Prasanna Balachandran

arXiv2026

Are LLMs Ready for Neural-integrated Mechanistic Modeling? A Benchmark and Agentic Framework

Large language models (LLMs) have shown promise in constructing mechanistic models from data. However, existing evaluations largely focus on simplified settings and fail to capture the complexity of real-world scientific modeling. In practice, such modeling often involves neural-integrated formulations, where a mechanistic model component and a neural network component are jointly constructed, leading to a significantly more complex search space. Motivated by this gap, we introduce the Neural-Integrated Mechanistic Modeling (NIMM) benchmark, which evaluates LLM-generated neural-integrated mechanistic models across three scientific domains. Experiments on NIMM reveal that existing LLM-based approaches struggle to effectively explore this complex space, resulting in limited search stability and solution quality. To address this challenge, we propose NIMMGen, a tree-guided agentic framework that enables diversified exploration via branch-level search and improves solutions through atomic model refinement. Extensive experiments demonstrate that NIMMGen achieves state-of-the-art performance on NIMM, significantly improving search stability and solution quality.

cs.LG↗

arXiv2021

Computing and Memory Technologies based on Magnetic Skyrmions

Solitonic magnetic excitations such as domain walls and, specifically, skyrmionics enable the possibility of compact, high density, ultrafast,all-electronic, low-energy devices, which is the basis for the emerging area of skyrmionics. The topological winding of skyrmion spins affects their overall lifetime, energetics and dynamical behavior. In this review, we discuss skyrmionics in the context of the present day solid state memory landscape, and show how their size, stability and mobility can be controlled by material engineering, as well as how they can be nucleated and detected. Ferrimagnetsnear their compensation points are important candidates for this application, leading to detailed exploration of amorphous CoGd as well as the study of emergent materials such as Mn$_4$N and Inverse Heusler alloys. Along with material properties, geometrical parameters such as film thickness, defect density and notches can be used to tune skyrmion properties, such as their size and stability. Topology, however, can be a double-edged sword, especially for isolated metastable skyrmions, as it brings stability at the cost of additional damping and deflective Magnus forces compared to domain walls. Skyrmion deformation in response to forces also makes them intrinsically slower than domain walls. We explore potential analog applications of skyrmions, including temporal memory at low density, and decorrelator for stochastic computing at a higher density that capitalizes on their interactions. We summarize the main challenges to achieve a skyrmionics technology, including maintaining positional stability with very high accuracy, electrical readout, especially for small ferrimagnetic skyrmions, deterministic nucleation and annihilation, and overall integration with digital circuits with the associated circuit overhead.

cond-mat.mtrl-sci↗