aiqbench analysis
This paper gives a theoretical account of when repeated retraining stays stable in multi-agent settings where each deployed model shifts the data future models train on. It replaces the usual single scalar contraction constant with a "reflexivity matrix" and shows that the spectral radius crossing one separates convergence to a stable equilibrium from linear excess loss. The results are proven bounds rather than empirical benchmarks, so their practical value depends on how well the per-agent sensitivity and curvature parameters can be estimated in real systems.
The Reflexivity Threshold: A Phase Transition for Multi-Agent Performative Prediction
arXiv:2610.29577v1 Announce Type: new Abstract: Many learning systems are performative: once deployed, their predictions change the data distribution on which future models are trained. In multi-agent settings, this feedback is routed through a network of interacting learners, so stability depends not only on the strength of performativity but also on who affects whom. We study repeated retraining in multi-agent performative prediction and identify a spectral threshold that separates stable learning from worst-case learning obstruction. Prior work often collapses cross-agent feedback into a scalar contraction constant, missing this network structure. We instead introduce the reflexivity matrix $\Gamma$, which records how strongly one agent's deployment can move another agent's best response, defined from per-agent gradient sensitivities and curvature parameters. The threshold occurs when the spectral radius of this matrix crosses one. Below the threshold, repeated retraining converges geometrically to a unique performatively stable equilibrium, with cumulative excess loss bounded uniformly in the time horizon. Above the threshold, we construct noisy supercritical instances in which every local-information learner incurs expected cumulative excess loss growing linearly for horizons up to order $1/(\rho(\Gamma)-1)$. Together, these results characterize the threshold behavior through upper and lower bounds and recover the single-agent scalar threshold from prior work as a special case.
Abstract:Many learning systems are performative: once deployed, their predictions change the data distribution on which future models are trained. In multi-agent settings, this feedback is routed through a network of interacting learners, so stability depends not only on the strength of performativity but also on who affects whom. We study repeated retraining in multi-agent performative prediction and identify a spectral threshold that separates stable learning from worst-case learning obstruction. Prior work often collapses cross-agent feedback into a scalar contraction constant, missing this network structure. We instead introduce the reflexivity matrix $\Gamma$, which records how strongly one agent's deployment can move another agent's best response, defined from per-agent gradient sensitivities and curvature parameters. The threshold occurs when the spectral radius of this matrix crosses one. Below the threshold, repeated retraining converges geometrically to a unique performatively stable equilibrium, with cumulative excess loss bounded uniformly in the time horizon. Above the threshold, we construct noisy supercritical instances in which every local-information learner incurs expected cumulative excess loss growing linearly for horizons up to order $1/(\rho(\Gamma)-1)$. Together, these results characterize the threshold behavior through upper and lower bounds and recover the single-agent scalar threshold from prior work as a special case.
Accepted at NeurIPS 2026. 20 pages, including appendix and checklist. Code: this https URL
Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
arXiv:2610.29577 [cs.LG]
(or arXiv:2610.29577v1 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2610.29577 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 309 words
Related coverage
The value of information and the emergence of communication in evolved e-puck robots
arXiv · Oct 12, 2026
Honesty in Causal Forests: When It Helps and When It Hurts
arXiv · Jul 20, 2026
Honesty in Causal Forests: When It Helps and When It Hurts
arXiv · Jun 3, 2026
Learning Reflexive Behavior for Contact-Rich Manipulation
arXiv · Oct 5, 2026