Periodic Reporting for period 1 - VM-AI (Virus Machines and Artificial Intelligence)
Okres sprawozdawczy: 2025-04-01 do 2026-06-30
The project will combine ideas from Artificial Intelligence (AI), Spiking Neural P (SNP) systems, and Virus Machines (VMs) that are a branch of Natural Computing that aims in developing unconventional computing paradigms inspired by biological structures and behaviors. The ""mixing"" of the first two seemingly different research fields comes from neurons, both fields have strong connections with modelling of neurons: the AI field is based on the artificial neural networks (ANN) architecture, while the SNP systems are based on the neurons of the brain metaphor, a closer approximation to the spiking activity. Recently, several bridges have been constructed from the spiking neurons with a young computing paradigm called virus machines, which abstracts viral replication and transmission, gaining attention in recent years due to the novelty computing behavior and the higher control of the computation due to an instruction network. We propose virus machines and artificial intelligence to focus attention on domains in which virus machines might offer superior performance over conventional and neural-like systems within artificial intelligence. We consider potential pathways towards virus machines and develop applications in time series prediction and image processing. Moreover, we explore new bridges and discussions between this and other well-established unconventional computing paradigms.
Despite its early conclusion, the project achieved relevant technical and scientific results aligned with its objective of developing virus-inspired approaches for artificial intelligence. A coherent scientific framework was defined, integrating virus machines into AI models to create novel learning systems by means of virus machines.
Novel algorithms and methodologies were developed and implemented in proof-of-concept prototypes to assess feasibility. The first ever learning systems with virus machines has been defined as Echo Virus Machine. Initial experiments and simulations were conducted, supported by dedicated data processing pipelines. The results demonstrated the technical viability of virus-inspired mechanisms to enhance effectivity and consistency in time series prediction, even more if it is chaotic.
Key limitations and boundary conditions were identified such as time efficiency and explicability, providing clear directions for future research. The project delivered reusable technical assets and scientifically validated results, forming a solid foundation for further development despite the reduced project duration.
Novel algorithms and methodologies were developed and implemented in proof-of-concept prototypes to assess feasibility. The first ever learning systems with virus machines has been defined as Echo Virus Machine. Initial experiments and simulations were conducted, supported by dedicated data processing pipelines. The results demonstrated the technical viability of virus-inspired mechanisms to enhance effectivity and consistency in time series prediction, even more if it is chaotic.
Key limitations and boundary conditions were identified such as time efficiency and explicability, providing clear directions for future research. The project delivered reusable technical assets and scientifically validated results, forming a solid foundation for further development despite the reduced project duration.
The project developed Echo Virus Machines (EVMs), a novel AI methodology inspired by both virus-based computational models and echo state neural networks, targeting time-series prediction. Proof-of-concept implementations demonstrated the feasibility of EVMs and their ability to exploit inherent parallelism for temporal processing tasks. In parallel, the theoretical foundations of virus machines were extended, including formal results on their computational power and the definition of new machine variants addressing a wider range of applications.
The results open new research directions in parallel, bio-inspired AI and temporal modelling. Further uptake requires continued theoretical and experimental research, large-scale validation, and exploration of additional application domains to fully assess impact and scalability.
The results open new research directions in parallel, bio-inspired AI and temporal modelling. Further uptake requires continued theoretical and experimental research, large-scale validation, and exploration of additional application domains to fully assess impact and scalability.