arXiv subjects

George Pappas

Publications and source records attributed to George Pappas.

At least 19 recordsLinked to original sources

arXiv2026

Strategic Decision Support for AI Agents

Traditionally, decision support studies how humans use machine learning models to make better decisions. In modern agentic systems, this division of roles is increasingly reversed: AI agents act on behalf of users, while humans and tools becomes support mechanisms around them. This role reversal brings reliability concerns to the forefront, since agentic errors can be consequential and agent behavior must remain aligned with human goals and constraints. Departing from the classical view of decision support, we revisit its two basic principles, the cost--value tradeoff of seeking support and the role of uncertainty quantification, in a setting where AI agents are the central actors. We propose a framework for strategic decision support for AI agents through an optimization problem that minimizes support usage subject to controlling a counterfactual missed-support error: the probability that the agent acts alone on instances where support would have materially improved its output. At the population level, we show that the optimal policy is a threshold rule on the value of support. Building on this structure, we develop an online algorithm that adaptively thresholds such a score and uses randomized exploration to control missed-support error without distributional assumptions. We further introduce a calibration-on-the-fly method that reduces unnecessary support calls online. We instantiate this framework across diverse scenarios, including information gathering, human--AI collaboration, and tool use, showing how each can be modeled through the same strategic decision-support lens. Experiments across these settings show that our method reliably controls the target error while substantially reducing support usage in practice.

cs.AI↗

arXiv2026

Conformal Risk-Averse Decision Making with Action Conditional Guarantee

Reliable decision making pipelines powered by machine learning models require uncertainty quantification (UQ) methods that come with explicit safety guarantees. Conformal prediction provides such UQ by wrapping ML predictions into prediction sets, and recent work by Kiyani et al. (2025b) established that these sets can be translated into optimal risk-averse decision policies -- yet only inheriting marginal safety guarantees. We generalize and strengthen their results by (i) introducing action-conditional conformal prediction, which yields safety guarantees conditioned explicitly on each action taken by the decision maker, (ii) showing that action-conditional prediction sets serve as a proxy for the feasible decision space for risk-averse decision makers aiming to optimize action-conditional value-at-risk, and (iii) proposing a principled finite-sample algorithm based on pinball-loss minimization, connecting the framework of Gibbs et al. (2025) to action-conditional guarantees. Experiments on two real-world datasets confirm that our approach significantly improves action-conditional performance over conformal baselines.

stat.ML↗

arXiv2026

InfoSFT: Learn More and Forget Less with Information-Aware Token Weighting

Supervised fine-tuning (SFT) provides the standard approach for teaching LLMs new behaviors from offline expert demonstrations. However, standard SFT uniformly fits all samples -- including those with low likelihood under the base model -- which can disproportionately drive training updates toward overfitting specific samples rather than learning the target behavior. Moreover, adapting to these unlikely samples induces substantial policy shifts that degrade prior capabilities. Existing methods mitigate this by filtering, regenerating, or down-weighting low-likelihood data. In doing so, they often suppress precisely the novel behaviors the base model has yet to learn. We propose InfoSFT, a principled weighting scheme for the SFT objective that concentrates learning signals on maximally informative, medium-confidence tokens -- those neither overly familiar to the base model nor too unlikely to cause instability. Requiring only a one-line modification to the standard token-wise loss, InfoSFT demonstrably improves generalization over vanilla SFT and likelihood-weighted baselines across math, code, and chain-of-thought tasks with diverse model families, while better preserving pre-existing capabilities.

cs.LG↗

arXiv2026

Beyond Binary Success: Sample-Efficient and Statistically Rigorous Robot Policy Comparison

Generalist robot manipulation policies are becoming increasingly capable, but are limited in evaluation to a small number of hardware rollouts. This strong resource constraint in real-world testing necessitates both more informative performance measures and reliable and efficient evaluation procedures to properly assess model capabilities and benchmark progress in the field. This work presents a novel framework for robot policy comparison that is sample-efficient, statistically rigorous, and applicable to a broad set of evaluation metrics used in practice. Based on safe, anytime-valid inference (SAVI), our test procedure is sequential, allowing the evaluator to stop early when sufficient statistical evidence has accumulated to reach a decision at a pre-specified level of confidence. Unlike previous work developed for binary success, our unified approach addresses a wide range of informative metrics: from discrete partial credit task progress to continuous measures of episodic reward or trajectory smoothness, spanning both parametric and nonparametric comparison problems. Through extensive validation on simulated and real-world evaluation data, we demonstrate up to 70% reduction in evaluation burden compared to standard batch methods and up to 50% reduction compared to state-of-the-art sequential procedures designed for binary outcomes, with no loss of statistical rigor. Notably, our empirical results show that competing policies can be separated more quickly when using fine-grained task progress than binary success metrics.

cs.RO↗

arXiv2026

When to Trust the Cheap Check: Weak and Strong Verification for Reasoning

Reasoning with LLMs increasingly unfolds inside a broader verification loop. Internally, systems use cheap checks, such as self-consistency or proxy rewards, which we call weak verification. Externally, users inspect outputs and steer the model through feedback until results are trustworthy, which we call strong verification. These signals differ sharply in cost and reliability: strong verification can establish trust but is resource-intensive, while weak verification is fast and scalable but noisy and imperfect. We formalize this tension through weak--strong verification policies, which decide when to accept or reject based on weak verification and when to defer to strong verification. We introduce metrics capturing incorrect acceptance, incorrect rejection, and strong-verification frequency. Over population, we show that optimal policies admit a two-threshold structure and that calibration and sharpness govern the value of weak verifiers. Building on this, we develop an online algorithm that provably controls acceptance and rejection errors without assumptions on the query stream, the language model, or the weak verifier.

cs.LG↗

arXiv2026

Multi-Round Human-AI Collaboration with User-Specified Requirements

As humans increasingly rely on multiround conversational AI for high stakes decisions, principled frameworks are needed to ensure such interactions reliably improve decision quality. We adopt a human centric view governed by two principles: counterfactual harm, ensuring the AI does not undermine human strengths, and complementarity, ensuring it adds value where the human is prone to err. We formalize these concepts via user defined rules, allowing users to specify exactly what harm and complementarity mean for their specific task. We then introduce an online, distribution free algorithm with finite sample guarantees that enforces the user-specified constraints over the collaboration dynamics. We evaluate our framework across two interactive settings: LLM simulated collaboration on a medical diagnostic task and a human crowdsourcing study on a pictorial reasoning task. We show that our online procedure maintains prescribed counterfactual harm and complementarity violation rates even under nonstationary interaction dynamics. Moreover, tightening or loosening these constraints produces predictable shifts in downstream human accuracy, confirming that the two principles serve as practical levers for steering multi-round collaboration toward better decision quality without the need to model or constrain human behavior.

cs.LG↗

arXiv2026

Robust Policy Optimization to Prevent Catastrophic Forgetting

Large language models are commonly trained through multi-stage post-training: first via RLHF, then fine-tuned for other downstream objectives. Yet even small downstream updates can compromise earlier learned behaviors (e.g., safety), exposing a brittleness known as catastrophic forgetting. This suggests standard RLHF objectives do not guarantee robustness to future adaptation. To address it, most prior work designs downstream-time methods to preserve previously learned behaviors. We argue that preventing this requires pre-finetuning robustness: the base policy should avoid brittle high-reward solutions whose reward drops sharply under standard fine-tuning. We propose Fine-tuning Robust Policy Optimization (FRPO), a robust RLHF framework that optimizes reward not only at the current policy, but across a KL-bounded neighborhood of policies reachable by downstream adaptation. The key idea is to ensure reward stability under policy shifts via a max-min formulation. By modifying GRPO, we develop an algorithm with no extra computation, and empirically show it substantially reduces safety degradation across multiple base models and downstream fine-tuning regimes (SFT and RL) while preserving downstream task performance. We further study a math-focused RL setting, demonstrating that FRPO preserves accuracy under subsequent fine-tuning.

cs.LG↗

arXiv2026

eCP: Equivariant Conformal Prediction with pre-trained models

Conformal prediction, a post-hoc, distribution-free, finite-sample method of uncertainty quantification that offers formal coverage guarantees under the assumption of data exchangeability. Unfortunately, the resulting uncertainty regions can grow significantly in long horizon missions, rendering the statistical guarantees uninformative. To that end, we propose infusing CP with geometric information via group-averaging of the pretrained predictor to distribute the non-conformity mass across the orbits. Each sample now is treated as a representative of an orbit, thus uncertainty can be mitigated by other samples entangled to it via the orbit inducing elements of the symmetry group. Our approach provably yields contracted non-conformity scores in increasing convex order, implying improved exponential-tail bounds and sharper conformal prediction sets in expectation, especially at high confidence levels. We then propose an experimental design to test these theoretical claims in pedestrian trajectory prediction.

cs.LG↗

arXiv2026

On transferring safety certificates across dynamical systems

Control barrier functions (CBFs) provide a powerful tool for enforcing safety constraints in control systems, but their direct application to complex, high-dimensional dynamics is often challenging. In many settings, safety certificates are more naturally designed for simplified or alternative system models that do not exactly match the dynamics of interest. This paper addresses the problem of transferring safety guarantees between dynamical systems with mismatched dynamics. We propose a transferred control barrier function (tCBF) framework that enables safety constraints defined on one system to be systematically enforced on another system using a simulation function and an explicit margin term. The resulting transferred barrier accounts for model mismatch and induces a safety condition that can be enforced on the target system via a quadratic-program-based safety filter. The proposed approach is general and does not require the two systems to share the same state dimension or dynamics. We demonstrate the effectiveness of the framework on a quadrotor navigation task with the transferred barrier ensuring collision avoidance for the target system, while remaining minimally invasive to a nominal controller. These results highlight the potential of transferred control barrier functions as a general mechanism for enforcing safety across heterogeneous dynamical systems.

eess.SY↗

arXiv2025

Multipole moments do not uniquely characterize spacetimes beyond general relativity

Spacetimes in general relativity can be uniquely decomposed into a set of multipole moments. Given the usefulness of moments in the categorization of radiation patterns, tidal deformations, and other phenomena associated with compact objects, a number of studies have explored their construction in beyond-Einstein theories of gravity. It is shown here that uniqueness does not necessarily extend across theories: by comparing a few static and spherically-symmetric solutions in different theories, we find that two distinct objects can possess the same Geroch-Hansen moments. Moreover, two metrics can match and yet take different moments. Implications of this result are explored in the context of black-hole shadows and ``universal'' relations hinging on moment computations.

gr-qc↗

arXiv2025

Verification of Sequential Convex Programming for Parametric Non-convex Optimization

We introduce a verification framework to exactly verify the worst-case performance of sequential convex programming (SCP) algorithms for parametric non-convex optimization. The verification problem is formulated as an optimization problem that maximizes a performance metric (e.g., the suboptimality after a given number of iterations) over parameters constrained to be in a parameter set and iterate sequences consistent with the SCP update rules. Our framework is general, extending the notion of SCP to include both conventional variants such as trust-region, convex-concave, and prox-linear methods, and algorithms that combine convex subproblems with rounding steps, as in relaxing and rounding schemes. Unlike existing analyses that may only provide local guarantees under limited conditions, our framework delivers global worst-case guarantees--quantifying how well an SCP algorithm performs across all problem instances in the specified family. Applications in control, signal processing, and operations research demonstrate that our framework provides, for the first time, global worst-case guarantees for SCP algorithms in the parametric setting.

math.OC↗

arXiv2025

Robust Decision Making with Partially Calibrated Forecasts

Calibration has emerged as a foundational goal in ``trustworthy machine learning'', in part because of its strong decision theoretic semantics. Independent of the underlying distribution, and independent of the decision maker's utility function, calibration promises that amongst all policies mapping predictions to actions, the uniformly best policy is the one that ``trusts the predictions'' and acts as if they were correct. But this is true only of \emph{fully calibrated} forecasts, which are tractable to guarantee only for very low dimensional prediction problems. For higher dimensional prediction problems (e.g. when outcomes are multiclass), weaker forms of calibration have been studied that lack these decision theoretic properties. In this paper we study how a conservative decision maker should map predictions endowed with these weaker (``partial'') calibration guarantees to actions, in a way that is robust in a minimax sense: i.e. to maximize their expected utility in the worst case over distributions consistent with the calibration guarantees. We characterize their minimax optimal decision rule via a duality argument, and show that surprisingly, ``trusting the predictions and acting accordingly'' is recovered in this minimax sense by \emph{decision calibration} (and any strictly stronger notion of calibration), a substantially weaker and more tractable condition than full calibration. For calibration guarantees that fall short of decision calibration, the minimax optimal decision rule is still efficiently computable, and we provide an empirical evaluation of a natural one that applies to any regression model solved to optimize squared error.

stat.ML↗

arXiv2025

Human-AI Collaborative Uncertainty Quantification

AI predictive systems are increasingly embedded in decision making pipelines, shaping high stakes choices once made solely by humans. Yet robust decisions under uncertainty still rely on capabilities that current AI lacks: domain knowledge not captured by data, long horizon context, and reasoning grounded in the physical world. This gap has motivated growing efforts to design collaborative frameworks that combine the complementary strengths of humans and AI. This work advances this vision by identifying the fundamental principles of Human AI collaboration within uncertainty quantification, a key component of reliable decision making. We introduce Human AI Collaborative Uncertainty Quantification, a framework that formalizes how an AI model can refine a human expert's proposed prediction set with two goals: avoiding counterfactual harm, ensuring the AI does not degrade correct human judgments, and complementarity, enabling recovery of correct outcomes the human missed. At the population level, we show that the optimal collaborative prediction set follows an intuitive two threshold structure over a single score function, extending a classical result in conformal prediction. Building on this insight, we develop practical offline and online calibration algorithms with provable distribution free finite sample guarantees. The online method adapts to distribution shifts, including human behavior evolving through interaction with AI, a phenomenon we call Human to AI Adaptation. Experiments across image classification, regression, and text based medical decision making show that collaborative prediction sets consistently outperform either agent alone, achieving higher coverage and smaller set sizes across various conditions.

cs.AI↗

arXiv2025

An Einstein ring fingerprint around SMBHs illuminated by BLR spectral lines

The continuum emission from the hot and ionized inner regions of a supermassive black hole (SMBH) accretion disk that is strongly lensed by the light-ring (i.e., the unstable photon orbit), is always superimposed on that of the locally emitting plasma near the innermost stable circular orbit (ISCO), masking strong-gravity effects and making their study difficult. A cleaner image of the light-ring against a non-luminous background, not affected by the emission and all the dynamical effects taking place near the ISCO, would thus be preferable. A more distant SMBH illumination source that could accommodate this can be provided by the unique spectral lines emitted by the cooler parts of the accretion disk, the so-called Broad Line Region (BLR). Spectral line emission from the transitional region between the inner disk and the outer BLR may be ideal for obtaining a cleaner image of the light-ring, and better suited for strong gravity tests. A crucial first order effect of a BLR spectral line illumination of the SMBHs in galactic centers, and a possible smoking gun signal of strong SMBH lensing, will be an Einstein ring, whose presence could be evident even in unresolved sources where only the total line profile can be acquired. In resolved sources, the combination of information from the image and the spectrum can even facilitate the measurement of the mass of the SMBH.

gr-qc↗

arXiv2025

SMBH shadows: gravity fingerprints revealed by spectral line background radiation

SMBHs at the centers of galaxies, illuminated by their accretion disks, can reveal valuable information on BH spacetimes via their shadows, which can be used for strong-gravity tests. However, the continuum emission from their highly turbulent hot plasma is expected to be strongly time-varying and with very inhomogeneous brightness. This can mask important SMBH-related effects, like the appearance of the light-ring, rendering them ineffective as probes of strong gravitational lensing physics. Besides being an inhomogeneous and strongly time-varying ''illuminator'', the hot plasma emission extends all the way to the ISCO. This then leads to the superposition of the strongly-lensed radiation from the area of the light-ring to the continuum emission from the ISCO, effectively making gravitational lensing physics hard to separate from accretion disk Astrophysics. These problems could be overcome if one utilizes the spectral line radiation field emanating from the cooler parts of the extended accretion disk, the so-called BLR, and especially its expected neutral phase, as a more distant, but still adequately strong SMBH illuminator, typically found at $r\sim (10^2-10^4)\, R_s$. This kind of illumination can provide a cleaner image of the light-ring region, and thus allow more information on the spacetime geometry around the SMBH to be obtained. Here, we examine some of the benefits of such an illumination in discerning strong-gravity physics near SMBHs and their observability. We expand on the fact that such emission can provide a smoking gun signal of lensing, in the form of an Einstein ring. To first order, the imaging of the Einstein ring and its spectroscopic signature can facilitate the measurement of the SMBH mass, while the second order effects associated with the light-ring can constrain the SMBH spin, and even identify deviations from the Kerr spacetime.

gr-qc↗

arXiv2025

Learning Acceleration Algorithms for Fast Parametric Convex Optimization with Certified Robustness

We develop a machine-learning framework to learn hyperparameter sequences for accelerated first-order methods (e.g., the step size and momentum sequences in accelerated gradient descent) to quickly solve parametric convex optimization problems with certified robustness. We obtain a strong form of robustness guarantee -- certification of worst-case performance over all parameters within a set after a given number of iterations -- through regularization-based training. The regularization term is derived from the performance estimation problem (PEP) framework based on semidefinite programming, in which the hyperparameters appear as problem data. We show how to use gradient-based training to learn the hyperparameters for several first-order methods: accelerated versions of gradient descent, proximal gradient descent, and alternating direction method of multipliers. Through various numerical examples from signal processing, control, and statistics, we demonstrate that the quality of the solution can be dramatically improved within a budget of iterations, while also maintaining strong robustness guarantees. Notably, our approach is highly data-efficient in that we only use ten training instances in all of the numerical examples.

math.OC↗

arXiv2025

Deep Equivariant Multi-Agent Control Barrier Functions

With multi-agent systems increasingly deployed autonomously at scale in complex environments, ensuring safety of the data-driven policies is critical. Control Barrier Functions have emerged as an effective tool for enforcing safety constraints, yet existing learning-based methods often lack in scalability, generalization and sampling efficiency as they overlook inherent geometric structures of the system. To address this gap, we introduce symmetries-infused distributed Control Barrier Functions, enforcing the satisfaction of intrinsic symmetries on learnable graph-based safety certificates. We theoretically motivate the need for equivariant parametrization of CBFs and policies, and propose a simple, yet efficient and adaptable methodology for constructing such equivariant group-modular networks via the compatible group actions. This approach encodes safety constraints in a distributed data-efficient manner, enabling zero-shot generalization to larger and denser swarms. Through extensive simulations on multi-robot navigation tasks, we demonstrate that our method outperforms state-of-the-art baselines in terms of safety, scalability, and task success rates, highlighting the importance of embedding symmetries in safe distributed neural policies.

eess.SY↗

arXiv2025

Conformal Prediction Beyond the Seen: A Missing Mass Perspective for Uncertainty Quantification in Generative Models

Uncertainty quantification (UQ) is essential for safe deployment of generative AI models such as large language models (LLMs), especially in high stakes applications. Conformal prediction (CP) offers a principled uncertainty quantification framework, but classical methods focus on regression and classification, relying on geometric distances or softmax scores: tools that presuppose structured outputs. We depart from this paradigm by studying CP in a query only setting, where prediction sets must be constructed solely from finite queries to a black box generative model, introducing a new trade off between coverage, test time query budget, and informativeness. We introduce Conformal Prediction with Query Oracle (CPQ), a framework characterizing the optimal interplay between these objectives. Our finite sample algorithm is built on two core principles: one governs the optimal query policy, and the other defines the optimal mapping from queried samples to prediction sets. Remarkably, both are rooted in the classical missing mass problem in statistics. Specifically, the optimal query policy depends on the rate of decay, or the derivative, of the missing mass, for which we develop a novel estimator. Meanwhile, the optimal mapping hinges on the missing mass itself, which we estimate using Good Turing estimators. We then turn our focus to implementing our method for language models, where outputs are vast, variable, and often under specified. Fine grained experiments on three real world open ended tasks and two LLMs, show CPQ applicability to any black box LLM and highlight: (1) individual contribution of each principle to CPQ performance, and (2) CPQ ability to yield significantly more informative prediction sets than existing conformal methods for language uncertainty quantification.

cs.LG↗