A5013593379- Handwritten Text Recognition Techniques
- Natural Language Processing Techniques
- Speech Recognition and Synthesis
- Neural Networks and Applications
- Machine Learning and Algorithms
- Topic Modeling
- Image Retrieval and Classification Techniques
- Music and Audio Processing
- Human Motion and Animation
- Advanced Image and Video Retrieval Techniques
- Sparse and Compressive Sensing Techniques
- Image Processing and 3D Reconstruction
- Video Analysis and Summarization
- Web Data Mining and Analysis
- Tensor decomposition and applications
- Text and Document Classification Technologies
- Speech and Audio Processing
- Human Pose and Action Recognition
- Time Series Analysis and Forecasting
- Semantic Web and Ontologies
- Machine Learning and Data Classification
- Recommender Systems and Techniques
- Algorithms and Data Compression
- Face and Expression Recognition
- Functional Brain Connectivity Studies
Centrale Marseille
2016-2024
Aix-Marseille Université
2016-2024
Centre National de la Recherche Scientifique
2013-2024
Laboratoire d’Informatique et Systèmes
2017-2024
Château Gombert
2016-2024
Université de Toulon
2019-2024
Institut de Neurosciences de la Timone
2024
Laboratoire d’Informatique Fondamentale de Marseille
2015-2017
Innate Pharma (France)
2016-2017
Sorbonne Université
2006-2015
This article provides an overview of the first BIOASQ challenge, a competition on large-scale biomedical semantic indexing and question answering (QA), which took place between March September 2013. assesses ability systems to semantically index very large numbers scientific articles, return concise user-understandable answers given natural language questions by combining information from articles ontologies.The 2013 comprised two tasks, Task 1a 1b. In participants were asked automatically...
LSHTC is a series of challenges which aims to assess the performance classification systems in large-scale large number classes (up hundreds thousands). This paper describes dataset that have been released along series. The details construction datsets and design tracks as well evaluation measures we implemented quick overview results. All these datasets are available online runs may still be submitted on server challenges.
Large margin learning of Continuous Density HMMs with a partially labeled dataset has been extensively studied in the speech and handwriting recognition fields. Yet due to non-convexity optimization problem, previous works usually rely on severe approximations so that it is still an open problem. We propose new algorithm relies non-convex bundle methods allows tackling original problem as is. It proved converge solution accuracy ε rate O (1/ε). provide experimental results gained demonstrate...
Object segmentation is a crucial problem that usually solved by using supervised learning approaches over very large datasets composed of both images and corresponding object masks. Since the masks have to be provided at pixel level, building such dataset for any new domain can time-consuming. We present ReDO, model able extract objects from without annotation in an unsupervised way. It relies on idea it should possible change textures or colors changing overall distribution dataset....
We investigate a new approach for online handwritten shape recognition. Interesting features of this include learning without manual tuning, from very few training samples, incremental characters, and adaptation to the user-specific needs. The proposed system can deal with two-dimensional graphical shapes such as Latin Asian command gestures, symbols, small drawings, geometric shapes. It be used building block series recognition tasks many applications.
Active learning (AL) consists of asking human annotators to annotate automatically selected data that are assumed bring the most benefit in creation a classifier.AL allows learn accurate systems with much less annotated than what is required by pure supervised algorithms, hence limiting tedious effort annotating large collection data.We experimentally investigate behavior several AL strategies for sequence labeling tasks (in partially-labeled scenario) tailored on Partially-Labeled...
This paper provides an overview of the workshop Web-Scale Classification: Web Classification in Big Data Era which was held New York City, on February 28th as a seventh International Conference Search and Mining. The goal to discuss assess recent research focusing classification mining Web-scale category systems. brought together members several communities such web mining, machine learning, text social media mining.
This paper focuses on designing a handwriting recognition system dealing with on-line signal, i.e. temporal signal captured through an electronic pen or digitalized tablet. We present here some new results concerning hybrid based Hidden Markov Models (HMMs) and Neural Networks (NNs), which has already been presented in several contributions. In our approach, letter-model is Left-Right HMM, whose emission probability densities are approximated mixtures of predictive multilayer perceptrons....
We consider the problem of learning when obtaining training labels is costly, which usually tackled in literature using active-learning techniques. These approaches provide strategies to choose examples label before or during training. are based on heuristics even theoretical measures, but not learned as they directly used design a model aims at \textit{learning strategies} meta-learning setting. More specifically, we pool-based setting, where system observes all dataset and has subset...
Abstract The cerebral processing of voice information is known to engage, in human as well non-human primates, “temporal areas” (TVAs) that respond preferentially conspecific vocalizations. However, how represented by neuronal populations these areas, particularly speaker identity information, remains poorly understood. Here, we used a deep neural network (DNN) generate high-level, small-dimension representational space for identity—the ‘voice latent space’ (VLS)—and examined its linear...
Single-cell RNA sequencing (scRNASeq) data plays a major role in advancing our understanding of developmental biology. An important current question is how to classify transcriptomic profiles obtained from scRNASeq experiments into the various cell types and identify lineage relationship for individual cells. Because fast accumulation datasets high dimensionality data, it has become challenging explore annotate single-cell by hand. To overcome this challenge, automated classification methods...
In this paper we study the use of confidence measures for an on-line handwriting recognizer. We investigate various and their integration in isolated word recognition system as well a sentence system. tasks, rejection mechanism is designed order to reject outputs recognizer that are possibly wrong, which case badly written words, out-of-vocabulary words or general drawing. allows rejecting parts decoded sentence.
The rise in the use of social media networks has increased prevalence cyberbullying, and time is paramount to reduce negative effects that derive from those behaviours on any platform. This paper aims study early detection problem a general perspective by carrying out experiments over two independent datasets (Instagram Vine), exclusively using users' comments. We used textual information comments baseline models (fixed, threshold, dual models) apply three different methods improving...
This paper investigates the cooperation of online and off-line handwriting word recognition systems. Our goal is to improve a mature system by, exploiting complementary information present in representation built from signal. After describing HMM based systems, we propose formal framework, which allows different strategies for combining two HMM. These schemes are then evaluated on UNIPEN database, both isolated character tasks.