431 research outputs found
Semi-supervised learning and applications : a game-theoretic perspective
No full textThis thesis is focused on semi-supervised learning (SSL) algorithms, a family of methods lying in between supervised and unsupervised learning. The main characteristic of SSL algorithms is that they exploit at the same time the structure of the data (their features) and the available labeling information to estimates the boundaries of the classes/clusters. For this reason, they are particularly suitable in a regime of scarcity of labeled data or in the cases whether the data annotation is expensive or time-consuming. Here, we will exploit a recent algorithm, rooted in the evolutionary game-theory, named “Graph Transduction Games”. The GTG algorithm explicitly models an SSL problem as a non-cooperative game where players represent the data and the strategies the possible labels. A player chooses a strategy and receives a payoff which is proportional to the choice of the other players and to their similarities. The game is iterated until all the players have chosen their best strategy, and no one has any incentive to change his/her choice. The final labeling is then a property that emerges by the players interactions, hence from the data. During the labeling process, the similarities between all the data are taken into account, creating a context in which similar points affect each other in deciding the final labeling assignment. The neighboring players (data), hence the context, help in situations in which intrinsic ambiguities in the data may lead to inconsistent class assignments. Within this thesis, the GTG algorithm and the context in which players are playing will be explored into applications like bioinformatics, natural language processing, computer vision, and pure machine learning problems
Context aware nonnegative matrix factorization clustering
No full textIn this article we propose a method to refine the clustering results obtained with the nonnegative matrix factorization (NMF) technique, imposing consistency constraints on the final labeling of the data. The research community focused its effort on the initialization and on the optimization part of this method, without paying attention to the final cluster assignments. We propose a game theoretic framework in which each object to be clustered is represented as a player, which has to choose its cluster membership. The information obtained with NMF is used to initialize the strategy space of the players and a weighted graph is used to model the interactions among the players. These interactions allow the players to choose a cluster which is coherent with the clusters chosen by similar players, a property which is not guaranteed by NMF, since it produces a soft clustering of the data. The results on common benchmarks show that our model is able to improve the performances of many NMF formulations
Hypergraph Isomorphism Using Association Hypergraphs
No full textAssociation graphs represent a classical tool to deal with the graph matching problem and recently the idea has been generalized to the case of hypergraphs. In this article, the potential of this approach is explored. The proposed framework uses a class of dynamical systems derived from the Baum-Eagon inequality in order to find the maximum (maximal) clique in the association hypergraph, that corresponds to the maximum (maximal) isomorphism between the hypergraphs to be matched. The proposed approach has extensively been tested with experiments on a large synthetic dataset, including hypergraphs of different cardinalities, order and connectivities. In particular the isomorphism version of the problem has been analyzed. The results obtained are impressive in terms of correctness, thus showing that, despite its simplicity, the Baum-Eagon dynamics has an outstanding capacity of finding globally optimal solutions and solving the hypergraph isomorphism problem.Association graphs represent a classical tool to deal with the graph matching problem and recently the idea has been generalized to the case of hypergraphs. In this article, the potential of this approach is explored. The proposed framework uses a class of dynamical systems derived from the Baum-Eagon inequality in order to find the maximum (maximal) clique in the association hypergraph, that corresponds to the maximum (maximal) isomorphism between the hypergraphs to be matched. The proposed approach has extensively been tested with experiments on a large synthetic dataset, including hypergraphs of different cardinalities, order and connectivities. In particular the isomorphism version of the problem has been analyzed. The results obtained are impressive in terms of correctness, thus showing that, despite its simplicity, the Baum-Eagon dynamics has an outstanding capacity of finding globally optimal solutions and solving the hypergraph isomorphism problem. (C) 2019 Elsevier B.V. All rights reserved
Chapter 3 - Group Detection and Tracking Using Sociological Features
No full textThis chapter describes the most common features and definitions from the sociological science used to detect and track groups of people that are interacting. The necessity of having reliable algorithms to cope with these problems is gaining increasing interest, especially in the fields related to security and video surveillance. Answering the question of “who is present and with whom he/she is interacting in a scene?” is nowadays of utmost importance. Other domains require having good algorithms to face these problems, for example, activity recognition, social robotics, and automatic behavior analysis. The success of detection and tracking algorithms relies on the engineering of the features. In this context, the literature of sociological sciences gives us a set of well-established assumptions and constraints to create more reliable and plausible features and detection algorithms. In this chapter we will describe the existing features of the following two categories: the low-level category used to determine the spatial properties of each person in a scene (person position and head/body orientation), and the high-level category that agglomerates or uses the low-level features to implement sociological and biological definitions (frustum of visual attention). We will see how these features are used by the popular methods of group detection, such as game theory-based and probabilistic approaches. Finally, we will analyze a tracking model that can be integrated with the analyzed features and the described detection methods. The experimental part provides a comprehensive comparison of the performances of different algorithms to detect and track groups on standard and publicly available benchmarks
Speaker clustering using dominant sets
No full textSpeaker clustering is the task of forming speaker-specific groups based on a set of utterances. In this paper, we address this task by using Dominant Sets (DS). DS is a graphbased clustering algorithm with interesting properties that fits well to our problem and has never been applied before to speaker clustering. We report on a comprehensive set of experiments on the TIMIT dataset against standard clustering techniques and specific speaker clustering methods. Moreover, we compare performances under different features by using ones learned via deep neural network directly on TIMIT and other ones extracted from a pre-trained VGGVox net. To asses the stability, we perform a sensitivity analysis on the free parameters of our method, showing that performance is stable under parameter changes. The extensive experimentation carried out confirms the validity of the proposed method, reporting state-of-the-art results under three different standard metrics. We also report reference baseline results for speaker clustering on the entire TIMIT dataset for the first time
On association graph techniques for hypergraph matching
No full textAssociation graph techniques represent a classical approach to tackle the graph matching problem and recently the idea has been generalized to the case of hypergraphs. In this paper, we explore the potential of this approach in conjunction with a class of dynamical systems derived from the Baum-Eagon inequality. In particular, we focus on the pure isomorphism case and show, with extensive experiments on a large synthetic dataset, that despite its simplicity the Baum-Eagon dynamics does an excellent job at finding globally optimal solutions
Chapter 12 - Detecting conversational groups in images using clustering games
No full textDetecting groups of people in images is of great importance in many different contexts such as video surveillance, activity recognition and social robotics. Standing conversational groups (a.k.a. F-formations) represent a well-studied class of social interactions which play a prominent role in everyday human interactions. An F-formation is a type of social aggregation occurring when two or more persons are engaged in a conversation, of the type taking place, e.g., at a cocktail party or at a coffee break. Essentially, an F-formation defines a set of constraints on how the interactants have to be mutually located and oriented, and also the plausible zone in which the interactions may occur. In this chapter, we will describe an approach to detecting groups of conversing people in images based on game theory. The approach improves upon existing methods by building a stochastic model of social attention which captures the likelihood that two individuals take part in a conversation. This is used to derive a payoff function between detected individuals which defines the underlying clustering game. As it turns out, the stable equilibrium points of this game represent maximally coherent groups, and we used simple and effective evolutionary game dynamics to extract them. Extensive experimental results on several publicly available benchmark datasets demonstrate the superiority of the proposed approach over standard methods
MEMEX_KG: Knowledge Graphs about the cities of Lisbon, Barcelona and Paris
No full textKnowledge Graphs about the cities of Lisbon, Barcelona and Paris. The datasets are used in the following paper:
Mohamed, Hebatallah A., Sebastiano Vascon, Feliks Hibraj, Stuart James, Diego Pilutti, Alessio Del Bue, and Marcello Pelillo. "Geolocation of Cultural Heritage using Multi-View Knowledge Graph Embedding." arXiv preprint arXiv:2209.03638 (2022)
Two-dimensional impurity imaging in polar ice cores sparks new demand for automated image analysis
No full textDue to its micron-scale resolution and micro-destructiveness, laser ablation inductively coupled plasma mass spectrometry (LA-ICP-MS) is especially suited for the analysis of the oldest and highly thinned sections of polar ice cores. State-of-the-art 2D elemental imaging by LA-ICP-MS has great potential for investigating the location of impurities on the ice sample surface and is crucial to avoid misinterpretation of ultra-fine resolution signals. The impurity imaging with LA-ICP-MS comprises several millions of laser shots fired over just a few square mm. This technique combines new chemical images with visual analysis and, in so doing raises new questions that may be answered through techniques in automated image analysis and computer vision. As an illustration of this new frontier, a selected set of key problems is presented, with first examples of how automated image analysis techniques can help solving them. This concerns the relationship between impurity localization and the grain boundary network as well as the paleoclimate significance of single line profiles along the main core axis. Ultimately, this demonstrates that it is the right time to spark an intensified exchange among the two scientific communities of computer vision and ice core science
Transductive Label Augmentation for Improved Deep Network Learning
No full textA major impediment to the application of deep learning to real-world problems is the scarcity of labeled data. Small training sets are in fact of no use to deep networks as, due to the large number of trainable parameters, they will very likely be subject to overfitting phenomena. On the other hand, the increment of the training set size through further manual or semi-automatic labellings can be costly, if not possible at times. Thus, the standard techniques to address this issue are transfer learning and data augmentation, which consists of applying some sort of 'transformation' to existing labeled instances to let the training set grow in size. Although this approach works well in applications such as image classification, where it is relatively simple to design suitable transformation operators, it is not obvious how to apply it in more structured scenarios. Motivated by the observation that in virtually all application domains it is easy to obtain unlabeled data, in this paper we take a different perspective and propose a label augmentation approach. We start from a small, curated labeled dataset and let the labels propagate through a larger set of unlabeled data using graph transduction techniques. This allows us to naturally use (second-order) similarity information which resides in the data, a source of information which is typically neglected by standard augmentation techniques. In particular, we show that by using known game theoretic transductive processes we can create larger and accurate enough labeled datasets which use results in better trained neural networks. Preliminary experiments are reported which demonstrate a consistent improvement over standard image classification datasets
- …
