A5046897413- Robotics and Sensor-Based Localization
- Web Data Mining and Analysis
- Advanced Database Systems and Queries
- Domain Adaptation and Few-Shot Learning
- Data Management and Algorithms
- Advanced Vision and Imaging
- Advanced Image and Video Retrieval Techniques
- Semantic Web and Ontologies
- Multimodal Machine Learning Applications
- Algorithms and Data Compression
- Natural Language Processing Techniques
- Human Mobility and Location-Based Analysis
- Topic Modeling
- Indoor and Outdoor Localization Technologies
- Robotic Path Planning Algorithms
- Caching and Content Delivery
- Image Processing Techniques and Applications
- Machine Learning and Data Classification
- Text and Document Classification Technologies
- Image Retrieval and Classification Techniques
- Recommender Systems and Techniques
- Machine Learning and Algorithms
- Optical measurement and interference techniques
- Transportation and Mobility Innovations
- COVID-19 diagnosis using AI
Naver (South Korea)
2020-2024
Xerox (France)
2006-2016
Xerox (United States)
2006
Every year, for ten years now, the IPIN competition has aimed at evaluating real-world indoor localisation systems by testing them in a realistic environment, with movement, using EvAAL framework. The provided unique overview of state-of-the-art systems, technologies, and methods positioning navigation purposes. Through fair comparison performance achieved each system, was able to identify most promising approaches pinpoint critical working conditions. In 2020, included 5 diverse off-site...
Despite impressive performance for high-level downstream tasks, self-supervised pre-training methods have not yet fully delivered on dense geometric vision tasks such as stereo matching or optical flow. The application of concepts, instance discrimination masked image modeling, to is an active area research. In this work, we build the recent cross-view completion framework, a variation modeling that leverages second view from same scene which makes it well suited binocular tasks....
The overwhelming majority of existing domain adaptation methods makes an assumption freely available source data. An equal access to both and target data it possible measure the discrepancy between their distributions build representations common domains. In reality, such a simplifying rarely holds, since are routinely subject legal contractual constraints owners customers. When can not be accessed, decision making procedures often for nevertheless. These presented in form classification,...
IPIN 2019 Competition, sixth in a series of competitions, was held at the CNR Research Area Pisa (IT), integrated into program Conference. It included two on-site real-time Tracks and three off-site Tracks. The four presented this paper were set same environment, made buildings close together for total usable area 1000 m <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> outdoors 6000 indoors over floors, with path length exceeding 500 m....
Finding domain invariant features is critical for successful adaptation and transfer learning. However, in the case of unsupervised adaptation, there a significant risk overfitting on source training data. Recently, regularization was proposed deep models by (Ganin Lempitsky, 2015). We build their work suggesting more appropriate denoising autoencoders. Our model remains can be computed closed form. On standard text classification tasks, our approach yields state art results, with an...
We address the problem of indoor localization based on WiFi signal strengths. develop a semi-supervised deep learning method able to train prediction model from small set annotated observations and massive non-annotated ones. Our is variational autoencoder network. complement network with an additional component structural projection further improve accuracy in complex, multi-building multi-floor environment. consider several different compositions which combine classification regression...
Semantic segmentation plays a fundamental role in broad variety of computer vision applications, providing key information for the global understanding an image. Yet, state-of-the-art models rely on large amount annotated samples, which are more expensive to obtain than tasks such as image classification. Since unlabelled data is instead significantly cheaper obtain, it not surprising that Unsupervised Domain Adaptation reached success within semantic community. This survey effort summarize...
Collaborative tagging systems are now deployed extensively to help users share and organize resources. Tag prediction recommendation can simplify streamline the user experience, by modeling preferences, predictive accuracy be significantly improved. However, previous methods typically model behavior based only on a log of prior tags, neglecting other behaviors information in social systems, e.g., commenting items connecting with users. On hand, little is known about connection correlations...
Existing approaches for learning local image descriptors have shown remarkable achievements in a wide range of geometric tasks. However, most them require perpixel correspondence-level supervision, which is difficult to acquire at scale and high quality. In this paper, we propose explicitly integrate two matching priors single loss order learn without supervision. Given images depicting the same scene, extract pixel build correlation volume. The first prior enforces consistency matches...
We address the problem of tight XML schemas and propose regular tree automata to model data. show that is more powerful than DTDs closed under main algebraic operations. introduce query algebra based on model, discuss optimization pruning techniques. Finally we conversion schema into DTDs.
Masked Image Modeling (MIM) has recently been established as a potent pre-training paradigm. A pretext task is constructed by masking patches in an input image, and this masked content then predicted neural network using visible sole input. This leads to state-of-the-art performance when finetuned for high-level semantic tasks, e.g. image classification object detection. In paper we instead seek learn representations that transfer well wide variety of 3D vision lower-level geometric...
XML Schema language has been proposed to replace Document Type Definitions (DTDs) as schema mechanism for data. This consistently extends grammar-based constructions with constraint- and pattern-based ones have a higher expressive power than DTDs. As schemas remain optional XML, we address the problem of extraction. We model extended context-free grammars develop novel extraction algorithm inspired by methods grammatical inference. The copes also determinism requirement imposed DTDs languages.
Documentum Enterprise Content Integration (ECI) services is a content integration middleware that provides one-query access to the Intranet and Internet resources. The ECI Adapter technology offers an interface any application for data metadata extraction from unstructured Web pages. It unique frame-work of wrapper production, automatic recovery maintenance, developed at Xerox Research Centre Europe based on state-of-art algorithms machine learning grammatical inference. In this presentation...
We propose a new deep learning architecture for the tasks of semantic segmentation and depth prediction from RGB-D images. revise state art based on RGB feature fusion, where both modalities are assumed to be available at train test time. fusion is replaced with common representation. Combined an encoder-decoder type network, can jointly learn models estimation their This representation, inspired by multi-view learning, offers several important advantages, such as using one modality time...