Once-for-All: Scalable Simultaneous Forecasting via Equilibrium State Estimation
Beinan Xu, Andy Song, Jiti Gao, Feng Liu
Abstract
We introduce Equilibrium State Estimation (ESE), a novel paradigm for simultaneous prediction, where multiple interacting systems require separate yet coordinated forecasts. Such scenarios often arise in real-world settings such as economics and healthcare modeling. Unlike existing approaches that predict one system at a time, ESE forecasts all systems in a single pass. It first estimates the equilibrium state across systems, then generates holistic forecasts based on the difference between the current state and the estimated equilibrium. Extensive experiments on synthetic and real-world datasets, including currency exchange and COVID-19 spread modeling, demonstrate that ESE is at least as accurate as state-of-the-art (SOTA) methods while being significantly faster. In addition, ESE integrates seamlessly with conventional predictors, combining their accuracy with its exceptional efficiency and delivering a 10-70x speedup. With linear-time complexity, ESE scales far better than SOTA methods as the number of systems increases. Moreover, it remains accurate under diverse perturbations, establishing ESE as a fast, generalizable, robust, and scalable multi-prediction method.
AI Impact Assessments
(1 models)Scientific Impact Assessment: "Once-for-All: Scalable Simultaneous Forecasting via Equilibrium State Estimation"
1. Core Contribution
The paper introduces Equilibrium State Estimation (ESE), a paradigm for simultaneously forecasting multiple interacting systems by first estimating a collective equilibrium state, then predicting individual system trajectories based on deviations from that equilibrium. The key insight is decomposing the forecasting problem into two parts: (1) estimating an aggregate trend for the entire ensemble, and (2) distributing that aggregate across individual systems using attribute-informed equilibrium proportions. This avoids the need to train separate models per system or to handle high-dimensional multivariate outputs directly.
The approach draws conceptual inspiration from Nash equilibrium but operationalizes a statistical notion of equilibrium via cointegration testing. The method iteratively adjusts equilibrium estimates until they achieve a statistically significant long-run relationship with historical data (p < 0.05 via cointegration test).
2. Methodological Rigor
Strengths in methodology:
Concerns:
3. Potential Impact
Practical applications: The paper demonstrates ESE on two meaningful real-world domains—currency exchange rates (16 G20 currencies) and COVID-19 spread (up to 320 regions). The 10-70× speedup when integrated with SOTA methods is practically significant for operational forecasting systems.
Scalability advantage: The linear-time complexity (O(n·m·p) for n systems, m attributes, p time steps) is a genuine advantage over methods that scale quadratically or worse. Figure 3 convincingly demonstrates this linear scaling. For large-scale applications (hundreds of regions/systems), this is a meaningful contribution.
Integration capability: ESE's ability to wrap around existing forecasting methods (Eq. 8) is perhaps its most impactful feature. By reducing multi-system forecasting to single aggregate forecasting + proportional allocation, it enables any univariate forecaster to handle multi-system scenarios efficiently.
4. Timeliness & Relevance
The paper addresses a real gap in the forecasting literature. Multi-system forecasting scenarios are common (regional epidemics, financial markets, supply chains) but underserved by existing methods that either treat each system independently or model all variables jointly. The approach fills a practical niche between these extremes.
However, the timing relative to foundation models for time series (Time-MoE, AutoTimes) raises questions about whether the efficiency gains will remain relevant as large-scale pretrained models become more capable at zero-shot multi-target forecasting.
5. Strengths & Limitations
Key Strengths:
Notable Weaknesses:
6. Additional Observations
The paper is exceptionally thorough in its appendices, providing proofs, additional analyses, and extensive ablations. However, the main paper's clarity could benefit from a more concise presentation of the core algorithm. The connection to Nash equilibrium, while motivating, is somewhat loose—the actual mechanism is closer to constrained proportional allocation than game-theoretic equilibrium computation.
The paper would benefit from explicit confidence intervals or significance tests on the performance differences, as many improvements appear within noise margins. The computational cost comparisons are more convincingly significant than the accuracy comparisons.
Generated Jun 12, 2026
Comparison History (20)
Paper 2 introduces a fundamentally new paradigm (ESE) for simultaneous multi-system forecasting with strong theoretical grounding (equilibrium estimation), demonstrated 10-70x speedups, linear-time complexity, and broad applicability across diverse domains (economics, epidemiology). It addresses a clearly defined gap—scalable coordinated prediction—with rigorous experiments on both synthetic and real-world data. Paper 1 offers an incremental improvement to classification loss functions with benefits mainly in low-data regimes. Paper 2's broader cross-domain applicability, scalability advantages, and novel conceptual framework give it higher potential impact.
Paper 2 is more likely to have higher scientific impact: it introduces a broadly applicable, theoretically grounded metric (TQS) linking quantization to dynamical-systems stability, enabling a priori sensitivity estimation decoupled from specific PTQ choices and even usable for black-box/compiled models. This directly targets timely deployment constraints (edge/resource-limited inference) across many time-series and sequential models, with potential spillover to control and stability analysis. Paper 1’s ESE is innovative and useful for multi-system forecasting, but its impact is more domain-specific and depends on strong assumptions about equilibrium estimation.
Paper 1 likely has higher impact due to a clearly novel, scalable paradigm (Equilibrium State Estimation) that enables simultaneous multi-system forecasting with strong claimed efficiency gains (10–70×, linear-time) and broad applicability (economics, epidemiology, other interacting dynamical systems). If validated, this combines methodological innovation with immediate real-world utility and cross-domain relevance. Paper 2 addresses an important, timely problem (continual learning) but the contribution appears more incremental/phenomenological around replay timing and consolidation, with less clearly specified algorithmic novelty or demonstrated breadth compared to Paper 1’s general-purpose, efficiency-driven framework.
While Paper 1 offers impressive scalability and speedups for practical forecasting, Paper 2 tackles a fundamental bottleneck in scientific discovery: extracting symbolic governing equations from high-dimensional, noisy data. By bridging representation learning with symbolic regression and providing theoretical identifiability guarantees, Paper 2 has a profound potential impact across physics, neuroscience, and biology, making it more transformative for fundamental science.
Paper 1 introduces a fundamentally new paradigm (ESE) for simultaneous multi-system forecasting with broad applicability across economics, healthcare, and other domains. Its linear-time complexity, 10-70x speedup over SOTA, and seamless integration with existing predictors make it highly practical and widely adoptable. Paper 2 makes a solid contribution to ecological monitoring with PULSE, but its scope is narrower—focused specifically on Orthoptera bioacoustics. While valuable for biodiversity research, Paper 1's methodological generality, scalability advantages, and cross-domain relevance give it higher potential for broad scientific impact.
While Paper 1 presents a highly efficient and novel algorithmic approach to simultaneous forecasting, Paper 2 details a foundational IEEE standard for machine learning arithmetic. Establishing universally adopted low-precision hardware formats has a profound, industry-wide impact, directly influencing the design of future AI accelerators and the execution of virtually all large-scale ML models.
Paper 1 likely has higher scientific impact due to a more distinct, scalable paradigm (Equilibrium State Estimation) addressing simultaneous forecasting across interacting systems with clear efficiency gains (10–70x speedup, linear-time scaling) and demonstrated applicability to high-stakes domains (economics, epidemiology). Its “plug-in” compatibility with conventional predictors increases adoption potential and cross-domain reach. Paper 2 extends a well-developed area (unsupervised GNN-based graph clustering) with iterative self-learning/context graphs; impact may be narrower and results appear more conditional (notably when clusters are balanced), reducing broad, immediate real-world leverage.
Paper 1 introduces a novel paradigm (ESE) for simultaneous multi-system forecasting with strong theoretical foundations, demonstrated scalability (linear-time complexity), significant speedups (10-70x), and broad applicability across economics and healthcare. It addresses a fundamental challenge in multi-system prediction with rigorous methodology and extensive experiments. Paper 2 is a narrower application study comparing GANs for Mars DEM segmentation with a negative result (GAN augmentation didn't help), limited novelty, and narrower impact scope.
Paper 2 likely has higher impact: it proposes a broadly applicable paradigm (Equilibrium State Estimation) for simultaneous forecasting across many interacting systems, with strong real-world relevance (economics, healthcare/COVID-19) and clear scalability gains (linear time, 10–70× speedup) while matching SOTA accuracy. Its applicability spans multiple domains and addresses a timely need for efficient multi-system prediction. Paper 1 is a solid, novel improvement for neural operators, but is more specialized to PDE/operator learning and offers a more incremental architectural refinement.
Paper 2 addresses a fundamental question in learning theory—generalizing beyond independence assumptions—by introducing the novel framework of simulatable processes. It provides deep theoretical contributions (recovering VC-dimension-based guarantees under dependent data, proving computational/statistical separations, and connecting to Kolmogorov complexity), which broadly impact learning theory, computational complexity, and the foundations of machine learning. Paper 1 offers a useful engineering contribution for multi-system forecasting with practical speedups, but its conceptual novelty and breadth of theoretical impact are more limited compared to Paper 2's foundational advances.