Hanul Kim | Searcharxiv

arXiv2026

Decision MetaMamba: Enhancing Selective SSM in Offline RL with Heterogeneous Sequence Mixing

Mamba-based models have drawn much attention in offline RL. However, their selective mechanism often detrimental when key steps in RL sequences are omitted. To address these issues, we propose a simple yet effective structure, called Decision MetaMamba (DMM), which replaces Mamba's token mixer with a dense layer-based sequence mixer and modifies positional structure to preserve local information. By performing sequence mixing that considers all channels simultaneously before Mamba, DMM prevents information loss due to selective scanning and residual gating. Extensive experiments demonstrate that our DMM delivers the state-of-the-art performance across diverse RL tasks. Furthermore, DMM achieves these results with a compact parameter footprint, demonstrating strong potential for real-world applications.

cs.LG↗

arXiv2026

MultiModalPFN: Extending Prior-Data Fitted Networks for Multimodal Tabular Learning

Recently, TabPFN has gained attention as a foundation model for tabular data. However, it struggles to integrate heterogeneous modalities such as images and text, which are common in domains like healthcare and marketing, thereby limiting its applicability. To address this, we present the Multi-Modal Prior-data Fitted Network (MMPFN), which extends TabPFN to handle tabular and non-tabular modalities in a unified manner. MMPFN comprises per-modality encoders, modality projectors, and pre-trained foundation models. The modality projectors serve as the critical bridge, transforming non-tabular embeddings into tabular-compatible tokens for unified processing. To this end, we introduce a multi-head gated MLP and a cross-attention pooler that extract richer context from non-tabular inputs while mitigates attention imbalance issue in multimodal learning. Extensive experiments on medical and general-purpose multimodal datasets demonstrate that MMPFN consistently outperforms competitive state-of-the-art methods and effectively exploits non-tabular modalities alongside tabular features. These results highlight the promise of extending prior-data fitted networks to the multimodal setting, offering a scalable and effective framework for heterogeneous data learning. The source code is available at https://github.com/too-z/MultiModalPFN.

cs.LG↗

arXiv2024

Decision MetaMamba: Enhancing Selective SSM in Offline RL with Heterogeneous Sequence Mixing

Mamba-based models have drawn much attention in offline RL. However, their selective mechanism often detrimental when key steps in RL sequences are omitted. To address these issues, we propose a simple yet effective structure, called Decision MetaMamba (DMM), which replaces Mamba's token mixer with a dense layer-based sequence mixer and modifies positional structure to preserve local information. By performing sequence mixing that considers all channels simultaneously before Mamba, DMM prevents information loss due to selective scanning and residual gating. Extensive experiments demonstrate that our DMM delivers the state-of-the-art performance across diverse RL tasks. Furthermore, DMM achieves these results with a compact parameter footprint, demonstrating strong potential for real-world applications. Code is available at https://github.com/too-z/decision-metamamba

cs.LG↗

arXiv2023

Yield Stress Fluids Solidifying in Capillary Imbibition

When subjected to an external stress that exceeds the yield stress ($\sigma_\mathrm{Y}$), yield stress fluids (YSFs) undergo a solid-to-liquid transition. Despite the extensive studies, there has been limited attention to the process of liquid-to-solid transition. This work examines the solidification of YSFs through capillary imbibition, easily observed in the processes of wetting, coating, spreading, and wicking. During gradual deceleration of the capillary rise, YSFs display an unexpected flowing behavior, even when subjected to stresses below the $\sigma_\mathrm{Y}$. We propose a model with numerical solutions based on rheological properties of YSFs and slip to capture this unusual, yet universal behavior.

cond-mat.soft↗

arXiv2019

Genuine Ohmic van der Waals contact between indium and MoS2

The formation of an ideal van der Waals (vdW) contacts at metal/transition-metal dichalcogenide (TMDC) interfaces is a critical step for the development of high-performance and energy-efficient electronic and optoelectronic applications based on the two-dimensional (2D) semiconductors. In overcoming the key chal-lenges of the conventional metal deposition process that leads to an uncontrol-lable Schottky barrier height and high contact resistance, notable advances were recently made by transferring atomically flat metal thin films or thermally evapo-rating indium/gold alloy. However, the realization of an ideal vdW contact be-tween an elemental metal and TMDC through the evaporation process is yet to be demonstrated, and particularly the evidence of an Ohmic contact between three-dimensional metallic electrodes and TMDCs is still unavailable. Herein, we report the fabrication of atomically clean metal/TMDC contacts by evaporating metals at a relatively low thermal energy and subsequently cooling the substrate holder down to 100 K by liquid nitrogen, achieving for the indium (In)/molybdenum disulfide (MoS2) case an accumulation-type Ohmic contact with a metal-induced electron doping density of 10$^{12}$/cm$^2$. We find that the transport at the In/MoS2 contact is dominated by the field-emission mechanism over a wide temperature range from 2.4 to 300 K, and the contact resistance reaches 600 Ohm um and 1,000 Ohm um at cryogenic temperatures for the few-layer and monolayer MoS2 cases, respectively. Based on first-principles calculations, we find that the na-ture of the ideal In/MoS2 vdW contact is characterized by the formation of in-gap states within TMDC together with the abrupt and rigid shift of the TMDC band.

cond-mat.mes-hall↗

arXiv2017

Energy dissipation mechanism revealed by spatially resolved Raman thermometry of graphene/hexagonal boron nitride heterostructure devices

Understanding the energy transport by charge carriers and phonons in two-dimensional (2D) van der Waals heterostructures is essential for the development of future energy-efficient 2D nanoelectronics. Here, we performed in situ spatially resolved Raman thermometry on an electrically biased graphene channel and its hBN substrate to study the energy dissipation mechanism in graphene/hBN heterostructures. By comparing the temperature profile along the biased graphene channel with that along the hBN substrate, we found that the thermal boundary resistance between the graphene and hBN was in the range of (1-2) x 10^(-7) m^(2) KW^(-1) from ~100 C to the onset of graphene break-down at ~600 C in air. Consideration of an electro-thermal transport model together with the Raman thermometry conducted in air showed that local doping occurred under a strong electric field played a crucial role in the energy dissipation of the graphene/hBN device up to T ~ 600 C.

cond-mat.mes-hall↗