arXiv subjects

Vassil Alexandrov

Publications and source records attributed to Vassil Alexandrov.

6 recordsLinked to original sources

arXiv2026

TokaMind: A Multi-Modal Transformer Foundation Model for Tokamak Plasma Dynamics

We present TokaMind, to our knowledge the first open-source foundation model for tokamak plasma dynamics, based on a Multi-Modal Transformer (MMT) and pretrained on heterogeneous diagnostics from the publicly available MAST dataset. TokaMind supports multiple data modalities (time-series, 2D profiles, and videos) with different sampling rates, robust missing-signal handling, and efficient task adaptation via selectively loading and freezing four model components. To represent multi-modal signals, we use a lightweight fixed-basis Discrete Cosine Transform embedding (DCT3D) and provide a clean interface for alternative embeddings (e.g., Variational Autoencoders). We evaluate TokaMind on the recently introduced MAST benchmark TokaMark, which comprises 14 tasks with heterogeneous reconstruction and forecasting objectives. Our results show that fine-tuned TokaMind outperforms the strongest benchmark baseline on all but one task. Compared with training the same architecture from scratch under a matched epoch budget, warm-start adaptation is most beneficial on demanding downstream settings, including long-horizon forecasting and high-dimensional equilibrium objectives. These findings highlight the value of multi-modal pretraining for tokamak plasma dynamics and provide a practical, extensible foundation for future fusion modeling tasks. Training code and model weights are publicly available at github.com/UKAEA-IBM-STFC-Fusion-FMs/tokamind and huggingface.co/UKAEA-IBM-STFC, respectively.

physics.plasm-ph↗

arXiv2025

Fast Linear Solvers via AI-Tuned Markov Chain Monte Carlo-based Matrix Inversion

Large, sparse linear systems are pervasive in modern science and engineering, and Krylov subspace solvers are an established means of solving them. Yet convergence can be slow for ill-conditioned matrices, so practical deployments usually require preconditioners. Markov chain Monte Carlo (MCMC)-based matrix inversion can generate such preconditioners and accelerate Krylov iterations, but its effectiveness depends on parameters whose optima vary across matrices; manual or grid search is costly. We present an AI-driven framework recommending MCMC parameters for a given linear system. A graph neural surrogate predicts preconditioning speed from $A$ and MCMC parameters. A Bayesian acquisition function then chooses the parameter sets most likely to minimise iterations. On a previously unseen ill-conditioned system, the framework achieves better preconditioning with 50\% of the search budget of conventional methods, yielding about a 10\% reduction in iterations to convergence. These results suggest a route for incorporating MCMC-based preconditioners into large-scale systems.

cs.LG↗

arXiv2024

On Advanced Monte Carlo Methods for Linear Algebra on Advanced Accelerator Architectures

In this paper we present computational experiments with the Markov Chain Monte Carlo Matrix Inversion ($(\text{MC})^2\text{MI}$) on several accelerator architectures and investigate their impact on performance and scalability of the method. The method is used as a preconditioner and for solving the corresponding system of linear equations iterative methods, such as generalized minimal residuals (GMRES) or bi-conjugate gradient (stabilized) (BICGstab), are used. Numerical experiments are carried out to highlight the benefits and deficiencies of both approaches and to assess their overall usefulness in light of scalability of the method.

math.NA↗

arXiv2023

A Practitioner's Guide to Quantum Algorithms for Optimisation Problems

Quantum computing is gaining popularity across a wide range of scientific disciplines due to its potential to solve long-standing computational problems that are considered intractable with classical computers. One promising area where quantum computing has potential is in the speed-up of NP-hard optimisation problems that are common in industrial areas such as logistics and finance. Newcomers to the field of quantum computing who are interested in using this technology to solve optimisation problems do not have an easily accessible source of information on the current capabilities of quantum computers and algorithms. This paper aims to provide a comprehensive overview of the theory of quantum optimisation techniques and their practical application, focusing on their near-term potential for noisy intermediate scale quantum devices. Two main paradigms for quantum hardware are then discussed: quantum annealing and gate-based quantum computing. While quantum annealers are effective for some optimisation problems, they have limitations and cannot be used for universal quantum computation. In contrast, gate-based quantum computers offer the potential for universal quantum computation, but they face challenges with hardware limitations and accurate gate implementation. The paper provides a detailed mathematical discussion with references to key works in the field, as well as a more practical discussion with relevant examples. The most popular techniques for quantum optimisation on gate-based quantum computers, the quantum approximate optimisation (QAO) algorithm and the quantum alternating operator ansatz (QAOA) framework, are discussed in detail. The paper concludes with a discussion of the challenges facing quantum optimisation techniques and the need for further research and development to identify new, effective methods for achieving quantum advantage.

quant-ph↗

arXiv2014

Automating Fault Tolerance in High-Performance Computational Biological Jobs Using Multi-Agent Approaches

Background: Large-scale biological jobs on high-performance computing systems require manual intervention if one or more computing cores on which they execute fail. This places not only a cost on the maintenance of the job, but also a cost on the time taken for reinstating the job and the risk of losing data and execution accomplished by the job before it failed. Approaches which can proactively detect computing core failures and take action to relocate the computing core's job onto reliable cores can make a significant step towards automating fault tolerance. Method: This paper describes an experimental investigation into the use of multi-agent approaches for fault tolerance. Two approaches are studied, the first at the job level and the second at the core level. The approaches are investigated for single core failure scenarios that can occur in the execution of parallel reduction algorithms on computer clusters. A third approach is proposed that incorporates multi-agent technology both at the job and core level. Experiments are pursued in the context of genome searching, a popular computational biology application. Result: The key conclusion is that the approaches proposed are feasible for automating fault tolerance in high-performance computing systems with minimal human intervention. In a typical experiment in which the fault tolerance is studied, centralised and decentralised checkpointing approaches on an average add 90% to the actual time for executing the job. On the other hand, in the same experiment the multi-agent approaches add only 10% to the overall execution time.

cs.DC↗

arXiv2013

Can Agent Intelligence be used to Achieve Fault Tolerant Parallel Computing Systems?

The work reported in this paper is motivated towards validating an alternative approach for fault tolerance over traditional methods like checkpointing that constrain efficacious fault tolerance. Can agent intelligence be used to achieve fault tolerant parallel computing systems? If so, "What agent capabilities are required for fault tolerance?", "What parallel computational tasks can benefit from such agent capabilities?" and "How can agent capabilities be implemented for fault tolerance?" need to be addressed. Cognitive capabilities essential for achieving fault tolerance through agents are considered. Parallel reduction algorithms are identified as a class of algorithms that can benefit from cognitive agent capabilities. The Message Passing Interface is utilized for implementing an intelligent agent based approach. Preliminary results obtained from the experiments validate the feasibility of an agent based approach for achieving fault tolerance in parallel computing systems.

cs.DC↗