Searcharxiv⌕ Search

arXiv subjects

Mirco Milletarì

Publications and source records attributed to Mirco Milletarì.

4 recordsLinked to original sources

A Hierarchical Statistical Framework for Evaluating Agent Reliability

Agents operating in complex scenarios often face uncertainty due to incomplete information about the environment they operate in, ambiguous user queries, complex business rules and the inherently stochastic dynamics of the governing policy. Several new benchmarks have been proposed to evaluate LLM-based agents on complex tasks in multi-tool environments, including settings with policy constraints and varying levels of intent ambiguity. These conditions make it important to evaluate agentic systems systematically and quantify their performance under different conditions to assess their reliability in production. This need is evident in the recently introduced Thinkingbox benchmark, where even the strongest models struggle to perform consistently across diverse business scenarios. This paper presents a statistical framework for evaluating uncertainty and reliability in agentic systems. We define a hierarchy of scenarios, instances, and samples to assess model performance and variability systematically. We apply this framework to common metrics found in the literature: mean pass rate, $pass@k$, and $passhat^k$, with uncertainty estimates for each metric. Aggregating metrics at different levels of the hierarchy reveals different structures and properties of the policy model; we connect this hierarchical aggregation with the known process of \emph{coarse-graining} as intended in the physics and statistical literature. We independently derive exact finite-sample expressions for the sampling variances of the unbiased $pass@k$ and $passhatk$ estimators; related results also appear in recent submissions. Our central contribution is a constructive derivation of hierarchical uncertainty propagation, making explicit its correspondence with coarse-graining and separating execution noise from task heterogeneity.

cs.AI↗

Non-divergent Fermi velocity for interacting graphene at the Dirac point

Recent experiments reveal a significant increase in the graphene Fermi velocity close to charge neutrality. This has widely been interpreted as a confirmation of the logarithmic divergence of the graphene Fermi velocity predicted by a perturbative approach. In this work, we reconsider this problem using functional bosonization techniques calculating the effects of electron interactions on the density of states non-perturbatively. We find that the renormalized velocity is {\it finite} and independent of the high energy cut-off, and we argue that the experimental observations are better understood in terms of an anomalous dimension. Our results also represent a bosonized solution for interacting Weyl fermions in (2+1) dimensions at half-filing.

cond-mat.str-el↗

Transient Features in Charge Fractionalization, Local Equilibration and Non-equilibrium Bosonization

In quantum Hall edge states and in other one-dimensional interacting systems, charge fractionalization can occur due to the fact that an injected charge pulse decomposes into eigenmodes propagating at different velocities. If the original charge pulse has some spatial width due to injection with a given source-drain voltage, a finite time is needed until the separation between the fractionalized pulses is larger than their width. In the formalism of non-equilibrium bosonization, the above physics is reflected in the separation of initially overlapping square pulses in the effective scattering phase. When expressing the single particle Green's function as a functional determinant of counting operators containing the scattering phase, the time evolution of charge fractionalization is mathematically described by functional determinants with overlapping pulses. We develop a framework for the evaluation of such determinants, describe the system's equilibration dynamics, and compare our theoretical results with recent experimental findings.

cond-mat.str-el↗

Shot Noise Signatures of Charge Fractionalization in the $ν=2$ Quantum Hall edge

We investigate the effect of interactions on shot noise in $ν=2$ quantum Hall edges, where a repulsive coupling between co-propagating edge modes is expected to give rise to charge fractionalization. Using the method of non-equilibrium bosonization, we find that even asymptotically the edge distribution function depends in a sensitive way on the interaction strength between the edge modes. We compute shot noise and Fano factor from the asymptotic distribution function, and from comparison with a reference model of fractionalized excitations we find that the Fano factor can be close to the value of the fractionalized charge.

cond-mat.mes-hall↗