arXiv subjects

Achim Streit

Publications and source records attributed to Achim Streit.

At least 37 records · Page 2Linked to original sources

arXiv2025

Jigsaw: Training Multi-Billion-Parameter AI Weather Models with Optimized Model Parallelism

AI-based methods have revolutionized atmospheric forecasting, with recent successes in medium-range forecasting spurring the development of climate foundation models. Accurate modeling of complex atmospheric dynamics at high spatial resolutions and longer lead times requires large neural networks and gigabyte-sized data samples, making accelerator memory and I/O-bandwidth the bottlenecks for model training. We introduce WeatherMixer, a multi-layer-perceptron-based architecture whose workload scales linearly with input size, allowing the model to learn global weather phenomena at accuracies similar to numerical weather prediction. To cope with the computational demand, we propose Jigsaw, a novel model parallelization scheme that employs both domain and tensor parallelism, eliminating memory redundancy. Jigsaw exceeds state-of-the-art performance in strong scaling in compute-communication-limited systems and achieves superscalar weak scaling in I/O-bandwidth-limited systems. We scale training to 256 GPUs, reaching peak performances of 9 and 11 PFLOPs, 23% and 28% of theoretical peaks, achieving 68% and 72% scaling efficiency versus 51% without model parallelism.

cs.LG↗

arXiv2025

QML Essentials -- A framework for working with Quantum Fourier Models

In this work, we propose a framework in the form of a Python package, specifically designed for the analysis of Quantum Machine Learning models. This framework is based on the PennyLane simulator and facilitates the evaluation and training of Variational Quantum Circuits. It provides additional functionality ranging from the ability to add different types of noise to the classical simulation, over different parameter initialisation strategies, to the calculation of expressibility and entanglement for a given model. As an intrinsic property of Quantum Fourier Models, it provides two methods for calculating the corresponding Fourier spectrum: one via the Fast Fourier Transform and another analytical method based on the expansion of the expectation value using trigonometric polynomials. It also provides a set of predefined approaches that allow a fast and straightforward implementation of Quantum Machine Learning models. With this framework, we extend the PennyLane simulator with a set of tools that allow researchers a more convenient start with Quantum Fourier Models and aim to unify the analysis of Variational Quantum Circuits.

quant-ph↗

arXiv2025

From Hope to Heuristic: Realistic Runtime Estimates for Quantum Optimisation in NHEP

Noisy Intermediate-Scale Quantum (NISQ) computers, despite their limitations, present opportunities for near-term quantum advantages in Nuclear and High-Energy Physics (NHEP) when paired with specially designed quantum algorithms and processing units. This study focuses on core algorithms that solve optimisation problems through the quadratic Ising or quadratic unconstrained binary optimisation model, specifically quantum annealing and the Quantum Approximate Optimisation Algorithm (QAOA). In particular, we estimate runtimes and scalability for the task of particle track reconstruction, a key computing challenge in NHEP, and investigate how the classical parameter space in QAOA, along with techniques like a Fourier-analysis based heuristic, can facilitate future quantum advantages. The findings indicate that lower frequency components in the parameter space are crucial for effective annealing schedules, suggesting that heuristics can improve resource efficiency while achieving near-optimal results. Overall, the study highlights the potential of NISQ computers in NHEP and the significance of co-design approaches and heuristic techniques in overcoming challenges in quantum algorithms.

quant-ph↗

arXiv2025

Model Fusion via Neuron Transplantation

Ensemble learning is a widespread technique to improve the prediction performance of neural networks. However, it comes at the price of increased memory and inference time. In this work we propose a novel model fusion technique called \emph{Neuron Transplantation (NT)} in which we fuse an ensemble of models by transplanting important neurons from all ensemble members into the vacant space obtained by pruning insignificant neurons. An initial loss in performance post-transplantation can be quickly recovered via fine-tuning, consistently outperforming individual ensemble members of the same model capacity and architecture. Furthermore, NT enables all the ensemble members to be jointly pruned and jointly trained in a combined model. Comparing it to alignment-based averaging (like Optimal-Transport-fusion), it requires less fine-tuning than the corresponding OT-fused model, the fusion itself is faster and requires less memory, while the resulting model performance is comparable or better. The code is available under the following link: https://github.com/masterbaer/neuron-transplantation.

cs.LG↗

arXiv2024

A Comparative Study of Pruning Methods in Transformer-based Time Series Forecasting

The current landscape in time-series forecasting is dominated by Transformer-based models. Their high parameter count and corresponding demand in computational resources pose a challenge to real-world deployment, especially for commercial and scientific applications with low-power embedded devices. Pruning is an established approach to reduce neural network parameter count and save compute. However, the implications and benefits of pruning Transformer-based models for time series forecasting are largely unknown. To close this gap, we provide a comparative benchmark study by evaluating unstructured and structured pruning on various state-of-the-art multivariate time series models. We study the effects of these pruning strategies on model predictive performance and computational aspects like model size, operations, and inference time. Our results show that certain models can be pruned even up to high sparsity levels, outperforming their dense counterpart. However, fine-tuning pruned models is necessary. Furthermore, we demonstrate that even with corresponding hardware and software support, structured pruning is unable to provide significant time savings.

cs.LG↗

arXiv2024

FAIR Digital Objects for the Realization of Globally Aligned Data Spaces

The FAIR principles are globally accepted guidelines for improved data management practices with the potential to align data spaces on a global scale. In practice, this is only marginally achieved through the different ways in which organizations interpret and implement these principles. The concept of FAIR Digital Objects provides a way to realize a domain-independent abstraction layer that could solve this problem, but its specifications are currently diverse, contradictory, and restricted to semantic models. In this work, we introduce a rigorously formalized data model with a set of assertions using formal expressions to provide a common baseline for the implementation of FAIR Digital Objects. The model defines how these objects enable machine-actionable decisions based on the principles of abstraction, encapsulation, and entity relationship to fulfill FAIR criteria for the digital resources they represent. We provide implementation examples in the context of two use cases and explain how our model can facilitate the (re)use of data across domains. We also compare how our model assertions are met by FAIR Digital Objects as they have been described in other projects. Finally, we discuss our results' adoption criteria, limitations, and perspectives in the big data context. Overall, our work represents an important milestone for various communities working towards globally aligned data spaces through FAIRification.

cs.DB↗

arXiv2024

Harnessing Orthogonality to Train Low-Rank Neural Networks

This study explores the learning dynamics of neural networks by analyzing the singular value decomposition (SVD) of their weights throughout training. Our investigation reveals that an orthogonal basis within each multidimensional weight's SVD representation stabilizes during training. Building upon this, we introduce Orthogonality-Informed Adaptive Low-Rank (OIALR) training, a novel training method exploiting the intrinsic orthogonality of neural networks. OIALR seamlessly integrates into existing training workflows with minimal accuracy loss, as demonstrated by benchmarking on various datasets and well-established network architectures. With appropriate hyperparameter tuning, OIALR can surpass conventional training setups, including those of state-of-the-art models.

cs.LG↗

arXiv2024

AB-Training: A Communication-Efficient Approach for Distributed Low-Rank Learning

Communication bottlenecks severely hinder the scalability of distributed neural network training, particularly in high-performance computing (HPC) environments. We introduce AB-training, a novel data-parallel method that leverages low-rank representations and independent training groups to significantly reduce communication overhead. Our experiments demonstrate an average reduction in network traffic of approximately 70.31\% across various scaling scenarios, increasing the training potential of communication-constrained systems and accelerating convergence at scale. AB-training also exhibits a pronounced regularization effect at smaller scales, leading to improved generalization while maintaining or even reducing training time. We achieve a remarkable 44.14 : 1 compression ratio on VGG16 trained on CIFAR-10 with minimal accuracy loss, and outperform traditional data parallel training by 1.55\% on ResNet-50 trained on ImageNet-2012. While AB-training is promising, our findings also reveal that large batch effects persist even in low-rank regimes, underscoring the need for further research into optimized update mechanisms for massively distributed training.

cs.LG↗

arXiv2024

Modeling Distributed Computing Infrastructures for HEP Applications

Predicting the performance of various infrastructure design options in complex federated infrastructures with computing sites distributed over a wide area network that support a plethora of users and workflows, such as the Worldwide LHC Computing Grid (WLCG), is not trivial. Due to the complexity and size of these infrastructures, it is not feasible to deploy experimental test-beds at large scales merely for the purpose of comparing and evaluating alternate designs. An alternative is to study the behaviours of these systems using simulation. This approach has been used successfully in the past to identify efficient and practical infrastructure designs for High Energy Physics (HEP). A prominent example is the Monarc simulation framework, which was used to study the initial structure of the WLCG. New simulation capabilities are needed to simulate large-scale heterogeneous computing systems with complex networks, data access and caching patterns. A modern tool to simulate HEP workloads that execute on distributed computing infrastructures based on the SimGrid and WRENCH simulation frameworks is outlined. Studies of its accuracy and scalability are presented using HEP as a case-study. Hypothetical adjustments to prevailing computing architectures in HEP are studied providing insights into the dynamics of a part of the WLCG and candidates for improvements.

cs.DC↗

arXiv2024

ReCycle: Fast and Efficient Long Time Series Forecasting with Residual Cyclic Transformers

Transformers have recently gained prominence in long time series forecasting by elevating accuracies in a variety of use cases. Regrettably, in the race for better predictive performance the overhead of model architectures has grown onerous, leading to models with computational demand infeasible for most practical applications. To bridge the gap between high method complexity and realistic computational resources, we introduce the Residual Cyclic Transformer, ReCycle. ReCycle utilizes primary cycle compression to address the computational complexity of the attention mechanism in long time series. By learning residuals from refined smoothing average techniques, ReCycle surpasses state-of-the-art accuracy in a variety of application use cases. The reliable and explainable fallback behavior ensured by simple, yet robust, smoothing average techniques additionally lowers the barrier for user acceptance. At the same time, our approach reduces the run time and energy consumption by more than an order of magnitude, making both training and inference feasible on low-performance, low-power and edge computing devices. Code is available at https://github.com/Helmholtz-AI-Energy/ReCycle

cs.LG↗

arXiv2023

Resource-aware Research on Universe and Matter: Call-to-Action in Digital Transformation

Given the urgency to reduce fossil fuel energy production to make climate tipping points less likely, we call for resource-aware knowledge gain in the research areas on Universe and Matter with emphasis on the digital transformation. A portfolio of measures is described in detail and then summarized according to the timescales required for their implementation. The measures will both contribute to sustainable research and accelerate scientific progress through increased awareness of resource usage. This work is based on a three-days workshop on sustainability in digital transformation held in May 2023.

physics.comp-ph↗

arXiv2023

Massively Parallel Genetic Optimization through Asynchronous Propagation of Populations

We present Propulate, an evolutionary optimization algorithm and software package for global optimization and in particular hyperparameter search. For efficient use of HPC resources, Propulate omits the synchronization after each generation as done in conventional genetic algorithms. Instead, it steers the search with the complete population present at time of breeding new individuals. We provide an MPI-based implementation of our algorithm, which features variants of selection, mutation, crossover, and migration and is easy to extend with custom functionality. We compare Propulate to the established optimization tool Optuna. We find that Propulate is up to three orders of magnitude faster without sacrificing solution accuracy, demonstrating the efficiency and efficacy of our lazy synchronization approach. Code and documentation are available at https://github.com/Helmholtz-AI-Energy/propulate

cs.NE↗

arXiv2022

Learning Tree Structures from Leaves For Particle Decay Reconstruction

In this work, we present a neural approach to reconstructing rooted tree graphs describing hierarchical interactions, using a novel representation we term the Lowest Common Ancestor Generations (LCAG) matrix. This compact formulation is equivalent to the adjacency matrix, but enables learning a tree's structure from its leaves alone without the prior assumptions required if using the adjacency matrix directly. Employing the LCAG therefore enables the first end-to-end trainable solution which learns the hierarchical structure of varying tree sizes directly, using only the terminal tree leaves to do so. In the case of high-energy particle physics, a particle decay forms a hierarchical tree structure of which only the final products can be observed experimentally, and the large combinatorial space of possible trees makes an analytic solution intractable. We demonstrate the use of the LCAG as a target in the task of predicting simulated particle physics decay structures using both a Transformer encoder and a Neural Relational Inference encoder Graph Neural Network. With this approach, we are able to correctly predict the LCAG purely from leaf features for a maximum tree-depth of $8$ in $92.5\%$ of cases for trees up to $6$ leaves (including) and $59.7\%$ for trees up to $10$ in our simulated dataset.

physics.comp-ph↗

arXiv2022

HyDe: The First Open-Source, Python-Based, GPU-Accelerated Hyperspectral Denoising Package

As with any physical instrument, hyperspectral cameras induce different kinds of noise in the acquired data. Therefore, Hyperspectral denoising is a crucial step for analyzing hyperspectral images (HSIs). Conventional computational methods rarely use GPUs to improve efficiency and are not fully open-source. Alternatively, deep learning-based methods are often open-source and use GPUs, but their training and utilization for real-world applications remain non-trivial for many researchers. Consequently, we propose HyDe: the first open-source, GPU-accelerated Python-based, hyperspectral image denoising toolbox, which aims to provide a large set of methods with an easy-to-use environment. HyDe includes a variety of methods ranging from low-rank wavelet-based methods to deep neural network (DNN) models. HyDe's interface dramatically improves the interoperability of these methods and the performance of the underlying functions. In fact, these methods maintain similar HSI denoising performance to their original implementations while consuming nearly ten times less energy. Furthermore, we present a method for training DNNs for denoising HSIs which are not spatially related to the training dataset, i.e., training on ground-level HSIs for denoising HSIs with other perspectives including airborne, drone-borne, and space-borne. To utilize the trained DNNs, we show a sliding window method to effectively denoise HSIs which would otherwise require more than 40 GB. The package can be found at: \url{https://github.com/Helmholtz-AI-Energy/HyDe}.

cs.CV↗

arXiv2021

Accelerating Neural Network Training with Distributed Asynchronous and Selective Optimization (DASO)

With increasing data and model complexities, the time required to train neural networks has become prohibitively large. To address the exponential rise in training time, users are turning to data parallel neural networks (DPNN) to utilize large-scale distributed resources on computer clusters. Current DPNN approaches implement the network parameter updates by synchronizing and averaging gradients across all processes with blocking communication operations. This synchronization is the central algorithmic bottleneck. To combat this, we introduce the Distributed Asynchronous and Selective Optimization (DASO) method which leverages multi-GPU compute node architectures to accelerate network training. DASO uses a hierarchical and asynchronous communication scheme comprised of node-local and global networks while adjusting the global synchronization rate during the learning process. We show that DASO yields a reduction in training time of up to 34% on classical and state-of-the-art networks, as compared to other existing data parallel training methods.

cs.LG↗

arXiv2020

HeAT -- a Distributed and GPU-accelerated Tensor Framework for Data Analytics

To cope with the rapid growth in available data, the efficiency of data analysis and machine learning libraries has recently received increased attention. Although great advancements have been made in traditional array-based computations, most are limited by the resources available on a single computation node. Consequently, novel approaches must be made to exploit distributed resources, e.g. distributed memory architectures. To this end, we introduce HeAT, an array-based numerical programming framework for large-scale parallel processing with an easy-to-use NumPy-like API. HeAT utilizes PyTorch as a node-local eager execution engine and distributes the workload on arbitrarily large high-performance computing systems via MPI. It provides both low-level array computations, as well as assorted higher-level algorithms. With HeAT, it is possible for a NumPy user to take full advantage of their available resources, significantly lowering the barrier to distributed data analysis. When compared to similar frameworks, HeAT achieves speedups of up to two orders of magnitude.

cs.DC↗

arXiv2019

German-Russian Astroparticle Data Life Cycle Initiative

A data life cycle (DLC) is a high-level data processing pipeline that involves data acquisition, event reconstruction, data analysis, publication, archiving, and sharing. For astroparticle physics a DLC is particularly important due to the geographical and content diversity of the research field. A dedicated and experiment spanning analysis and data centre would ensure that multi-messenger analyses can be carried out using state-of-the-art methods. The German-Russian Astroparticle Data Life Cycle Initiative (GRADLCI) is a joint project of the KASCADE-Grande and TAIGA collaborations, aimed at developing a concept and creating a DLC prototype that takes into account the data processing features specific for the research field. An open science system based on the KASCADE Cosmic Ray Data Centre (KCDC), which is a web-based platform to provide the astroparticle physics data for the general public, must also include effective methods for distributed data storage algorithms and techniques to allow the community to perform simulations and analyses with sophisticated machine learning methods. The aim is to achieve more efficient analyses of the data collected in different, globally dispersed observatories, as well as a modern education to Big Data Scientist in the synergy between basic research and the information society. The contribution covers the status and future plans of the initiative.

astro-ph.IM↗

arXiv2018

Russian-German Astroparticle Data Life Cycle Initiative

Modern large-scale astroparticle setups measure high-energy particles, gamma rays, neutrinos, radio waves, and the recently discovered gravitational waves. Ongoing and future experiments are located worldwide. The data acquired have different formats, storage concepts, and publication policies. Such differences are a crucial point in the era of Big Data and of multi-messenger analysis in astroparticle physics. We propose an open science web platform called ASTROPARTICLE.ONLINE which enables us to publish, store, search, select, and analyze astroparticle data. In the first stage of the project, the following components of a full data life cycle concept are under development: describing, storing, and reusing astroparticle data; software to perform multi-messenger analysis using deep learning; and outreach for students, post-graduate students, and others who are interested in astroparticle physics. Here we describe the concepts of the web platform and the first obtained results, including the meta data structure for astroparticle data, data analysis by using convolution neural networks, description of the binary data, and the outreach platform for those interested in astroparticle physics. The KASCADE-Grande and TAIGA cosmic-ray experiments were chosen as pilot examples.

astro-ph.IM↗