New Techniques in Deep Representation Learning

Thesis (Ph.D.)--University of Washington, 2016-06

Learning representations for Information Retrieval

La recherche d'informations s'intéresse, entre autres, à répondre à des questions comme: est-ce qu'un document est pertinent à une requête ? Est-ce que deux requêtes ou deux documents sont similaires ? Comment la similarité entre deux requêtes ou documents peut être utilisée pour améliorer l'estimation de la pertinence ? Pour donner réponse à ces questions, il est nécessaire d'associer chaque document et requête à des représentations interprétables par ordinateur. Une fois ces représentations estimées, la similarité peut correspondre, par exemple, à une distance ou une divergence qui opère dans l'espace de représentation. On admet généralement que la qualité d'une représentation a un impact direct sur l'erreur d'estimation par rapport à la vraie pertinence, jugée par un humain. Estimer de bonnes représentations des documents et des requêtes a longtemps été un problème central de la recherche d'informations. Le but de cette thèse est de proposer des nouvelles méthodes pour estimer les représentations des documents et des requêtes, la relation de pertinence entre eux et ainsi modestement avancer l'état de l'art du domaine. Nous présentons quatre articles publiés dans des conférences internationales et un article publié dans un forum d'évaluation. Les deux premiers articles concernent des méthodes qui créent l'espace de représentation selon une connaissance à priori sur les caractéristiques qui sont importantes pour la tâche à accomplir. Ceux-ci nous amènent à présenter un nouveau modèle de recherche d'informations qui diffère des modèles existants sur le plan théorique et de l'efficacité expérimentale. Les deux derniers articles marquent un changement fondamental dans l'approche de construction des représentations. Ils bénéficient notamment de l'intérêt de recherche dont les techniques d'apprentissage profond par réseaux de neurones, ou deep learning, ont fait récemment l'objet. Ces modèles d'apprentissage élicitent automatiquement les caractéristiques importantes pour la tâche demandée à partir d'une quantité importante de données. Nous nous intéressons à la modélisation des relations sémantiques entre documents et requêtes ainsi qu'entre deux ou plusieurs requêtes. Ces derniers articles marquent les premières applications de l'apprentissage de représentations par réseaux de neurones à la recherche d'informations. Les modèles proposés ont aussi produit une performance améliorée sur des collections de test standard. Nos travaux nous mènent à la conclusion générale suivante: la performance en recherche d'informations pourrait drastiquement être améliorée en se basant sur les approches d'apprentissage de représentations.

Learning representations for Information Retrieval

La recherche d'informations s'intéresse, entre autres, à répondre à des questions comme: est-ce qu'un document est pertinent à une requête ? Est-ce que deux requêtes ou deux documents sont similaires ? Comment la similarité entre deux requêtes ou documents peut être utilisée pour améliorer l'estimation de la pertinence ? Pour donner réponse à ces questions, il est nécessaire d'associer chaque document et requête à des représentations interprétables par ordinateur. Une fois ces représentations estimées, la similarité peut correspondre, par exemple, à une distance ou une divergence qui opère dans l'espace de représentation. On admet généralement que la qualité d'une représentation a un impact direct sur l'erreur d'estimation par rapport à la vraie pertinence, jugée par un humain. Estimer de bonnes représentations des documents et des requêtes a longtemps été un problème central de la recherche d'informations. Le but de cette thèse est de proposer des nouvelles méthodes pour estimer les représentations des documents et des requêtes, la relation de pertinence entre eux et ainsi modestement avancer l'état de l'art du domaine. Nous présentons quatre articles publiés dans des conférences internationales et un article publié dans un forum d'évaluation. Les deux premiers articles concernent des méthodes qui créent l'espace de représentation selon une connaissance à priori sur les caractéristiques qui sont importantes pour la tâche à accomplir. Ceux-ci nous amènent à présenter un nouveau modèle de recherche d'informations qui diffère des modèles existants sur le plan théorique et de l'efficacité expérimentale. Les deux derniers articles marquent un changement fondamental dans l'approche de construction des représentations. Ils bénéficient notamment de l'intérêt de recherche dont les techniques d'apprentissage profond par réseaux de neurones, ou deep learning, ont fait récemment l'objet. Ces modèles d'apprentissage élicitent automatiquement les caractéristiques importantes pour la tâche demandée à partir d'une quantité importante de données. Nous nous intéressons à la modélisation des relations sémantiques entre documents et requêtes ainsi qu'entre deux ou plusieurs requêtes. Ces derniers articles marquent les premières applications de l'apprentissage de représentations par réseaux de neurones à la recherche d'informations. Les modèles proposés ont aussi produit une performance améliorée sur des collections de test standard. Nos travaux nous mènent à la conclusion générale suivante: la performance en recherche d'informations pourrait drastiquement être améliorée en se basant sur les approches d'apprentissage de représentations.

Learning user and product distributed representations using a sequence model for sentiment analysis

In product reviews, it is observed that the distribution of polarity ratings over reviews written by different users or evaluated based on different products are often skewed in the real world. As such, incorporating user and product information would be helpful for the task of sentiment classification of reviews. However, existing approaches ignored the temporal nature of reviews posted by the same user or evaluated on the same product. We argue that the temporal relations of reviews might be potentially useful for learning user and product embedding and thus propose employing a sequence model to embed these temporal relations into user and product representations so as to improve the performance of document-level sentiment analysis. Specifically, we first learn a distributed representation of each review by a one-dimensional convolutional neural network. Then, taking these representations as pretrained vectors, we use a recurrent neural network with gated recurrent units to learn distributed representations of users and products. Finally, we feed the user, product and review representations into a machine learning classifier for sentiment classification. Our approach has been evaluated on three large-scale review datasets from the IMDB and Yelp. Experimental results show that: (1) sequence modeling for the purposes of distributed user and product representation learning can improve the performance of document-level sentiment classification; (2) the proposed approach achieves state-of-The-Art results on these benchmark datasets.

Learning representations for graph-structured socio-technical systems

The recent widespread use of social media platforms and web services has led to a vast amount of behavioral data that can be used to model socio-technical systems. A significant part of this data can be represented as graphs or networks, which have become the prevalent mathematical framework for studying the structure and the dynamics of complex interacting systems. However, analyzing and understanding these data presents new challenges due to their increasing complexity and diversity. For instance, the characterization of real-world networks includes the need of accounting for their temporal dimension, together with incorporating higher-order interactions beyond the traditional pairwise formalism. The ongoing growth of AI has led to the integration of traditional graph mining techniques with representation learning and low-dimensional embeddings of networks to address current challenges. These methods capture the underlying similarities and geometry of graph-shaped data, generating latent representations that enable the resolution of various tasks, such as link prediction, node classification, and graph clustering. As these techniques gain popularity, there is even a growing concern about their responsible use. In particular, there has been an increased emphasis on addressing the limitations of interpretability in graph representation learning. This thesis contributes to the advancement of knowledge in the field of graph representation learning and has potential applications in a wide range of complex systems domains. We initially focus on forecasting problems related to face-to-face contact networks with time-varying graph embeddings. Then, we study hyperedge prediction and reconstruction with simplicial complex embeddings. Finally, we analyze the problem of interpreting latent dimensions in node embeddings for graphs. The proposed models are extensively evaluated in multiple experimental settings and the results demonstrate their effectiveness and reliability, achieving state-of-the-art performances and providing valuable insights into the properties of the learned representations.

Representação de linhas de transmissão monofásicas por meio de variáveis de estado: comparação das soluções numérica e analítica

Coordenação de Aperfeiçoamento de Pessoal de Nível Superior (CAPES)

ImageLearn - Decoding Britain's Landscape

Abstract Ordnance Survey, our national mapping organisation, collects vast amounts of high-resolution aerial imagery covering the entirety of the country. Currently, photogrammetrists and surveyors use this to manually capture real-world objects and characteristics for a relatively small number of features. Arguably, the vast archive of imagery that we have obtained portraying the whole of Great Britain is highly underutilised and could be ‘mined’ for much more information. Over the last year the ImageLearn project has investigated the potential of "representation learning" to automatically extract relevant features from aerial imagery. Representation learning is a form of data-mining in which the feature-extractors are learned using machine-learning techniques, rather than being manually defined. At the beginning of the project we conjectured that representations learned could help with processes such as object detection and identification, change detection and social landscape regionalisation of Britain. This seminar will give an overview of the project and highlight some of our research results.

Apprendimento non supervisionato di rappresentazioni e legami di similarità tra eventi menzionati nella letteratura biomedica

L'estrazione automatica degli eventi biomedici dalla letteratura scientifica ha catturato un forte interesse nel corso degli ultimi anni, dimostrandosi in grado di riconoscere interazioni complesse e semanticamente ricche espresse all'interno del testo. Purtroppo però, esistono davvero pochi lavori focalizzati sull'apprendimento di embedding o di metriche di similarità per i grafi evento. Questa lacuna lascia le relazioni biologiche scollegate, impedendo l'applicazione di tecniche di machine learning che potrebbero dare un importante contributo al progresso scientifico. Approfittando dei vantaggi delle recenti soluzioni di deep graph kernel e dei language model preaddestrati, proponiamo Deep Divergence Event Graph Kernels (DDEGK), un metodo non supervisionato e induttivo in grado di mappare gli eventi all'interno di uno spazio vettoriale, preservando le loro similarità semantiche e strutturali. Diversamente da molti altri sistemi, DDEGK lavora a livello di grafo e non richiede nè etichette e feature specifiche per un determinato task, nè corrispondenze note tra i nodi. A questo scopo, la nostra soluzione mette a confronto gli eventi con un piccolo gruppo di eventi prototipo, addestra delle reti di cross-graph attention per andare a individuare i legami di similarità tra le coppie di nodi (rafforzando l'interpretabilità), e impiega dei modelli basati su transformer per la codifica degli attributi continui. Sono stati fatti ampi esperimenti su dieci dataset biomedici. Mostriamo che le nostre rappresentazioni possono essere utilizzate in modo efficace in task quali la classificazione di grafi, clustering e visualizzazione e che, allo stesso tempo, sono in grado di semplificare il task di semantic textual similarity. Risultati empirici dimostrano che DDEGK supera significativamente gli altri modelli che attualmente detengono lo stato dell'arte.

Computational Representation of Collaborative Learning Flow Patterns Using IMS Learning Design

The identification and integration of reusable and customizable CSCL (Computer Supported Collaborative Learning) may benefit from the capture of best practices in collaborative learning structuring. The authors have proposed CLFPs (Collaborative Learning Flow Patterns) as a way of collecting these best practices. To facilitate the process of CLFPs by software systems, the paper proposes to specify these patterns using IMS Learning Design (IMS-LD). Thus, teachers without technical knowledge can particularize and integrate CSCL tools. Nevertheless, the support of IMS-LD for describing collaborative learning activities has some deficiencies: the collaborative tools that can be defined in these activities are limited. Thus, this paper proposes and discusses an extension to IMS-LD that enables to specify several characteristics of the use of tools that mediate collaboration. In order to obtain a Unit of Learning based on a CLFP, a three stage process is also proposed. A CLFP-based Unit of Learning example is used to illustrate the process and the need of the proposed extension.

Visuotactile learning and body representation: an ERP study with rubber hands and rubber objects

We studied how the integration of seen and felt tactile stimulation modulates somatosensory processing, and investigated whether visuotactile integration depends on temporal contiguity of stimulation, and its coherence with a pre-existing body representation. During training, participants viewed a rubber hand or a rubber object that was tapped either synchronously with stimulation of their own hand, or in an uncorrelated fashion. In a subsequent test phase, somatosensory event-related potentials (ERPs) were recorded to tactile stimulation of the left or right hand, to assess how tactile processing was affected by previous visuotactile experience during training. An enhanced somatosensory N140 component was elicited after synchronous, compared with uncorrelated, visuotactile training, irrespective of whether participants viewed a rubber hand or rubber object. This early effect of visuotactile integration on somatosensory processing is interpreted as a candidate electrophysiological correlate of the rubber hand illusion that is determined by temporal contiguity, but not by pre-existing body representations. ERPmodulations were observed beyond 200msec post-stimulus, suggesting an attentional bias induced by visuotactile training. These late modulations were absent when the stimulation of a rubber hand and the participant’s own hand was uncorrelated during training, suggesting that pre-existing body representations may affect later stages of tactile processing.

Neural systems underlying learning and representation of global motion

We demonstrate performance-related changes in cortical and cerebellar activity. The largest learning-dependent changes were observed in the anterior lateral cerebellum, where the extent and intensity of activation correlated inversely with psychophysical performance. After learning had occurred (a few minutes), the cerebellar activation almost disappeared; however, it was restored when the subjects were presented with a novel, untrained direction of motion for which psychophysical performance also reverted to chance level. Similar reductions in the extent and intensity of brain activations in relation to learning occurred in the superior colliculus, anterior cingulate, and parts of the extrastriate cortex. The motion direction-sensitive middle temporal visual complex was a notable exception, where there was an expansion of the cortical territory activated by the trained stimulus. Together, these results indicate that the learning and representation of visual motion discrimination are mediated by different, but probably interacting, neuronal subsystems.

Representation and learning of visual information for pose recognition

Recovering position from sensor information is an important problem in mobile robotics, known as localisation. Localisation requires a map or some other description of the environment to provide the robot with a context to interpret sensor data. The mobile robot system under discussion is using an artificial neural representation of position. Building a geometrical map of the environment with a single camera and artificial neural networks is difficult. Instead it would be simpler to learn position as a function of the visual input. Usually when learning images, an intermediate representation is employed. An appropriate starting point for biologically plausible image representation is the complex cells of the visual cortex, which have invariance properties that appear useful for localisation. The effectiveness for localisation of two different complex cell models are evaluated. Finally the ability of a simple neural network with single shot learning to recognise these representations and localise a robot is examined.