities, building a data management system only requires parameter configuration and minimal code development within DOMAS. This paper firstly discusses the development of advanced light sources in China and the associated demands and challenges in data management, prompting a reconsideration of data management software framework design. It then outlines the architecture of the framework, detailing its components and functions. Lastly, it highlights the application progress and effectiveness of DOMAS when deployed for the High Energy Photon Source (HEPS) and Beijing Synchrotron Radiation Facility (BSRF).

TH-iSSD: Design and Implementation of a Generic and Reconfigurable Near-Data Processing Framework

We present the design and implementation of TH-iSSD, a near-data processing framework to address the data movement problem. TH-iSSD does not pose any restriction to the hardware selection and is highly reconfigurable—its core components, such as the on-device compute unit (e.g., FPGA, embedded CPUs) and data collectors (e.g., camera, sensors), can be easily replaced to adapt to different use cases. TH-iSSD achieves this goal by incorporating highly flexible computation and data paths. In the data path, TH-iSSD adopts an efficient device-level data switch that exchanges data with both host CPUs and peripheral sensors; it also enables direct accesses between the sensing, computation, and storage hardware components, which completely eliminates the redundant data movement overhead, and thus delivers both high performance and energy efficiency. In the computation path, TH-iSSD provides an abstraction offilestreamfor developers, which abstracts a collection of data along with the related computation task as a file. Since existing applications are familiar with POSIX-like interfaces, they can be ported on top of our platform with minimal code modification. Moreover, TH-iSSD also introduces mechanisms includingpipelined near-data processingandpriority-aware I/O schedulingto make TH-iSSD perform more effectively. We deploy TH-iSSD to accelerate two types of applications: the content-based information retrieval system and the edge zero-streaming system. Our experimental results show that TH-iSSD achieves up to 1.6× higher throughput and 36% lower latency thancompute-centricdesigns.

Leveraging PyCaret for Time Series Analysis-A Low Code Approach

Time series analysis is a fundamental task in various domains, ranging from finance to healthcare and beyond. Traditional methods for time series analysis often require significant manual effort and expertise. PyCaret, a low-code machine learning library, offers a simplified approach to time series analysis, enabling practitioners to build robust models with minimal code. In this paper, we delve into PyCaret’s capabilities for time series analysis, exploring its methods and comparing them with traditional Python packages. Through examples and case studies, we demonstrate how PyCaret streamlines the time series analysis workflow, making it accessible to a broader audience.

Extracting and visualizing hidden activations and computational graphs of PyTorch models with TorchLens

Deep neural network models (DNNs) are essential to modern AI and provide powerful models of information processing in biological neural networks. Researchers in both neuroscience and engineering are pursuing a better understanding of the internal representations and operations that undergird the successes and failures of DNNs. Neuroscientists additionally evaluate DNNs as models of brain computation by comparing their internal representations to those found in brains. It is therefore essential to have a method to easily and exhaustively extract and characterize the results of the internal operations of any DNN. Many models are implemented in PyTorch, the leading framework for building DNN models. Here we introduce TorchLens, a new open-source Python package for extracting and characterizing hidden-layer activations in PyTorch models. Uniquely among existing approaches to this problem, TorchLens has the following features: (1) it exhaustively extracts the results of all intermediate operations, not just those associated with PyTorch module objects, yielding a full record of every step in the model's computational graph, (2) it provides an intuitive visualization of the model's complete computational graph along with metadata about each computational step in a model's forward pass for further analysis, (3) it contains a built-in validation procedure to algorithmically verify the accuracy of all saved hidden-layer activations, and (4) the approach it uses can be automatically applied to any PyTorch model with no modifications, including models with conditional (if–then) logic in their forward pass, recurrent models, branching models where layer outputs are fed into multiple subsequent layers in parallel, and models with internally generated tensors (e.g., injections of noise). Furthermore, using TorchLens requires minimal additional code, making it easy to incorporate into existing pipelines for model development and analysis, and useful as a pedagogical aid when teaching deep learning concepts. We hope this contribution will help researchers in AI and neuroscience understand the internal representations of DNNs.

Generalizing Körner’s graph entropy to graphons

Körner introduced the notion of graph entropy in 1973 as the minimal code rate of a natural coding problem where not all pairs of letters can be distinguished in the alphabet. Later it turned out that it can be expressed as the solution of a minimization problem over the so-called vertex-packing polytope.In this paper we generalize this notion to graphons. We show that the analogous minimization problem provides an upper bound for graphon entropy. We also give a lower bound in the shape of a maximization problem. The main result of the paper is that for most graphons these two bounds actually coincide and hence precisely determine the entropy in question. Furthermore, graphon entropy has a nice connection to the fractional chromatic number and the fractional clique number.

Torchlens: A Python package for extracting and visualizing all hidden layer activations from arbitrary PyTorch models with minimal code

Torchlens: A Python package for extracting and visualizing all hidden layer activations from arbitrary PyTorch models with minimal code

Merak: An Efficient Distributed DNN Training Framework With Automated 3D Parallelism for Giant Foundation Models

Foundation models are in the process of becoming the dominant deep learning technology. Pretraining a foundation model is always time-consuming due to the large scale of both the model parameter and training dataset. Besides being computing-intensive, the pretraining process is extremely memory- and communication-intensive. These challenges make it necessary to apply 3D parallelism, which integrates data parallelism, pipeline model parallelism, and tensor model parallelism, to achieve high training efficiency. However, current 3D parallelism frameworks still encounter two issues: i) they are not transparent to model developers, requiring manual model modification to parallelize training, and ii) their utilization of computation resources, GPU memory, and network bandwidth is insufficient. We propose <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Merak</i> , an automated 3D parallelism deep learning training framework with high resource utilization. Merak automatically deploys 3D parallelism with an automatic model partitioner, which includes a graph-sharding algorithm and proxy node-based model graph. Merak also offers a non-intrusive API to scale out foundation model training with minimal code modification. In addition, we design a high-performance 3D parallel runtime engine that employs several techniques to exploit available training resources, including a shifted critical path pipeline schedule that increases computation utilization, stage-aware recomputation that makes use of idle worker memory, and sub-pipelined tensor model parallelism that overlaps communication and computation. Experiments on 64 GPUs demonstrate Merak's capability to speed up training performance over state-of-the-art 3D parallelism frameworks of models with 1.5, 2.5, 8.3, and 20 billion parameters by up to 1.42, 1.39, 1.43, and 1.61×, respectively.

Implementing commercial inverse design tools for compact, phase-encoded, plasmonic digital logic devices

Numerical simulations have become an essential design tool in the field of photonics, especially for nanophotonics. In particular, 3D finite-difference-time-domain (FDTD) simulations are popular for their powerful design capabilities. Increasingly, researchers are developing or using inverse design tools to improve device footprints and performance. These tools often make use of 3D FDTD simulations and the adjoint optimization method. In this work, we implement a commercial inverse design tool with these features for several plasmonic devices that push the boundaries of the tool. We design a logic gate with complex design requirements, as well as a y-splitter and waveguide crossing. With minimal code changes, we implement a design that incorporates phase encoded inputs in a dielectric-loaded surface plasmon polariton waveguide. The complexity of the requirements in conjunction with limitations in the inverse design tool force us to make concessions regarding the density of encoding and to use on-off keying to encode the outputs. We compare the performance of the inverse-designed devices to conventionally-designed devices with the same operational behavior. Finally, we discuss the limitations of the inverse design tools for realizing complex device designs and comment on what is possible at present and where improvements can be made.

A novel convergence enhancement method based on online dimension reduction optimization

Iterative steady-state solvers are widely used in computational fluid dynamics. Unfortunately, it is difficult to obtain steady-state solutions for unstable problems caused by physical instability and numerical instability. Optimization is a better choice for solving unstable problems because the steady-state solution is always the extreme point of optimization regardless of whether the problem is unstable or ill-conditioned, but it is difficult to solve partial differential equations (PDEs) due to too many optimization variables. In this study, we propose an online dimension reduction optimization method to enhance the convergence of the traditional iterative method to obtain the steady-state solutions of unstable problems. This method performs proper orthogonal decomposition (POD) on the snapshots collected from a few iteration steps of computational fluid dynamics (CFD) simulation, optimizes the POD mode coefficients to minimize the PDE residual to obtain a solution with a lower residual in the POD subspace, and then continues to iterate with the optimized solution as the initial value, repeating the above three steps until the residual converges. Several typical cases show that the proposed method can efficiently calculate the steady-state solution of unstable problems with both the high efficiency and robustness of the iterative method and the good convergence of the optimization method. In addition, this method avoids specific knowledge about the underlying numerical scheme of the CFD code and is easy to implement in almost any iterative solver with minimal code modification.

Leveraging HPC resources with distributed RDataFrame

The declarative approach to data analysis provides high-level abstractions for users to operate on their datasets in a much more ergonomic fashion compared to imperative interfaces. ROOT offers such a tool with RDataFrame, which has been tested in production environments and used in real-world analyses with optimal results. Its programming model acts by creating a computation graph with the operations issued by the user and executing it lazily only when the final results are queried. It has always been oriented towards parallelisation, with native support for multi-thread execution on a single machine. Recently, RDataFrame has been extended with a Python layer that is capable of steering and executing the RDataFrame computation graph over a set of distributed resources. In addition, such a layer requires minimal code changes for an RDataFrame application to run distributedly. The new tool effectively allows running a C++ event loop based on RDataFrame while leveraging common industry tools like Dask to schedule the usage of resources. This work presents results and insights gathered through the distributed RDataFrame tool running a physics analysis connecting multiple nodes with a Dask scheduler that requests resources from a Slurm cluster.

Impunity: Enforcing Physical Unit Consistency at Definition Time in Python

We introduce impunity, a Python library that enables static analysis of code annotations to ensure the consistency of physical units. It provides a framework for developers to annotate their Python code with physical units and automatically verifies if the units are compatible and adhere to predefined coherence rules. impunity comes as a decorator to apply on functions: it analyses the source code to check for consistency of physical dimensions, and applies minimal code rewriting if conversions are necessary. Overall, this approach takes the best of type-checking based methods and dynamic methods and provides a robust approach with no overhead at runtime.

Local and substrate-specific S-palmitoylation determines subcellular localization of G\u03b1o

Peripheral membrane proteins (PMPs) associate with cellular membranes through post-translational modifications like S-palmitoylation. The Golgi apparatus is generally viewed as the transitory station where palmitoyl acyltransferases (PATs) modify PMPs, which are then transported to their ultimate destinations such as the plasma membrane (PM). However, little substrate specificity among the many PATs has been determined. Here we describe the inherent partitioning of Gαo – α-subunit of heterotrimeric Go proteins – to PM and Golgi, independent from Golgi-to-PM transport. A minimal code within Gαo N-terminus governs its compartmentalization and re-coding produces G protein versions with shifted localization. We establish the S-palmitoylation at the outer nuclear membrane assay (“SwissKASH”) to probe substrate specificity of PATs in intact cells. With this assay, we show that PATs localizing to different membrane compartments display remarkable substrate selectivity, which is the basis for PMP compartmentalization. Our findings uncover a mechanism governing protein localization and establish the basis for innovative drug discovery.

Simulation of the five-qubit quantum error correction code on superconducting qubits

Experimental realization of stabilizer-based quantum error correction (QEC) codes that would yield superior logical qubit performance is one of the formidable task for state-of-the-art quantum processors. A major obstacle towards realizing this goal is the large footprint of QEC codes, even those with a small distance. We propose a circuit based on the minimal distance-3 QEC code, which requires only 5 data qubits and 5 ancilla qubits, connected in a ring with iSWAP gates implemented between neighboring qubits. Using a density-matrix simulation, we show that, thanks to its smaller footprint, the proposed code has a lower logical error rate than Surface-17 for similar physical error rates. We also estimate the performance of a neural network-based error decoder, which can be trained to accommodate the error statistics of a specific quantum processor by training on experimental data.

Three Combinatorial Perspectives on Minimal Codes

We develop three approaches of combinatorial flavor to study the structure of minimal codes and cutting blocking sets in finite geometry, each of which has a particular application. The first approach uses techniques from algebraic combinatorics, describing the supports in a linear code via the Alon--Füredi theorem and the combinatorial Nullstellensatz. The second approach combines methods from coding theory and statistics to compare the mean and variance of the nonzero weights in a minimal code. Finally, the third approach regards minimal codes as cutting blocking sets and studies these using the theory of spreads in finite geometry. By applying and combining these approaches with each other, we derive several new bounds and constraints on the parameters of minimal codes. Moreover, we obtain two new constructions of cutting blocking sets of small cardinality in finite projective spaces. In turn, these allow us to give explicit constructions of minimal codes having short length for the given field and dimension.

Analysis and prediction of water quality using deep learning and auto deep learning techniques

Natural water sources like ponds, lakes and rivers are facing a great threat because of activities like discharge of untreated industrial effluents, sewage water, wastes, etc. It is mandatory to examine the water quality to ensure that only safe water is available for consumption. Traditional methods of water quality inspection are a cumbersome process and hence, Artificial Intelligence (AI) can be used as a catalyst for this process. AutoDL is an upcoming field to automate deep learning pipelines and enables model creation and interpretation with minimal code. However, it is still in the nascent stage. This work explores the suitability of adopting AutoDL for Water Quality Assessment by drawing a comparison between AutoDL and a conventional models and analysis to foresee the quality of the water, an appropriate class based on Water Quality Index segregating water bodies into different classes. The accuracy of conventional DL is 1.8% higher than that of AutoDL for binary class water data. The accuracy of conventional DL is 1% higher than that of AutoDL for multiclass water data. The accuracy of conventional model was ~98% to ~99% whereas AutoDL method yielded ~96% to ~98%. However, the AutoDL model ease the task of finding the appropriate DL model and proved better efficiency without manual intervention.

A new multipath channel estimation and mitigation using annihilation filter combined tracking loop implementation in software GPS receivers

The multipath effect causes severe degradation in the positioning of commercial GPS receivers. Due to multipath error, the positioning accuracy could reach a few of 10 m. If the cumulative Multipath delay is less than 0.1–0.35 chips, then it is difficult to mitigate in GPS receivers. This causes severe degradation in GPS signals and can cause a measurement bias. To alleviate this problem, the estimation of multipath parameters using annihilating filter and its mitigation in the GPS tracking loop is proposed in this work. The estimation of randomly generated multipath signals can be performed in the receiver with a lower sampling rate when compared to the larger bandwidth of the GPS baseband signal. Here, the frequency components of the multipath signal in superimposed complex exponentials have been transformed from the time delay and the amplitude of the path observables. The Rayleigh fading model in the urban scenario has been simulated in which the amplitude and the phase of the number of paths (i.e. the frequency component of superimposed complex exponentials) are set and this fading signal is convolved with GPS signal that forms the multipath faded signal. In the GPS receiver post-processing stage, with the help of the annihilation filter, the multipath components are estimated, then an inverse/adaptive filter and compensation technique are further applied to mitigate the multipath component. The mean square error with the different number of paths with noisy environments is analyzed utilizing the cadzow denoising algorithm. The simulation results of the proposed technique employed in the tracking module of the software GPS receiver under severe multipath conditions indicate a substantial enhancement in the performance of the GPS receiver with minimal code and carrier phase error when compared to the least squares and adaptive blind equalization channel techniques. Moreover, the positioning accuracy is also calculated with the inclusion of multipath components in two satellites out of six satellites used in the simulation, the results showed that the annihilation filter improved the mean position accuracy up to 9.3023 m.