Journal of Science and Technique: Section on Information and Communication Technology

aDisRAE: ADAPTIVE DISCRIMINATIVE REPRESENTATION AUTOENCODER FOR FEW-SHOT CYBERATTACK DETECTION

2026-01-12T09:18:14+07:00

Due to the scarcity of labeled anomalous data, few-shot learning has emerged as a
critical paradigm for detecting novel and rare cyberattacks. The Discriminative Representation
Autoencoder (DisRAE) framework learns a latent space where anomalies are pushed away
from a central cluster of normal data, but it struggles with advanced attacks that closely mimic
benign behavior. These subtle anomalies are often mapped too close to the normal cluster,
leading to detection evasion. To address this limitation, this paper proposes the Adaptive
Discriminative Representation Autoencoder (aDisRAE). The framework enhances the training
objective by incorporating a prior outlier score that quantifies the subtlety of each anomaly.
This score guides an adaptive repulsion mechanism, applying a stronger force to anomalies
that most resemble normal data, ensuring a more effective separation in the latent space. The
experiments evaluate aDisRAE on three public benchmark datasets: NSL-KDD, CIC-IDS2017
and UNSW-NB15. The results show a notable improvement, raising AUC by up to 10% and
boosting robustness, especially against evasive attacks.

ENHANCING COPYRIGHT PROTECTION AND PROVENANCE IN NFTS WITH BLOCKCHAIN-INTEGRATED FREQUENCY-DOMAIN WATERMARKING

2026-01-12T09:18:20+07:00

The rapid expansion of the Non-Fungible Token (NFT) market has underscored
significant challenges in copyright protection and ownership authentication. While
blockchain technology ensures the immutability and transparency of token transactions, the
off-chain storage of metadata and original content remains a critical vulnerability, exposing
NFTs to risks such as data loss, manipulation, and copyright disputes. In response to these
challenges, this study proposes a blockchain-integrated watermarking framework that
embeds resilient copyright information into digital assets via a general frequency-domain
approach. The watermark is stored off-chain within the InterPlanetary File System (IPFS),
while its associated Content Identifier (CID) is anchored in a smart contract, ensuring
traceability of provenance and verification of authenticity. Comparative experiments with
the Least Significant Bit (LSB) method demonstrate the superior robustness of the proposed
frequency-domain technique against various attacks, including compression, noise, and
image manipulation. The proposed framework significantly enhances copyright protection,
facilitates transparent NFT provenance, and provides a scalable foundation for secure digital
asset management within blockchain-based ecosystems.

ENHANCING ADVERSARIAL ROBUSTNESS IN MACHINE LEARNING-BASED MALWARE DETECTION VIA ACTIVATION FUNCTION DESIGN

2026-01-12T09:18:25+07:00

In recent years, machine learning (ML) has significantly enhanced the efficiency of
malware detection systems. Despite achieving high performance, these models now face a
growing threat from adversarial attacks. Adversarial malware samples can be intricately
crafted to deceive detection models, resulting in misclassifications of malicious programs,
thereby allowing them to bypass security systems. Various techniques have been developed
to generate adversarial malware specifically designed to evade different ML-based detection
systems. This threat underscores the urgent need for solutions that enhance the resilience of
malware detection models against adversarial attacks. The paper evaluates and proposes an
empirical _{cost-efficient} adversarial defense strategy recommendation via activation function
design, that does not require computationally intensive methods such as adversarial training,
while boosting the inherent resilience of ML-based malware detection models against
black-box attacks. Results show that specific combinations, in particular Rectified Linear
Unit (ReLU) and Tanh, can significantly boost robustness without additional training or
inference setup. This work provides an empirical design aspect for building intrinsically
robust ML-based malware detection systems.

AMCF-NET: ADAPTIVE MULTI-SCALE CROSS-MODAL FUSION NETWORK FOR UAV-SATELLITE CROSS-VIEW LOCALIZATION

2026-01-12T09:18:29+07:00

Cross-view localization between Unmanned Aerial Vehicle (UAV) and satellite imagery
is crucial for autonomous navigation in GPS-denied environments. However, large domain
gaps, including viewpoint discrepancies, scale variations, and appearance differences — pose
significant challenges. In this paper, we propose the Adaptive Multi-scale Cross-modal Fusion
Network (AMCF-Net), a novel approach that effectively addresses these limitations through a
shared backbone architecture and adaptive fusion mechanisms. Unlike previous dual-backbone
approaches that process UAV and satellite images separately, our method employs a unified
FocalNet-Tiny backbone to extract cross-modal features, followed by a Spatially-adaptive Crossmodal
Feature Fusion (AMCF) module that dynamically combines multi-scale similarities
using learned adaptive weights. This shared representation learning enables better cross-modal
alignment and significantly reduces computational overhead. Comprehensive experiments on
the UL14 benchmark demonstrate that AMCF-Net achieves state-of-the-art performance, with a
Relative Distance Score (RDS) of 78.12% and meter-level accuracy of 27.25% at 3 m, 50.16%
at 5 m, 84.37% at 10 m, and finally 88.51% at 20 m. Ablation studies further validate the
effectiveness of the shared backbone and adaptive fusion mechanism, demonstrating significant
improvements over traditional separate processing approaches.

VINEURO: A MULTIMODAL EEG-BLOOD FUSION MODEL FOR ALZHEIMER’S RISK PREDICTION

2026-01-12T09:18:33+07:00

Early prediction of Alzheimer’s disease risk is crucial for timely intervention but
remains challenging in routine clinical practice. Electroencephalography (EEG) is
inexpensive and non-invasive, yet EEG alone often lacks sufficient sensitivity and
robustness for reliable early-stage risk estimation. In parallel, routine blood tests capture
peripheral immune, inflammatory, and metabolic changes associated with cognitive decline,
suggesting that combining EEG with blood-based biomarkers could yield more informative
risk stratification. In this work, ViNeuro, a multimodal EEG–blood model tailored to
Alzheimer’s risk prediction, is proposed. A single EEG foundation encoder, termed
ViNeuro-EEG, is first pretrained using the dual self-supervised objective of the EEGPT
model with the criss-cross backbone and learned positional encoding of the CBraMOD
model. Pretraining is conducted on a unified corpus of multi-channel clinical EEG data that
includes Vietnamese recordings from 108 Military Central Hospital and international
datasets. On top of this encoder, a multimodal extension, ViNeuro-MM, is constructed by
projecting routine blood biomarkers into the EEG embedding space and using them as
queries in a cross-attention layer over EEG tokens. The proposed framework is evaluated on
the PEARL-Neuro cohort for Alzheimer’s risk prediction. Compared to its EEG-only
counterpart, ViNeuro-MM achieves substantial performance gains, with relative
improvements of up to 24.72% in balanced accuracy, demonstrating that fusing routine
blood-based biomarkers with EEG foundation representations can markedly enhance early
Alzheimer’s risk prediction.

DEVELOPING OBJECT DETECTION ALGORITHM FOR OPTOELECTRONIC SYSTEMS ON SURFACE VESSELS USING DEEP LEARNING MODELS

2026-01-12T09:18:37+07:00

Automatic detection of surface vessels is an important task in maritime surveillance
and security. This paper proposes an improvement to the Ultralytics YOLOv8 model to
achieve higher accuracy and faster processing speed when recognizing surface vessels under
harsh lighting and weather conditions. The paper intergrates three main techniques: a new
C3Plus block, a Position-wise Spatial Attention (PSA) mechanism, and a Convolutional
Block Attention Module (CBAM) module to enhance the network’s feature learning ability.
Experiments on a diverse ship image dataset show that the improved model provides an
increase in mAP of about 3–6% compared to the original YOLOv8 while maintaining a
similar processing speed. In particular, in dark or noisy conditions, the CBAM and PSA
improvements help reduce missing objects and improve the model’s robustness.