Author name cluster
Possible papers associated with this exact author name in Arrow. This page groups case-insensitive exact name matches and is not a full identity disambiguation profile.
EAAI Journal 2025 Journal Article
EAAI Journal 2025 Journal Article
The catastrophic effects of Pine Wilt Disease (PWD) and the lack of a defined cure make it a danger to the world’s forests. The early detection of PWD becomes essential when establishing efficient approaches to mitigation. However, collecting data, labelling, and proposing a system for PWD detection is a challenge. This work presents an innovative PWD system and uses synthetic data sets to overcome these challenges. In particular, it emphasises two contributions: (1) the implementation of a multi-head external attention mechanism focused on improving computational effectiveness and model performance, and (2) the integration of synthetic data to address the problem of data scarcity in PWD detection. To solve the high computational cost of the self-attention function from the original transformer, we introduce external attention, which enables two linear transformations to take fewer amounts of parameters. Whereas the original self-attention is designed to capture relations inside a single sample, the external attention seeks inter-sample connections through memory modules. It is also computationally efficient compared to the wider context information, and it may consume as much computational power as the 1 by 1 convolution. Our external attention mechanism is multi-head attention, which enables the model to pay close attention to multiple features in the data and find different types of relationships. The proposed methodology achieves high performance across multiple metrics for both individual and ensemble datasets. For individual datasets, the model achieves accuracy (90%–98%), Global Detection Rate (GDR) (90%–98%), Matthews correlation coefficient (MCC) (0. 80–0. 96) and Kappa (0. 80–0. 96). For ensemble datasets, the model achieves accuracy (94%–96%), GDR (94%–96%), MCC (0. 88 - 0. 92) and Kappa (0. 88 - 0. 92). These results confirm that the presented technique is valid and effective in the early-stage detection of PWD in trees, and can be implemented in forest management.
AAAI Conference 2025 Conference Paper
Vision Transformers (ViTs) have achieved remarkable success in various computer vision tasks. However, ViTs have a huge computational cost due to their inherent reliance on multi-head self-attention (MHSA), prompting efforts to accelerate ViTs for practical applications. To this end, recent works aim to reduce the number of tokens, mainly focusing on how to effectively prune or merge them. Nevertheless, since ViT tokens are generated from non-overlapping grid patches, they usually do not convey sufficient semantics, making it incompatible with efficient ViTs. To address this, we propose ImagePiece, a novel re-tokenization strategy for Vision Transformers. Following the MaxMatch strategy of NLP tokenization, ImagePiece groups semantically insufficient yet locally coherent tokens until they convey meaning. This simple retokenization is highly compatible with previous token reduction methods, being able to drastically narrow down relevant tokens, enhancing the inference speed of DeiT-S by 54% (nearly 1.5x faster) while achieving a 0.39% improvement in ImageNet classification accuracy. For hyper-speed inference scenarios (with 251% acceleration), our approach surpasses other baselines by an accuracy over 8%.
AAAI Conference 2024 Conference Paper
Recent advances in vision language pretraining (VLP) have been largely attributed to the large-scale data collected from the web. However, uncurated dataset contains weakly correlated image-text pairs, causing data inefficiency. To address the issue, knowledge distillation have been explored at the expense of extra image and text momentum encoders to generate teaching signals for misaligned image-text pairs. In this paper, our goal is to resolve the misalignment problem with an efficient distillation framework. To this end, we propose ECLIPSE: Expediting Contrastive Language-Image Pretraining with Self-distilled Encoders. ECLIPSE features a distinctive distillation architecture wherein a shared text encoder is utilized between an online image encoder and a momentum image encoder. This strategic design choice enables the distillation to operate within a unified projected space of text embedding, resulting in better performance. Based on the unified text embedding space, ECLIPSE compensates for the additional computational cost of the momentum image encoder by expediting the online image encoder. Through our extensive experiments, we validate that there is a sweet spot between expedition and distillation where the partial view from the expedited online image encoder interacts complementarily with the momentum teacher. As a result, ECLIPSE outperforms its counterparts while achieving substantial acceleration in inference speed.
NeurIPS Conference 2022 Conference Paper
Pre-training vision-language models with contrastive objectives has shown promising results that are both scalable to large uncurated datasets and transferable to many downstream applications. Some following works have targeted to improve data efficiency by adding self-supervision terms, but inter-domain (image-text) contrastive loss and intra-domain (image-image) contrastive loss are defined on individual spaces in those works, so many feasible combinations of supervision are overlooked. To overcome this issue, we propose UniCLIP, a Unified framework for Contrastive Language-Image Pre-training. UniCLIP integrates the contrastive loss of both inter-domain pairs and intra-domain pairs into a single universal space. The discrepancies that occur when integrating contrastive loss between different domains are resolved by the three key components of UniCLIP: (1) augmentation-aware feature embedding, (2) MP-NCE loss, and (3) domain dependent similarity measure. UniCLIP outperforms previous vision-language pre-training methods on various single- and multi-modality downstream tasks. In our experiments, we show that each component that comprises UniCLIP contributes well to the final performance.
ICRA Conference 2010 Conference Paper
In this paper, we present a global localization and local pose error compensation method in a known structured environment using magnetic landmarks. In previous our research, it was possible to compensate the pose error (xe, ye, qe) of a mobile robot correctly on the surface of structured environment with magnetic landmarks. In this work, we propose a methodology of arranging magnetic landmarks on the map such that properly arranged magnetic patterns ease the global localization of a mobile agent. Among total six patterns of magnetic-bar in square arrangement, five unique landmarks are obtained. Therefore, a heuristic pattern search method is applied to build the virtual map using five landmarks. In order to obtain the global pose information, the robot identifies the pattern of magnets, and obtains the current global pose information by comparing the measured neighboring patterns with the map information that is saved in advance. Experimental results show the effectiveness of the magnetic-pattern landmarks for the global localization and local pose control of a mobile robot.