arXiv subjects

Christian Frey

Publications and source records attributed to Christian Frey.

13 recordsLinked to original sources

arXiv2026

FondueQA: A Benchmark for Evaluating Parametric Compensation in Multi-Hop QA

Large Language Models (LLMs) show remarkable performance based on parametric knowledge, i.e., their ability to recall facts and associations internalized in a model's weights. However, studying this capability on established multi-hop benchmarks may be complicated by benchmark contamination, as many of these benchmarks have been publicly available long enough to plausibly appear in modern pretraining corpora. As a result, correct answers under missing evidence may reflect either general parametric knowledge or memorization of benchmark-specific instances. To study how models use internal knowledge when retrieval evidence is incomplete, we introduce FondueQA, in which every required multi-hop fact is uniquely localized to a designated evidence document. This enables controlled document-level evidence ablations to study parametric compensation under incomplete retrieval. Our empirical evaluation reveals that parametric compensation strongly depends on both entity popularity and the structural role of missing evidence. Generally, models compensate more successfully for facts involving popular entities. On FondueQA, we show that bridging evidence can be more useful than the answer-bearing document itself. Moreover, models may recall a missing fact in isolation but fail to use it under partial evidence, particularly when the context contains plausible competing entities, which can reduce Exact Match by up to 15 percentage points.

cs.LG↗

arXiv2026

Muffin - A Shared Memory Buffer for Asynchronous AI Agents

The rise of Agentic AI systems has changed the landscape of artificial intelligence by enabling strong reasoning capabilities through specialized multi-agent interaction with structured self-reflection. However, in target-driven reasoning tasks, single- and multi-agent systems are often on par in terms of performance gains. A major challenge is that failures often arise from an ineffective exchange of critical information during inter-agent interaction and coordination. Different from existing multi-agent debate paradigms that rely on synchronous turn-taking and sharing information across different agents, we propose an inter-agent protocol based on shared Memory bUFFer INference - referred to as MUFFIN - in an asynchronous agentic framework. In the first step, an orchestrator routes queries based on their reasoning demands and delegates demanding ones to a multi-agent setting, where a shared memory buffer provides asynchronous exchange of intermediate findings while an agent's private reasoning remains local. Our experiments show that our framework Muffin outperforms established competitors, achieving the best accuracy in 17 out of 20 of our benchmark settings, and provides a better quality-cost trade-off than multi-agent debate: Muffin improves accuracy by 19.2 accuracy points while using 77.4% fewer tokens, without requiring full-trace sharing.

cs.AI↗

arXiv2026

MuPACE: Learning to Pace Muon for LoRA Fine-Tuning

Learning rate schedules are typically fixed before training begins, yet when fine-tuning large language models with low-rank adapters (LoRA), the appropriate scale varies substantially with the model, task, and adapter configuration. Practitioners re-tune a schedule for every new problem, yet a tuned schedule still cannot react to the training dynamics it encounters. We instead amortize learning rate selection across LoRA fine-tuning problems. We leave the base optimizer, Muon, untouched, so that its normalized update fixes the direction, and meta-learn only the global step size. Our method MuPACE is a small recurrent policy, meta-trained across a distribution of fine-tuning runs, that conditions on the observed trajectory and emits bounded adjustments to an interpretable analytic schedule. On 64 fine-tuning problems, MuPACE outperforms standard fixed schedules at a strong shared default on 64-78% of problems and is on par with the same schedules with per-task tuned peaks. Meta-trained only on models with up to 16 layers, MuPACE transfers one-shot to unseen tasks on models as deep as 64 layers, such as Qwen3-32B and Qwen3.8-27B, where a single run remains competitive with swept schedules.

cs.LG↗

arXiv2026

Towards Reproducibility in Predictive Process Mining: SPICE -- A Deep Learning Library

In recent years, Predictive Process Mining (PPM) techniques based on artificial neural networks have evolved as a method for monitoring the future behavior of unfolding business processes and predicting Key Performance Indicators (KPIs). However, many PPM approaches often lack reproducibility, transparency in decision making, usability for incorporating novel datasets and benchmarking, making comparisons among different implementations very difficult. In this paper, we propose SPICE, a Python framework that reimplements three popular, existing baseline deep-learning-based methods for PPM in PyTorch, while designing a common base framework with rigorous configurability to enable reproducible and robust comparison of past and future modelling approaches. We compare SPICE to original reported metrics and with fair metrics on 11 datasets.

cs.LG↗

arXiv2026

ValueGround: Evaluating Culture-Conditioned Visual Value Grounding in MLLMs

Cultural values are expressed not only through language but also through visual scenes and everyday social practices. Yet existing evaluations of cultural values in language models are almost entirely text-only, leaving it unclear whether culture-conditioned judgments remain stable when response options are visualized. We introduce ValueGround, a benchmark for evaluating culture-conditioned visual value grounding in multimodal large language models (MLLMs). Built from World Values Survey questions, ValueGround uses minimally contrastive image pairs to represent opposing response options while controlling irrelevant variation. Given a country, a question, and an image pair, a model must choose the image that best matches the country's value tendency without access to the original response-option texts. Experiments across six MLLMs and 13 countries show that models perform substantially worse with visualized response options than with the original textual options, with average accuracy dropping from 72.8% to 62.6%. Our benchmark provides a controlled testbed for studying cross-modal transfer of culture-conditioned value judgments.

cs.CL↗

arXiv2026

POP: Prior-Fitted First-Order Optimization Policies

Gradient-based optimizers are highly sensitive to design choices in their adaptive learning rate mechanisms. To address this limitation, we introduce POP, a meta-learned Reinforcement Learning (RL) policy that predicts adaptive learning rates for gradient descent, conditioned on the contextual information provided in the optimization trajectory. Our method introduces a novel RL reward formulation, a new function-scaling strategy for in-distribution generalization, and a novel prior that is used to sample millions of synthetic optimization problems. We evaluate POP on an established benchmark including 43 optimization functions of various complexity, where it significantly outperforms gradient-based methods. Our evaluation demonstrates strong generalization capabilities without task-specific tuning.

cs.LG↗

arXiv2026

Learning to Order: Task Sequencing as In-Context Optimization

Task sequencing (TS) is one of the core open problems in Deep Learning, arising in a plethora of real-world domains, from robotic assembly lines to autonomous driving. Unfortunately, prior work has not convincingly demonstrated the generalization ability of meta-learned TS methods to solve new TS problems, given few initial demonstrations. In this paper, we demonstrate that deep neural networks can meta-learn over an infinite prior of synthetically generated TS problems and achieve a few-shot generalization. We meta-learn a transformer-based architecture over datasets of sequencing trajectories generated from a prior distribution that samples sequencing problems as paths in directed graphs. In a large-scale experiment, we provide ample empirical evidence that our meta-learned models discover optimal task sequences significantly quicker than non-meta-learned baselines.

cs.LG↗

arXiv2025

Zhyper: Factorized Hypernetworks for Conditioned LLM Fine-Tuning

Large Language Model (LLM) conditioning refers to instructing an LLM to generate content in accordance with the norms and values of a specific culture, beliefs of a particular political orientation, or any desired text-specified semantic conditioning. Unfortunately, prompt engineering does not ensure that LLMs behave in accordance with a desired conditioning due to the inductive bias of the pre-training and alignment datasets. Prior works have focused on fine-tuning LLMs by directly conditioning the LoRA weights; however, such methods introduce a large number of parameters. As a remedy, we propose Zhyper, a parameter-efficient factorized hypernetwork framework that generates context-aware LoRA adapters from textual descriptions. Experiments on multiple benchmarks show that Zhyper achieves competitive performance with up to 26x fewer parameters than the state-of-the-art baselines. Furthermore, we extend Zhyper to cultural alignment, demonstrating improved generalization to out-of-domain settings and a better capturing of fine-grained contextual values.

cs.CL↗

arXiv2023

Fully Coupled Forced Response Analysis of Nonlinear Turbine Blade Vibrations in the Frequency Domain

For the first time, a fully-coupled Harmonic Balance method is developed for the forced response of turbomachinery blades. The method is applied to a state-of-the-art model of a turbine bladed disk with interlocked shrouds subjected to wake-induced loading. The recurrent opening and closing of the pre-loaded shroud contact causes a softening effect, leading to turning points in the amplitude-frequency curve near resonance. Therefore, the coupled solver is embedded into a numerical path continuation framework. Two variants are developed: the coupled continuation of the solution path, and the coupled re-iteration of selected solution points. While the re-iteration variant is slightly more costly per solution point, it has the important advantage that it can be run completely in parallel, which substantially reduces the wall clock time. It is shown that wake- and vibration-induced flow fields do not linearly superimpose, leading to a severe underestimation of the resonant vibration level by the influence-coefficient-based state-of-the-art methods (which rely on this linearity assumption).

cs.CE↗

arXiv2021

Development of a Fully-Coupled Harmonic Balance Method and a Refined Energy Method for the Computation of Flutter-Induced Limit Cycle Oscillations of Bladed Disks with Nonlinear Friction Contacts

Flutter stability is a dominant design constraint of modern gas and steam turbines. To further increase the feasible design space, flutter-tolerant designs are currently explored, which may undergo Limit Cycle Oscillations (LCOs) of acceptable, yet not vanishing, level. Bounded self-excited oscillations are a priori a nonlinear phenomenon, and can thus only be explained by nonlinear interactions such as dry stick-slip friction in mechanical joints. The currently available simulation methods for blade flutter account for nonlinear interactions, at most, in only one domain, the structure or the fluid, and assume the behavior in the other domain as linear. In this work, we develop a fully-coupled nonlinear frequency domain method which is capable of resolving nonlinear flow and structural effects. We demonstrate the computational performance of this method for a state-of-the-art aeroelastic model of a shrouded turbine blade row. Besides simulating limit cycles, we predict, for the first time, the phenomenon of nonlinear instability, i.e., a situation where the equilibrium point is locally stable, but for sufficiently strong perturbation (caused e.g. by an impact), the dry frictional dissipation cannot bound the flutter vibrations. This implies that linearized theory does not necessary lead to a conservative design of turbine blades. We show that this phenomenon is due to the nonlinear contact interactions at the tip shrouds, which cause a change of the vibrational deflection shape and frequency, which in turn leads to a loss of aeroelastic stability. Finally, we extend the well-known energy method to capture these effects, and conclude that it provides a good approximation and is useful for initializing the fully-coupled solver.

cs.CE↗

arXiv2020

Workspace monitoring and planning for safe mobile manipulation

In order to enable physical human-robot interaction where humans and (mobile) manipulators share their workspace and work together, robots have to be equipped with important capabilities to guarantee human safety. The robots have to recognize possible collisions with the human co-worker and react anticipatorily by adapting their motion to avert dangerous situations while they are executing their task. Therefore, methods have been developed that allow to monitor the workspace of mobile manipulators using multiple depth sensors to gather information about the robot environment. This encompasses both 3D information about obstacles in the close robot surroundings and the prediction of obstacle motions in the entire monitored space. Based on this information, a collision-free robot motion is planned and during the execution the robot continuously reacts to unforeseen dangerous situations by adapting its planned motion, slowing down or stopping. For the demonstration of a manufacturing scenario, the developed methods have been implemented on a prototypical mobile manipulator. The algorithms handle both robot platform and manipulator in a uniform manner so that an overall optimization of the path and of the collision avoidance behavior is possible. By integrating the monitoring, planning, and interaction control components, the task of grasping, placing and delivering objects to humans in a shared workspace is demonstrated.

cs.RO↗

arXiv2020

GivEn -- Shape Optimization for Gas Turbines in Volatile Energy Networks

This paper describes the project GivEn that develops a novel multicriteria optimization process for gas turbine blades and vanes using modern "adjoint" shape optimization algorithms. Given the many start and shut-down processes of gas power plants in volatile energy grids, besides optimizing gas turbine geometries for efficiency, the durability understood as minimization of the probability of failure is a design objective of increasing importance. We also describe the underlying coupling structure of the multiphysical simulations and use modern, gradient based multicriteria optimization procedures to enhance the exploration of Pareto-optimal solutions.

math.OC↗

arXiv2018

Efficient Information Flow Maximization in Probabilistic Graphs

Reliable propagation of information through large networks, e.g., communication networks, social networks or sensor networks is very important in many applications concerning marketing, social networks, and wireless sensor networks. However, social ties of friendship may be obsolete, and communication links may fail, inducing the notion of uncertainty in such networks. In this paper, we address the problem of optimizing information propagation in uncertain networks given a constrained budget of edges. We show that this problem requires to solve two NP-hard subproblems: the computation of expected information flow, and the optimal choice of edges. To compute the expected information flow to a source vertex, we propose the F-tree as a specialized data structure, that identifies independent components of the graph for which the information flow can either be computed analytically and efficiently, or for which traditional Monte-Carlo sampling can be applied independently of the remaining network. For the problem of finding the optimal edges, we propose a series of heuristics that exploit properties of this data structure. Our evaluation shows that these heuristics lead to high quality solutions, thus yielding high information flow, while maintaining low running time.

cs.DS↗