Graph Convolutional Networks (GCNs) draws significant attention due to its power of representation learning on graphs. Recent works developed frameworks to train deep GCNs. Such works show impressive results in tasks like point cloud classification and segmentation, and protein interaction prediction. While for large-scale graphs, doing full-batch training by GCNs is still challenging especially when GCNs go deeper. By fully analyzing a clustering-based mini-batch training algorithm ClusterGCN, we propose random partition which is a more efficient and effective method to implement mini-batch training. Besides, selecting different permutation invariance function (such as max, mean or add) for neighbors’ information aggregation will result in every different results. Therefore, we propose to alleviate it by introducing a novel Generalized Aggregation Function. In this thesis, I analyze the drawbacks caused by ClusterGCN and discuss about its limits. I further compare the performance of ClusterGCN with random partition and the final experimental results show that simple random partition outperforms ClusterGCN with very obvious advantageous for node property prediction task. For the techniques which are commonly used to make GCNs go deeper, I demonstrate a better way of applying residual connections (pre-activation) to stack more layers for GCNs. Last, I show the complete work of training deeper GCNs with generalized aggregators and display the promising results over several datasets from the Open Graph Benchmark (OGB).

Text classification is a fundamental natural language processing task. However, in real-world applications, class distributions are usually skewed, e.g., due to inherent class imbalance. In addition, the task difficulty changes based on the underlying language. When rich morphological structure and high ambiguity are exhibited, natural language understanding can become challenging. For example, Arabic, ranked the fifth most widely used language, has a rich morphological structure and high ambiguity that result from Arabic orthography. Thus, Arabic natural language processing is challenging. Several studies employ Long Short- Term Memory (LSTM) networks and Convolutional Neural Networks (CNNs), but Graph Convolutional Networks (GCNs) have not yet been investigated for the task. Sequence- based models can successfully capture semantics in local consecutive text sequences. On the other hand, graph-based models can preserve global co-occurrences that capture non- consecutive and long-distance semantics. A text representation approach that combines local and global information can enhance performance in practical class imbalance text classification scenarios. Yet, multi-view graph-based text representations have received limited attention. In this research, first we introduce Multi-view Minority Class Text Graph Convolutional Network (MMCT-GCN), a transductive multi-view text classification model that captures textual graph representations for the minority class alongside sequence-based text representations. Experimental results show that MMCT-GCN obtains consistent improvements over baselines. Second, we develop an Arabic Bidirectional Encoder Representations from Transformers (BERT) Graph Convolutional Network (AraBERT-GCN), a hybrid model that combines the large-scale pre-trained models that encode the local context and semantics alongside graph-based features that are capable of extracting the global word co-occurrences in non-consecutive extended semantics by only one or two hops. Experimental results show that AraBERT-GCN outperforms the state-of-the-art (SOTA) on our Arabic text datasets. Finally, we propose an Arabic Multidimensional Edge Graph Convolutional Network (AraMEGraph) designed for text classification that encapsulates richer and context-aware representations of word and phrase relationships, thus mitigating the impact of the complexity and ambiguity of the Arabic language. / Doctor of Philosophy / The text classification task is an important step in understanding natural language. However, this task has many challenges, such as uneven data distributions and language difficulty. For example, Arabic is the fifth most spoken language. It has many different word forms and meanings, which can make things harder to understand. Long Short-Term Memory (LSTM) networks and Convolutional Neural Networks (CNNs) are widely utilized for text classification. However, another kind of network called graph convolutional network (GCN) has yet to be explored for this task. Graph-based models keep track of how words are connected, even if they are not right next to each other in a sentence. This helps with better understanding the meaning of words. On the other hand, sequence-based models do well in understanding the meaning of words that are right next to each other. Mixing both types of information in text understanding can work better, especially when dealing with unevenly distributed data. In this research, we introduce a new text classification method called Multi-view Minority Class Text Graph Convolutional Network (MMCT-GCN). This model looks at text from different angles and combines information from graphs and sequence-based models. Our experiments show that this model performs better than other ones proposed in the literature. Additionally, we propose an Arabic BERT Graph Convolutional Network (AraBERT-GCN). It combines pre-trained models that understand words in context and graph features that look at how words relate to each other globally. This helps AraBERT- GCN do better than other models when working with Arabic text. Finally, we develop a special network called Arabic Multidimensional Edge Graph Convolutional Network (AraMEGraph) for Arabic text. It is designed to better understand Arabic and classify text more accurately. We do this by adding special edge features with multiple dimensions to help the network learn the relationships between words and phrases.

Neural architecture search is a popular method for automating architecture design. Bayesian optimization is a widely used approach for hyper-parameter optimization and can estimate a function with limited samples. However, Bayesian optimization methods are not preferred for architecture search as it expects vector inputs while graphs are high dimensional data. This thesis presents a Bayesian approach with Gaussian priors that use graph kernels specifically targeted to work in the higherdimensional graph space. We implemented three different graph kernels and show that on the NAS-Bench-101 dataset, an untrained graph convolutional network kernel outperforms previous methods significantly in terms of the best network found and the number of samples required to find it. We follow the AutoML guidelines to make this work reproducible. / Neural arkitektur sökning är en populär metod för att automatisera arkitektur design. Bayesian-optimering är ett vanligt tillvägagångssätt för optimering av hyperparameter och kan uppskatta en funktion med begränsade prover. Bayesianska optimeringsmetoder är dock inte att föredra för arkitektonisk sökning eftersom vektoringångar förväntas medan grafer är högdimensionella data. Denna avhandling presenterar ett Bayesiansk tillvägagångssätt med gaussiska prior som använder grafkärnor som är särskilt fokuserade på att arbeta i det högre dimensionella grafutrymmet. Vi implementerade tre olika grafkärnor och visar att det på NASBench- 101-data, till och med en otränad Grafkonvolutionsnätverk-kärna, överträffar tidigare metoder när det gäller det bästa nätverket som hittats och antalet prover som krävs för att hitta det. Vi följer AutoML-riktlinjerna för att göra detta arbete reproducerbart.

Graph convolutional networks (GCNs) showed promising results in learning from point cloud data. Applications of GCNs include point cloud classification, point cloud segmentation, point cloud upsampling, and more. Recently, the introduction of Deep Graph Convolutional Networks (DeepGCNs) allowed GCNs to go deeper, and thus resulted in better graph learning while avoiding the vanishing gradient problem in GCNs. By adapting impactful methods from convolutional neural networks (CNNs) such as residual connections, dense connections, and dilated convolutions, DeepGCNs allowed GCNs to learn better from non-Euclidean data. In addition, deep learning methods proved very effective in the task of point cloud upsampling. Unlike traditional optimization-based methods, deep learning-based methods to point cloud upsampling does not rely on priors nor hand-crafted features to learn how to upsample point clouds. In this thesis, I discuss the impact and show the performance results of DeepGCNs in the task of point cloud part segmentation on PartNet dataset. I also illustrate the significance of using GCNs as upsampling modules in the task of point cloud upsampling by introducing two novel upsampling modules: Multi-branch GCN and Clone GCN. I show quantitatively and qualitatively the performance results of our novel and versatile upsampling modules when evaluated on a new proposed standardized dataset: PU600, which is the largest and most diverse point cloud upsampling dataset currently in the literature.

要旨ファイルを差し替え（2023-01-23） / 京都大学 / 新制・課程博士 / 博士(薬学) / 甲第23144号 / 薬博第844号 / 新制||薬||242(附属図書館) / 京都大学大学院薬学研究科薬学専攻 / (主査)教授 高須 清誠, 教授 石濱 泰, 教授 大野 浩章 / 学位規則第4条第1項該当 / Doctor of Pharmaceutical Sciences / Kyoto University / DFAM

With increasing data and technology, multiple types of data from the same set of nodes have been generated. Since each data modality contains a unique aspect of the underlying mechanisms, multiple datatypes are integrated. In addition to multiple datatypes, networks are important to store information representing associations between entities such as genes of a protein-protein interaction network and authors of a citation network. Recently, some advanced approaches to graph-structured data leverage node associations and features simultaneously, called Graph Neural Network (GNN), but they have limitations for integrative approaches. The overall aim of this dissertation is to integrate multiple data modalities on graph-structured data to infer some context-specific gene regulation and predict outcomes of interest. To this end, first, we introduce a computational tool named CRINET to infer genome-wide competing endogenous RNA (ceRNA) networks. By integrating multiple data properly, we had a better understanding of gene regulatory circuitry addressing important drawbacks pertaining to ceRNA regulation. We tested CRINET on breast cancer data and found that ceRNA interactions and groups were significantly enriched in the cancer-related genes and processes. CRINET-inferred ceRNA groups supported the studies claiming the relation between immunotherapy and cancer. Second, we present SUPREME, a node classification framework, by comprehensively analyzing multiple data and associations between nodes with graph convolutions on multiple networks. Our results on survival analysis suggested that SUPREME could demystify the characteristics of classes with proper utilization of multiple data and networks. Finally, we introduce an attention-aware fusion approach, called GRAF, which fuses multiple networks and utilizes attention mechanisms on graph-structured data. Utilization of learned node- and association-level attention with network fusion allowed us to prioritize the edges properly, leading to improvement in the prediction results. Given the findings of all three tools and their outperformance over state-of-the-art methods, the proposed dissertation shows the importance of integrating multiple types of data and the exploitation of multiple graph structured data.

There exist a large number of rare genetic diseases in humans. Our knowledge of the specific gene variants whose presence in the genome of a person predisposes them towards developing a disease, called gene associations, is incomplete. Computational tools which can predict genes which may be associated with a rare disease have great utility in healthcare. However, a majority of existing prediction algorithms require a set of already known "seed genes'' to further discover novel associations for a disease. This drawback becomes more serious for rare genetic diseases, since a large proportion do not have any known gene associations. In this work, we develop an approach for disease-gene association prediction that overcomes the reliance on seed genes. Our approach uses the similarity of the observable biological characteristics of diseases (i.e., phenotypes) along with a global map of direct and indirect human protein interactions, to transfer associations from diseases whose gene associations have been discovered to diseases with no known gene associations. We formulate disease-gene association prediction over a multimodal network of diseases and genes, and develop an approach based on graph convolutional networks. We show how our model design considerations impact prediction performance. We demonstrate that our approach outperforms simpler graph machine learning and traditional machine learning approaches, as well as a competitive network propagation based approach for the task of predicting disease-gene associations. / Master of Science / There exist a large number of rare genetic diseases in humans. Our knowledge of the specific gene variants whose presence in the genome of a person predisposes them towards developing a disease, called gene associations, is incomplete. Computational tools which can predict genes which may be associated with a rare disease have great utility in healthcare. However, a majority of existing prediction algorithms require a set of already known "seed genes'' to further discover novel associations for a disease. This drawback becomes more serious for rare genetic diseases, since a large proportion do not have any known gene associations. In this work, we develop an approach for disease-gene association prediction that overcomes the reliance on seed genes. Our approach uses the similarity of the observable biological characteristics of diseases (i.e. disease phenotypes) along with a global map of direct and indirect human protein interactions, to transfer gene associations from diseases whose gene associations have been discovered, to diseases with no known associations. We implement an approach based on the field of graph machine learning, namely graph convolutional networks, to predict the genes associated with rare genetic diseases. We show how our predictor performs, compared to other approaches, and analyze some of the choices made in the design of the predictor, along with some properties of the outputs of our predictor.

A gene regulatory network (GRN) records the interactions among transcription factors and their target genes. GRNs are useful to study how transcription factors (TFs) control gene expression as cells transition between states during differentiation and development. Scientists usually construct GRNs by careful examination and study of the literature. This process is slow and painstaking and does not scale to large networks. In this thesis, we study the problem of inferring GRNs automatically from gene expression data. Recent data-driven approaches to infer GRNs increasingly rely on single-cell level RNA-sequencing (scRNA-seq) data. Most of these methods rely on unsupervised or association based strategies, which cannot leverage known regulatory interactions by design. To facilitate supervised learning, we propose a novel graph convolutional neural network (GCN) based autoencoder to infer new regulatory edges from a known GRN and scRNA-seq data. As the name suggests, a GCN-based autoencoder consists of an encoder that learns a low-dimensional embedding of the nodes (genes) in the input graph (the GRN) through a series of graph convolution operations and a decoder that aims to reconstruct the original graph as accurately as possible. We investigate several GCN-based architectures to determine the ideal encoder-decoder combination for GRN reconstruction. We systematically study the performance of these and other supervised learning methods on different mouse and human scRNA-seq datasets for two types of evaluation. We demonstrate that our GCN-based approach substantially outperforms traditional machine learning approaches. / Master of Science / In multi-cellular living organisms, stem cells differentiate into multiple cell types. Proteins called transcription factors (TFs) control the activity of genes to effect these transitions. It is possible to represent these interactions abstractly using a gene regulatory network (GRN). In a GRN, each node is a TF or a gene and each edge connects a TF to a gene or TF that it controls. New high-throughput technologies that can measure gene expression (activity) in individual cells provide rich data that can be used to construct GRNs. In this thesis, we take advantage of recent advances in the field of machine learning to develop a new computational method for computationally constructing GRNs. The distinguishing property of our technique is that it is supervised, i.e., it uses experimentally-known interactions to infer new regulatory connections. We investigate several variations of this approach to reconstruct a GRN as close to the original network as possible. We analyze and provide a rationale for the decisions made in designing, evaluating, and choosing the characteristics of our predictor. We show that our predictor has a reconstruction accuracy that is superior to other supervised-learning approaches.

Knowledge Graphs are fascinating concepts in machine learning as they can hold usefully structured information in the form of entities and their relations. Despite the valuable applications of such graphs, most knowledge bases remain incomplete. This missing information harms downstream applications such as information retrieval and opens a window for research in statistical relational learning tasks such as node classification and link prediction. This work proposes a deep learning framework based on existing relational convolutional (R-GCN) layers to learn on highly multi-relational data characteristic of realistic knowledge graphs for node property classification tasks. We propose a deep and improved variant, Deep-RGCNs, with dense and residual skip connections between layers. These skip connections are known to be very successful with popular deep CNN-architectures such as ResNet and DenseNet. In our experiments, we investigate and compare the performance of Deep-RGCN with different baselines on multi-relational graph benchmark datasets, AIFB and MUTAG, and show how the deep architecture boosts the performance in the task of node property classification. We also study the training performance of Deep-RGCNs (with N layers) and discuss the gradient vanishing and over-smoothing problems common to deeper GCN architectures.

Instruction Scheduling is a back-end compiler optimisation technique that can provide significant performance gains. It refers to ordering instructions in a particular order to reduce latency for processors with instruction-level parallelism. At the present typical compilers use heuristics to perform instruction scheduling and solve other related non-polynomial complete problems. This thesis aims to present a machine learning-based approach to challenge heuristic methods concerning performance. In this thesis, a novel reinforcement learning (RL) based model for the instruction scheduling problem is developed including modelling features of processors such as forwarding, resource utilisation and treatment of the action space. An efficient optimal scheduler is presented to be used for an optimal schedule length based reward function, however, this is not used in the final results as a heuristic based reward function was deemed to be sufficient and faster to compute. Furthermore, an RL agent that interacts with the model of the problem is presented using three different types of graph neural networks for the state processing: graph conventional networks, graph attention networks, and graph attention based on the work of Lee et al. A simple two-layer neural network is also used for generating embeddings for the resource utilisation stages. The proposed solution is validated against the modelled environment and favourable but not significant improvements were found compared to the most common heuristic method. Furthermore, it was found that having embeddings relating to resource utilisation was very important for the explained variance of the RL models. Additionally, a trained model was tested in an actual compiler, however, no informative results were found likely due to register allocation or other compiler stages that occur after instruction scheduling. Future work should include improving the scalability of the proposed solution. / Instruktionsschemaläggning är en optimeringsteknik för kompilatorer som kan ge betydande prestandavinster. Det handlar om att ordna instruktioner i en viss ordning för att minska latenstiden för processorer med parallellitet på instruktionsnivå. För närvarande använder vanliga kompilatorer heuristiker för att utföra schemaläggning av instruktioner och lösa andra relaterade ickepolynomiala kompletta problem. Denna avhandling syftar till att presentera en maskininlärningsbaserad metod för att utmana heuristiska metoder när det gäller prestanda. I denna avhandling utvecklas en ny förstärkningsinlärningsbaserad (RL) modell för schemaläggning av instruktioner, inklusive modellering av processorns egenskaper såsom vidarebefordran, resursutnyttjande och behandling av handlingsutrymmet. En effektiv optimal schemaläggare presenteras för att eventuellt användas för belöningsfunktionen, men denna används inte i de slutliga resultaten. Dessutom presenteras en RL-agent som interagerar med problemmodellen och använder tre olika typer av grafneurala nätverk för tillståndsprocessering: grafkonventionella nätverk, grafuppmärksamhetsnätverk och grafuppmärksamhet baserat på arbetet av Lee et al. Ett enkelt neuralt nätverk med två lager används också för att generera inbäddningar för resursanvändningsstegen. Den föreslagna lösningen valideras mot den modellerade miljön och gynnsamma men inte signifikanta förbättringar hittades jämfört med den vanligaste heuristiska metoden. Dessutom visade det sig att det var mycket viktigt för den förklarade variansen i RL-modellerna att ha inbäddningar relaterade till resursutnyttjande. Dessutom testades en tränad modell i en verklig kompilator, men inga informativa resultat hittades, sannolikt på grund av registerallokering eller andra kompilatorsteg som inträffar efter schemaläggning av instruktioner. Framtida arbete bör inkludera att förbättra skalbarheten hos den föreslagna lösningen.