arXiv subjects

Manfred Mücke

Publications and source records attributed to Manfred Mücke.

3 recordsLinked to original sources

arXiv2026

Loop-level surrogate modeling of dopant-distribution effects in Ba(Zr,Ti)O$_3$

Barium titanate-based perovskites are important candidates for lead-free dielectric and electromechanical technologies. In Zr-substituted BaTiO$_3$ (BZT), functional behavior is usually discussed in terms of the average Zr concentration, while the influence of dopant spatial distribution beyond average concentration is less understood and difficult to explore systematically. Here we present an accelerated materials-design workflow that links controlled dopant distributions to full field-driven response curves. We generate a broad set of Zr distributions spanning a continuum of nanoscale arrangements, with layers, rods, dots, and lamellae serving as representative end-member motifs, and encode each configuration using a compact, parametrized descriptor model. Effective-Hamiltonian molecular dynamics is used to compute polarization-electric-field and strain-field hysteresis loops, and we train a conditional autoencoder surrogate to predict complete loops directly from the distribution parameters. This surrogate enables rapid screening and dense, property-selective design maps at scales that are not feasible with direct simulations alone, and it supports targeted follow-up simulations in regions of interest. Using the predicted loop database, we screen the distribution space for multiple functional targets, including energy-storage performance, electromechanical response, and switching behavior, and identify the corresponding dopant distribution motif families. The resulting design maps show that dopant distribution is an independent tuning parameter that can strongly affect hysteresis behavior and loop-derived figures of merit: layer-like motifs, vertical lamellae, and nanoplate-like inclusions emerge in different performance regimes. More generally, predicting full response curves enables screening of other loop-derived targets and multi-objective design in substituted ferroelectrics.

cond-mat.mtrl-sci↗

arXiv2022

Compressed Hierarchical Representations for Multi-Task Learning and Task Clustering

In this paper, we frame homogeneous-feature multi-task learning (MTL) as a hierarchical representation learning problem, with one task-agnostic and multiple task-specific latent representations. Drawing inspiration from the information bottleneck principle and assuming an additive independent noise model between the task-agnostic and task-specific latent representations, we limit the information contained in each task-specific representation. It is shown that our resulting representations yield competitive performance for several MTL benchmarks. Furthermore, for certain setups, we show that the trained parameters of the additive noise model are closely related to the similarity of different tasks. This indicates that our approach yields a task-agnostic representation that is disentangled in the sense that its individual dimensions may be interpretable from a task-specific perspective.

cs.LG↗

arXiv2021

Understanding Cache Boundness of ML Operators on ARM Processors

Machine Learning compilers like TVM allow a fast and flexible deployment on embedded CPUs. This enables the use of non-standard operators, which are common in ML compression techniques. However, it is necessary to understand the limitations of typical compute-intense operators in ML workloads to design a proper solution. This is the first in-detail analysis of dense and convolution operators, generated with TVM, that compares to the fundamental hardware limits of embedded ARM processors. Thereby it explains the gap between computational peak performance, theoretical and measured, and real-world state-of-the-art results, created with TVM and openBLAS. Instead, one can see that single-precision general matrix multiply (GEMM) and convolutions are bound by L1-cache-read bandwidth. Explorations of 8-bit and bit-serial quantized operators show that quantization can be used to achieve relevant speedups compared to cache-bound floating-point operators. However, the performance of quantized operators highly depends on the interaction between data layout and bit packing.

cs.AR↗