arXiv subjects

Dongsu Kim

Publications and source records attributed to Dongsu Kim.

8 recordsLinked to original sources

arXiv2026

PENet+: A Lightweight Residual Transformer Framework for Efficient Image Steganalysis

Image steganalysis, the detection of hidden information embedded in digital images, is a core component of modern cybersecurity and digital forensics. Recent residual Transformer architectures, such as the Pixel-Difference-Convolution and Enhanced-Transformer-Network (PENet) [1], achieve strong detection accuracy, but their computational and memory demands hinder deployment in resource-constrained settings. We present PENet+, a lightweight steganalysis framework that preserves PENet's discriminative structure while substantially improving efficiency. Rather than redesigning or compressing the attention blocks, we retain PENet's self-attention topology for reproducibility and add a classifier-streamlining stage that progressively narrows the SPP-to-FC1 input channels (SPP: spatial pyramid pooling; FC1: first fully connected layer), yielding large reductions in parameters and FLOPs with negligible accuracy loss. We further refine the high-pass-filter (HPF) stem with an activation-aware mechanism that aggregates HPF responses early and selects a balanced SRM-Gabor top-K subset, and we replace PENet's backbone with a MobileNetV2-style inverted residual network. A balanced configuration with K=31 filters (16 Gabor + 15 SRM) matches or surpasses heavier settings at lower compute. Finally, we motivate PReLU from a steganalysis standpoint, arguing that preserving negative responses helps capture weak stego cues that ReLU suppresses. On a disjoint ALASKA2 JPEG QF90 protocol at 512x512 resolution (5,000 cover images for training, validation, and internal testing; a separate 19,000-cover evaluation set), PENet+ achieves up to 45.5% fewer parameters and about 97% fewer FLOPs than the re-evaluated PENet baseline, offering a computationally efficient direction for resource-constrained steganalysis. Device-level latency and power measurements remain future work.

cs.CV↗

arXiv2026

Sequential Behavioral Watermarking for LLM Agents

LLM-based agents act through sequences of executable decisions, but their trajectories provide little evidence of which agent or policy produced them, making provenance, ownership, and unauthorized reuse difficult to establish from observed behavior alone. This motivates watermarking signals embedded directly into agent behavior rather than only into generated text, since text watermarking cannot capture the action-level decisions that define agent execution. Recent agent watermarking methods address this gap by moving the watermark from generated text to behavioral choices. However, by treating each action step as an independent trial, they overlook trajectory structure and become fragile when trajectories are perturbed, truncated, or observed without reliable alignment. We propose SeqWM, a sequential behavioral watermarking framework that embeds signals into history-conditioned transition patterns and verifies trajectories position-agnostically against random-key baselines. Experiments across diverse agent benchmarks and LLM backbones show that SeqWM consistently achieves reliable detection while preserving agent utility, and remains robust under trajectory corruption where round-indexed behavioral watermarks collapse.

cs.CR↗

arXiv2026

Visualizing Millisecond Atomic Dynamics of Nanocrystals in Liquid

Atomic structures of nanomaterials are inherently dynamic, continuously reshaped through interactions with chemical species and external stimuli. Such dynamics are further amplified as the size and dimensionality of nanomaterials are reduced. Despite advances in analytical methods, it remains challenging to capture structural dynamics of nanomaterials in reactive environments with both atomic spatial resolution and commensurate temporal resolution. Here, we directly visualize atomic-scale dynamics of gold (Au) nanocrystals in reactive liquid environments with millisecond-speed liquid cell electron microscopy (EM) and deep-learning denoising. We uncover reversible fluctuations in local crystallinity of Au nanocrystals dependent on the surrounding chemical environment. These transient fluctuations, driven by interactions at nanocrystal-liquid interfaces, critically influence dissolution kinetics and grain boundary relaxation. By overcoming the spatiotemporal limitations in conventional liquid cell EM, our findings provide insights into how transient nanoscale structures dictate the stability and reactivity of nanomaterials.

cond-mat.mtrl-sci↗

arXiv2025

An involution for a Catalan-tangent number identity

We provide an involution proof of a Catalan-tangent number identity arising from the study of peak algebra that was found by Aliniaeifard and Li. In the course, we find a new combinatorial identity for the tangent numbers $T_{2n+1}$: $$ \sum_{k=0}^{n}(-1)^{k}{2n+1\choose 2k}2^{2n-2k}T_{2k+1}=(-1)^nT_{2n+1}. $$ Moreover, we derive two different $q$-analogs of the above identity from the combinatorial perspective.

math.CO↗

arXiv2020

Refined restricted inversion sequences

Recently, the study of patterns in inversion sequences was initiated by Corteel-Martinez-Savage-Weselcouch and Mansour-Shattuck independently. Motivated by their works and a double Eulerian equidistribution due to Foata (1977), we investigate several classical statistics on restricted inversion sequences that are either known or conjectured to be enumerated by {\em Catalan}, {\em Large Schröder}, {\em Baxter} and {\em Euler} numbers. One of the two highlights of our results is a fascinating bijection between $000$-avoiding inversion sequences and Simsun permutations, which together with Foata's V- and S-codes, provide a proof of a restriced double Eulerian equdistribution. The other one is a refinement of a conjecture due to Martinez and Savage that the cardinality of $\I_n(\geq,\geq,>)$ is the $n$-th Baxter number, which is proved via the so-called {\em obstinate kernel method} developed by Bousquet-Mélou.

math.CO↗

arXiv2019

A combinatorial bijection on $k$-noncrossing partitions

For any integer $k\geq2$, we prove combinatorially the following Euler (binomial) transformation identity $$ \NC_{n+1}^{(k)}(t)=t\sum_{i=0}^n{n\choose i}\NW_{i}^{(k)}(t), $$ where $\NC_{m}^{(k)}(t)$ (resp.~$\NW_{m}^{(k)}(t)$) is the sum of weights, $t^\text{number of blocks}$, of partitions of $\{1,\ldots,m\}$ without $k$-crossings (resp.~enhanced $k$-crossings). The special $k=2$ and $t=1$ case, asserting the Euler transformation of Motzkin numbers are Catalan numbers, was discovered by Donaghey 1977. The result for $k=3$ and $t=1$, arising naturally in a recent study of pattern avoidance in ascent sequences and inversion sequences, was proved only analytically.

math.CO↗

arXiv2016

A sextuple equidistribution arising in Pattern Avoidance

We construct an intriguing bijection between $021$-avoiding inversion sequences and $(2413,4213)$-avoiding permutations, which proves a sextuple equidistribution involving double Eulerian statistics. Two interesting applications of this result are also presented. Moreover, this result inspires us to characterize all permutation classes that avoid two patterns of length $4$ whose descent polynomial equals that of separable permutations.

math.CO↗

arXiv2010

A combinatorial approach to the power of 2 in the number of involutions

We provide a combinatorial approach to the largest power of $p$ in the number of permutations $π$ with $π^p=1$, for a fixed prime number $p$. With this approach, we find the largest power of $2$ in the number of involutions, in the signed sum of involutions and in the numbers of even or odd involutions.

math.CO↗