arXiv subjects

András Balogh

Publications and source records attributed to András Balogh.

5 recordsLinked to original sources

arXiv2026

One-way electron mirrors from single-layer stacking shifts in rhombohedral graphite

One-way mirrors or asymmetric scatterers are key components of photonic systems, protecting sources from back-reflections and routing light along a chosen path. Acoustic systems can also be engineered to show asymmetric reflection. However, a condensed matter platform acting as a one-way mirror without breaking linearity, time-reversal symmetry or energy conservation has been lacking. One route to achieving this is mode selection, where a probe coupled to only some modes sees a boundary as a mirror from one side and as transparent from the other. Here we show that single-layer stacking shifts in six-layer rhombohedral graphite provide this platform, acting as one-way mirrors for electrons. In scanning tunneling microscopy, 10 nm scale interference fringes appear only on the stacking fault side of a wall, tracking its Dirac band, the rhombohedral side remaining transparent. Layer-resolved tight-binding and wave-packet dynamics calculations reproduce this and trace the asymmetry to the mismatched layer polarization across the wall, with the tip supplying the selective coupling. Because stacking solitons glide under external fields, this mirror forms a movable element of electron optics defined by the crystal structure alone.

cond-mat.mes-hall↗

arXiv2026

ATAC: Augmentation-Based Test-Time Adversarial Correction for CLIP

Despite its remarkable success in zero-shot image-text matching, CLIP remains highly vulnerable to adversarial perturbations on images. As adversarial fine-tuning is prohibitively costly, recent works explore various test-time defense strategies; however, these approaches still exhibit limited robustness. In this work, we revisit this problem and propose a simple yet effective strategy: Augmentation-based Test-time Adversarial Correction (ATAC). Our method operates directly in the embedding space of CLIP, calculating augmentation-induced drift vectors to infer a semantic recovery direction and correcting the embedding based on the angular consistency of these latent drifts. Across a wide range of benchmarks, ATAC consistently achieves remarkably high robustness, surpassing that of previous state-of-the-art methods by nearly 50\% on average, all while requiring minimal computational overhead. Furthermore, ATAC retains state-of-the-art robustness in unconventional and extreme settings and even achieves nontrivial robustness against adaptive attacks. Our results demonstrate that ATAC is an efficient method in a novel paradigm for test-time adversarial defenses in the embedding space of CLIP. Code is available at: https://github.com/kylin0421/ATAC

cs.CV↗

arXiv2026

The role of stacking and strain in mean-field magnetic moments of multilayer graphene

Rhombohedral or ABC stacked multilayer graphene hosts a correlated magnetic ground state at charge neutrality, making it one of the simplest systems to investigate strong electronic correlations. We investigate this ground state in multilayer graphene structures using the Hubbard model in a distance dependent Slater-Koster tight binding framework. We show that by using a universal Hubbard-$U$ term, we can accurately capture the spin polarization predicted by hybrid density functional theory calculations for both hexagonal (ABA) and rhombohedral (ABC) stackings. Using this $U$ value, we calculate the magnetic moments of 3-8 layers of ABC and ABA graphene multilayers. We demonstrate that the structure and magnitude of these magnetic moments are robust when heterostructures are built from varying numbers of ABC and ABA multilayers. By applying different types of mechanical distortions, we study the behaviour of the magnetism in graphene systems under uniaxial strain and pressure. Our results establish a computationally efficient framework to investigate correlation-driven magnetism across arbitrary stacking configurations of graphite polytypes.

cond-mat.str-el↗

arXiv2026

Verification of the Implicit World Model in a Generative Model via Adversarial Sequences

Generative sequence models are typically trained on sample sequences from natural or formal languages. It is a crucial question whether -- or to what extent -- sample-based training is able to capture the true structure of these languages, often referred to as the ``world model''. Theoretical results indicate that we can hope for soundness at best, that is, generating valid sequences, but not necessarily all of them. However, it is still important to have practical tools that are able to verify whether a given sequence model is sound. In this study, we focus on chess, as it is a domain that provides enough complexity while having a simple rule-based world model. We propose adversarial sequence generation for verifying the soundness of the sequence model. Our adversaries generate valid sequences so as to force the sequence model to generate an invalid next move prediction. Apart from the falsification of soundness, this method is also suitable for a more fine-grained analysis of the failure modes and the effects of different choices during training. To demonstrate this, we propose a number of methods for adversarial sequence generation and evaluate the approach on a large set of chess models. We train models on random as well as high-quality chess games, using several training recipes. We find that none of the models are sound, but some training techniques and dataset choices are able to improve soundness remarkably. We also investigate the potential application of board state probes in both our training and attack methods. Our findings indicate that the extracted board states have no causal role in next token prediction in most of the models.

cs.LG↗

arXiv2024

How not to Stitch Representations to Measure Similarity: Task Loss Matching versus Direct Matching

Measuring the similarity of the internal representations of deep neural networks is an important and challenging problem. Model stitching has been proposed as a possible approach, where two half-networks are connected by mapping the output of the first half-network to the input of the second one. The representations are considered functionally similar if the resulting stitched network achieves good task-specific performance. The mapping is normally created by training an affine stitching layer on the task at hand while freezing the two half-networks, a method called task loss matching. Here, we argue that task loss matching may be very misleading as a similarity index. For example, it can indicate very high similarity between very distant layers, whose representations are known to have different functional properties. Moreover, it can indicate very distant layers to be more similar than architecturally corresponding layers. Even more surprisingly, when comparing layers within the same network, task loss matching often indicates that some layers are more similar to a layer than itself. We argue that the main reason behind these problems is that task loss matching tends to create out-of-distribution representations to improve task-specific performance. We demonstrate that direct matching (when the mapping minimizes the distance between the stitched representations) does not suffer from these problems. We compare task loss matching, direct matching, and well-known similarity indices such as CCA and CKA. We conclude that direct matching strikes a good balance between the structural and functional requirements for a good similarity index.

cs.LG↗