Beyond Immediate Computation: Temporal-Decoupled Computing for Artificial Intelligence Through Historical Computation Reuse, Present Execution, and Future Cost-Aware Scheduling (TDC)
Contemporary artificial intelligence (AI) systems predominantly associate computational execution with the arrival of a user request. Although asynchronous execution, predictive computation, caching, deferred workload scheduling, and energy-aware resource management have challenged aspects of this assumption, computational demand and computational execution are still frequently treated as closely coupled events. This paper introduces Temporal-Decoupled Computation (TDC), a general computational framework that explicitly separates the time at which a computational requirement is expressed from the time and mechanism through which the corresponding computation is satisfied. Under TDC, an AI system is not required to execute every newly received request through entirely new computation at the moment of arrival. Instead, it can autonomously determine whether an appropriate computational result or intermediate state already exists from previous computation, whether immediate execution is preferable, or whether eligible computation should be scheduled for a future period when computational resources, electricity, or infrastructure services are more economically or operationally favorable. The framework establishes three principal temporal pathways: Historical Computation Reuse, Present Computation, and Future Computation. These pathways are integrated through a Temporal Computation Decision Mechanism (TCDM), which evaluates computational requirements, result validity, information freshness, execution deadlines, expected resource availability, electricity prices, energy consumption, carbon intensity, and the potential cost of delayed execution. Unlike conventional workload deferral, TDC considers the relationship between a newly expressed computational requirement and computational work that may have been performed before the requirement arose, as well as work that may be performed after it arises. It therefore treats computational history, present execution, and future execution as interconnected resources within one temporal decision space. This paper makes a conceptual and theoretical contribution by formalizing TDC as a unified framework for AI computational orchestration. It distinguishes TDC from conventional asynchronous computing, predictive computing, semantic caching, and carbon-aware workload scheduling, while identifying their potential integration within the proposed architecture. To the best of our knowledge, this paper is the first to explicitly formulate this particular unified three-temporal-pathway framework as a general AI computation paradigm. The paper develops its conceptual foundations, mathematical formulation, system architecture, operational mechanisms, research propositions, and an experimental methodology for future empirical validation. The framework aims to establish a foundation for AI systems that optimize not only what should be computed, but also whether computation should be reused, performed immediately, or executed at a more advantageous time.
Authors
- Jincheng Zhang (ORCID: https://orcid.org/0009-0005-1860-0009)
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-11-16
- DOI
- https://doi.org/10.5281/zenodo.24270957
- Primary Topic
- Cloud Computing and Resource Management
- Type
- preprint