A5044708805- Video Coding and Compression Technologies
- Advanced Vision and Imaging
- Advanced Data Compression Techniques
- Wireless Communication Security Techniques
- Natural Language Processing Techniques
- Topic Modeling
- Video Analysis and Summarization
- Advanced Image and Video Retrieval Techniques
- 3D Shape Modeling and Analysis
- Image and Video Quality Assessment
- Computer Graphics and Visualization Techniques
- Chaos-based Image/Signal Encryption
- Speech Recognition and Synthesis
- Face recognition and analysis
- Algorithms and Data Compression
- Multimedia Communication and Technology
- Advanced Image Processing Techniques
- Advanced Steganography and Watermarking Techniques
- Music and Audio Processing
- 3D Surveying and Cultural Heritage
- Video Surveillance and Tracking Methods
- Visual Attention and Saliency Detection
- Image Retrieval and Classification Techniques
- Face and Expression Recognition
- Digital Rights Management and Security
University of Lisbon
2016-2025
Instituto de Telecomunicações
2016-2025
Iscte – Instituto Universitário de Lisboa
2024
Weatherford College
2024
Instituto Politécnico de Lisboa
1990-2023
Governo do Estado de São Paulo
2023
Instituto Superior de Tecnologias Avançadas
2015-2023
Voith (Germany)
2023
Secretaria Regional do Ambiente e Recursos Naturais
2023
Zimmer Biomet (United States)
2023
Discriminative learning methods for classification perform well when training and test data are drawn from the same distribution. Often, however, we have plentiful labeled a source domain but wish to learn classifier which performs on target with different distribution little or no data. In this work investigate two questions. First, under what conditions can trained be expected data? Second, given small amount of data, how should combine it during large achieve lowest error at time? We...
Problems that involve interacting with humans, such as natural language understanding, have not proven to be solvable by concise, neat formulas like F = ma. Instead, the best approach appears embrace complexity of domain and address it harnessing power data: if other humans engage in tasks generate large amounts unlabeled, noisy data, new algorithms can used build high-quality models from data.
Discriminative learning methods are widely used in natural language processing. These work best when their training and test data drawn from the same distribution. For many NLP tasks, however, we confronted with new domains which labeled is scarce or non-existent. In such cases, seek to adapt existing models a resource-rich source domain resource-poor target domain. We introduce structural correspondence automatically induce correspondences among features different domains. our technique on...
Conditional random fields for sequence labeling offer advantages over both generative models like HMMs and classifiers applied at each position. Among tasks in language processing, shallow parsing has received much attention, with the development of standard evaluation datasets extensive comparison among methods. We show here how to train a conditional field achieve performance as good any reported base noun-phrase chunking method on CoNLL task, better than single model. Improved training...
We describe and evaluate experimentally a method for clustering words according to their distribution in particular syntactic contexts. Words are represented by the relative frequency distributions of contexts which they appear, entropy between those is used as similarity measure clustering. Clusters average context derived from given probabilities cluster membership. In many cases, clusters can be thought encoding coarse sense distinctions. Deterministic annealing find lowest distortion...
H.264/AVC, the result of collaboration between ISO/IEC Moving Picture Experts Group and ITU-T Video Coding Group, is latest standard for video coding. The goals this standardization effort were enhanced compression efficiency, network friendly representation interactive (video telephony) non-interactive applications (broadcast, streaming, storage, on demand). H.264/AVC provides gains in efficiency up to 50% over a wide range bit rates resolutions compared previous standards. Compared...
We formalize weighted dependency parsing as searching for maximum spanning trees (MSTs) in directed graphs. Using this representation, the algorithm of Eisner (1996) is sufficient over all projective O(n3) time. More surprisingly, representation extended naturally to non-projective using Chu-Liu-Edmonds (Chu and Liu, 1965; Edmonds, 1967) MST algorithm, yielding an O(n2) algorithm. evaluate these methods on Prague Dependency Treebank online large-margin learning techniques (Crammer et al.,...
We present an effective training algorithm for linearly-scored dependency parsers that implements online large-margin multi-class (Crammer and Singer, 2003; Crammer et al., 2003) on top of efficient parsing techniques trees (Eisner, 1996). The trained achieve a competitive accuracy both English Czech with no language specific enhancements.
We introduce confidence-weighted linear classifiers, which add parameter confidence information to classifiers. Online learners in this setting update both classifier parameters and the estimate of their confidence. The particular online algorithms we study here maintain a Gaussian distribution over vectors mean covariance with each instance. Empirical evaluation on range NLP tasks show that our algorithm improves other state art batch methods, learns faster setting, lends itself better...
In discussing the rationale behind vision for JPEG Pleno and how new standardization initiative aims to reinvent future of imaging, authors review plenoptic representation its underlying practical implications challenges in implementing real-world applications with an enhanced quality experience.
The inside-outside algorithm for inferring the parameters of a stochastic context-free grammar is extended to take advantage constituent information (constituent bracketing) in partially parsed corpus.Experiments on formal and natural language corpora show that new can achieve faster convergence better modeling hierarchical structure than original one.In particular, over 90% test set bracketing accuracy was achieved grammars inferred by our from training handparsed part-of-speech strings...
The recently completed ISO/IEC, International Standard 15938, formally called the Multimedia Content Description Interface (but better known as MPEG-7), provides a rich set of tools for completely describing multimedia content. standard wasn't just designed from content management viewpoint (classical archival information). It includes an innovative description media's content, which we can extract via analysis and processing. MPEG-7 also isn't aimed at any one application; rather, elements...
Prolog is a simple but powerful programming language founded on symbolic logic. The basic computational mechanism pattern matching process ("unification") operating general record structures ("terms" of logic). We briefly review the and compare it especially with pure Lisp. remainder paper discusses techniques for implementing efficiently; in particular we describe how to compile patterns involved process. These are as incorporated our DECsystem-10 compiler (written Prolog). code generates...
We present a two-stage multilingual dependency parser and evaluate it on 13 diverse languages. The first stage is based the unlabeled parsing models described by McDonald Pereira (2006) augmented with morphological features for subset of second takes output from labels all edges in graph appropriate syntactic categories using globally trained sequence classifier over components graph. report results CoNLL-X shared task (Buchholz et al., 2006) data sets an error analysis.
Abstract Background We present a model for tagging gene and protein mentions from text using the probabilistic sequence framework of conditional random fields (CRFs). Conditional probability P ( t | o ) tag given an observation directly, have previously been employed successfully other tasks. The mechanics CRFs their relationship to maximum entropy are discussed in detail. Results employ diverse feature set containing standard orthographic features combined with expert form biological term...
In recent years, practical Wyner-Ziv (WZ) video coding solutions have been proposed with promising results. Most of the available in literature model correlation noise (CN) between original frame and its estimation made at decoder, which is so-called side information (SI), by a given distribution whose relevant parameters are estimated using an offline process, assuming that SI encoder or originals decoder. The major goal this paper to propose more realistic WZ approach performing online CN...
MPEG-21 is an open standards-based framework for multimedia delivery and consumption. It aims to enable the use of resources across a wide range networks devices. We discuss MPEG-21's parts, achievements, ongoing activities, opportunities new technologies.
Open source tools have recently reached a level of maturity which makes them suitable for building large-scale real-world systems. At the same time, field machine learning has developed large body powerful algorithms diverse applications. However, true potential these methods is not used, since existing implementations are openly shared, resulting in software with low usability, and weak interoperability. We argue that this situation can be significantly improved by increasing incentives...