arXiv subjects

Xingchen Chen

Publications and source records attributed to Xingchen Chen.

4 recordsLinked to original sources

arXiv2026

Delta-JEPA: Learning Action-Sensitive World Models via Latent Difference Decoding

Learning visual world models for planning requires compact latent dynamics that remain sensitive to actions, yet reconstruction-free joint-embedding objectives can collapse to action-insensitive representations. We propose Delta-JEPA, an end-to-end reconstruction-free world model that augments latent forward prediction with a Latent Difference Action Decoder (LDAD). Unlike inverse decoders that infer actions from concatenated endpoint embeddings, LDAD reconstructs the executed action from the latent displacement between consecutive observations. This displacement-level supervision directly regularizes transition geometry: adjacent embeddings cannot collapse without losing action information, and different actions are encouraged to induce distinguishable latent changes for rollout-based planning. Delta-JEPA uses only latent prediction and action reconstruction, avoiding pixel reconstruction and distribution-matching regularizers. Across four visual continuous-control tasks, Delta-JEPA improves planning over JEPA-based and representation-learning world model baselines. Ablations show that displacement-based action decoding is consistently more effective than endpoint concatenation, and action-sensitivity analyses show clearer action-conditioned latent responses. These results indicate that supervising latent differences is a simple and effective mechanism for collapse-resistant and action-sensitive world model learning.

cs.AI↗

arXiv2025

Dynamic Deep Graph Learning for Incomplete Multi-View Clustering with Masked Graph Reconstruction Loss

The prevalence of real-world multi-view data makes incomplete multi-view clustering (IMVC) a crucial research. The rapid development of Graph Neural Networks (GNNs) has established them as one of the mainstream approaches for multi-view clustering. Despite significant progress in GNNs-based IMVC, some challenges remain: (1) Most methods rely on the K-Nearest Neighbors (KNN) algorithm to construct static graphs from raw data, which introduces noise and diminishes the robustness of the graph topology. (2) Existing methods typically utilize the Mean Squared Error (MSE) loss between the reconstructed graph and the sparse adjacency graph directly as the graph reconstruction loss, leading to substantial gradient noise during optimization. To address these issues, we propose a novel \textbf{D}ynamic Deep \textbf{G}raph Learning for \textbf{I}ncomplete \textbf{M}ulti-\textbf{V}iew \textbf{C}lustering with \textbf{M}asked Graph Reconstruction Loss (DGIMVCM). Firstly, we construct a missing-robust global graph from the raw data. A graph convolutional embedding layer is then designed to extract primary features and refined dynamic view-specific graph structures, leveraging the global graph for imputation of missing views. This process is complemented by graph structure contrastive learning, which identifies consistency among view-specific graph structures. Secondly, a graph self-attention encoder is introduced to extract high-level representations based on the imputed primary features and view-specific graphs, and is optimized with a masked graph reconstruction loss to mitigate gradient noise during optimization. Finally, a clustering module is constructed and optimized through a pseudo-label self-supervised training mechanism. Extensive experiments on multiple datasets validate the effectiveness and superiority of DGIMVCM.

cs.LG↗

arXiv2022

Low-Energy Electron Microscopy contrast of stacking boundaries: comparing twisted few-layer graphene and strained epitaxial graphene on silicon carbide

Stacking domain boundaries occur in Van der Waals heterostacks whenever there is a twist angle or lattice mismatch between subsequent layers. Not only can these domain boundaries host topological edge states, imaging them has been instrumental to determine local variations in twisted bilayer graphene. Here, we analyse the mechanisms causing stacking domain boundary contrast in Bright Field Low-Energy Electron Microscopy (BF-LEEM) for both graphene on SiC, where domain boundaries are caused by strain and for twisted few layer graphene. We show that when domain boundaries are between the top two graphene layers, BF-LEEM contrast is observed due to amplitude contrast and corresponds well to calculations of the contrast based purely on the local stacking in the domain boundary. Conversely, for deeper-lying domain boundaries, amplitude contrast only provides a weak distinction between the inequivalent stackings in the domains themselves. However, for small domains phase contrast, where electrons from different parts of the unit cell interfere causes a very strong contrast. We derive a general rule-of-thumb of expected BF-LEEM contrast for domain boundaries in Van der Waals materials.

cond-mat.mes-hall↗

arXiv2021

Imaging moiré deformation and dynamics in twisted bilayer graphene

In twisted bilayer graphene (TBG) a moiré pattern forms that introduces a new length scale to the material. At the 'magic' twist angle of 1.1°, this causes a flat band to form, yielding emergent properties such as correlated insulator behavior and superconductivity [1-4]. In general, the moiré structure in TBG varies spatially, influencing the local electronic properties [5-9] and hence the outcome of macroscopic charge transport experiments. In particular, to understand the wide variety observed in the phase diagrams and critical temperatures, a more detailed understanding of the local moiré variation is needed [10]. Here, we study spatial and temporal variations of the moiré pattern in TBG using aberration-corrected Low Energy Electron Microscopy (AC-LEEM) [11,12]. The spatial variation we find is lower than reported previously. At 500°C, we observe thermal fluctuations of the moiré lattice, corresponding to collective atomic displacements of less than 70pm on a time scale of seconds [13], homogenizing the sample. Despite previous concerns, no untwisting of the layers is found, even at temperatures as high as 600°C [14,15]. From these observations, we conclude that thermal annealing can be used to decrease the local disorder in TBG samples. Finally, we report the existence of individual edge dislocations in the atomic and moiré lattice. These topological defects break translation symmetry and are anticipated to exhibit unique local electronic properties.

cond-mat.mes-hall↗