Deep learning (DL) models have significantly transformed machine learning (ML), particularly with their prowess in classification tasks. However, these models struggle to differentiate between in-distribution (ID) and out-of-distribution (OOD) data at the testing phase. This challenge has curtailed their deployment in sensitive fields like biotechnology, where misidentifying OOD data, such as unclear or unknown bacterial genomic sequences, as known ID classes could lead to dire consequences. To address this, we propose an approach to make DL models OOD-sensitive by exploiting the configuration of the logit space embeddings, into the model’s decision-making process. Leveraging the effect observed in recent studies that there is minimal overlap between the embeddings of ID and OOD data, we use a density estimator to model the ID logit distribution based on the training data. This allows us to reliably flag data that do not match the ID distribution as OOD. Our methodology is designed to be independent of the specific data or model architecture and can seamlessly augment existing trained models without the need to expose them to OOD data. Testing our method on widely recognized image datasets, we achieve leading-edge results, including a substantial 10% enhancement in the area under the receiver operating characteristic curve (AUCROC) on the Google genome dataset.

Link prediction is a crucial task in many downstream applications of graph machine learning. To this end, Graph Neural Network (GNN) is a widely used technique for link prediction, mainly in transductive settings, where the goal is to predict missing links between existing nodes. However, many real-life applications require an inductive setting that accommodates for new nodes, coming into an existing graph. Thus, recently inductive link prediction has attracted considerable attention, and a multi-layer perceptron (MLP) is the popular choice of most studies to learn node representations. However, these approaches have limited expressivity and do not fully capture the graph’s structural signal. Therefore, in this work we propose LEAP, an inductive link prediction method based on LEArnable toPology augmentation. Unlike previous methods, LEAP models the inductive bias from both the structure and node features, and hence is more expressive. To the best of our knowledge, this is the first attempt to provide structural contexts for new nodes via learnable augmentation in inductive settings. Extensive experiments on seven real-world homogeneous and heterogeneous graphs demonstrates that LEAP significantly surpasses SOTA methods. The improvements are up to 22% and 17% in terms of AUC and average precision, respectively. The code and datasets are available on GitHub (1https://github.com/AhmedESamy/LEAP/).

Decentralized Learning (DL) enables privacy-preserving collaboration among organizations or users to enhance the performance of local deep learning models. However, model aggregation becomes challenging when client data is heterogeneous, and identifying compatible collaborators without direct data exchange remains a pressing issue. In this paper, we investigate the effectiveness of various similarity metrics in DL for identifying peers for model merging, conducting an empirical analysis across multiple datasets with distribution shifts. Our research provides insights into the performance of these metrics, examining their role in facilitating effective collaboration. By exploring the strengths and limitations of these metrics, we contribute to the development of robust DL methods.

The sets of possible heads and tails for relations in a Knowledge Graph and a type taxonomy of participating entities are important aspects of a Knowledge Graph and constitute a large portion of a Knowledge Graph’s ontology. Making the ontology explicit helps to ensure consistency of the knowledge represented in the graph and allows for less costly maintenance and update of graph content. However, Knowledge Graphs are often conveniently formed without an explicitly described ontology, using only (head, relation, tail)-tuples. For such graphs without predefined structure, we propose learning an ontology and a hierarchical type taxonomy from the graph itself, taking advantage of the reciprocity of entity types and sets of possible heads (relational domains) and tails (relational ranges). Experiments on real-world datasets validate our approach and demonstrate the promise for leveraging machine learning methodologies to efficiently generate taxonomies and ontologies jointly for Knowledge Graphs (Our implementation can be found at https://tinyurl.com/5n6sw6yj).

Federated Learning (FL) is a promising framework for distributed learning when data is private and sensitive. However, the state-of-the-art solutions in this framework are not optimal when data is heterogeneous and non-IID. We propose a practical and robust approach to personalization in FL that adjusts to heterogeneous and non-IID data by balancing exploration and exploitation of several global models. To achieve our aim of personalization, we use a Mixture of Experts (MoE) that learns to group clients that are similar to each other, while using the global models more efficiently. We show that our approach achieves an accuracy up to 29.78% better than the state-of-the-art and up to 4.38% better compared to a local model in a pathological non-IID setting, even though we tune our approach in the IID setting.

As the research community focuses on improving the reliability of deep learning, identifying out-of-distribution (OOD) data has become crucial. Detecting OOD inputs during test/prediction allows the model to account for discriminative features unknown to the model. This capability increases the model's reliability since this model provides a class prediction solely at incoming data similar to the training one. Although OOD detection is well-established in computer vision, it is relatively unexplored in other areas, like time series-based human activity recognition (HAR). Since uncertainty has been a critical driver for OOD in vision-based models, the same component has proven effective in time-series applications. In this work, we propose an ensemble-based temporal learning framework to address the OOD detection problem in HAR with time-series data. First, we define different types of OOD for HAR that arise from realistic scenarios. Then we apply our ensemble-based temporal learning framework incorporating uncertainty to detect OODs for the defined HAR workloads. This particular formulation also allows a novel approach to fall detection. We train our model on non-fall activities and detect falls as OOD. Our method shows state-of-The-Art performance in a fall detection task using much lesser data. Furthermore, the ensemble framework outperformed the traditional deep-learning method (our baseline) on the OOD detection task across all the other chosen datasets.

Self-supervised graph representation learning (SSGRL) is a representation learning paradigm used to reduce or avoid manual labeling. An essential part of SSGRL is graph data augmentation. Existing methods usually rely on heuristics commonly identified through trial and error and are effective only within some application domains. Also, it is not clear why one heuristic is better than another. Moreover, recent studies have argued against some techniques (e.g., dropout: that can change the properties of molecular graphs or destroy relevant signals for graph-based document classification tasks). In this study, we propose a novel data-driven SSGRL approach that automatically learns a suitable graph augmentation from the signal encoded in the graph (i.e., the nodes' predictive feature and topological information). We propose two complementary approaches that produce learnable feature and topological augmentations. The former learns multi-view augmentation of node features, and the latter learns a high-order view of the topology. Moreover, the augmentations are jointly learned with the representation. Our approach is general that it can be applied to homogeneous and heterogeneous graphs. We perform extensive experiments on node classification (using nine homogeneous and heterogeneous datasets) and graph property prediction (using another eight datasets). The results show that the proposed method matches or outperforms the SOTA SSGRL baselines and performs similarly to semi-supervised methods. The anonymised source code is available at https://github.com/AhmedESamy/dsgrl/

We study the problem of training personalized deep learning models in a decentralized peer-to-peer setting, focusing on the setting where data distributions differ between the clients and where different clients have different local learning tasks. We study both covariate and label shift, and our contribution is an algorithm which for each client finds beneficial collaborations based on a similarity estimate for the local task. Our method does not rely on hyperparameters which are hard to estimate, such as the number of client clusters, but rather continuously adapts to the network topology using soft cluster assignment based on a novel adaptive gossip algorithm. We test the proposed method in various settings where data is not independent and identically distributed among the clients. The experimental evaluation shows that the proposed method performs better than previous state-of-the-art algorithms for this problem setting, and handles situations well where previous methods fail.

Link prediction between two nodes is a critical task in graph machine learning. Most approaches are based on variants of graph neural networks (GNNs) that focus on transductive link prediction and have high inference latency. However, many real-world applications require fast inference over new nodes in inductive settings where no information on connectivity is available for these nodes. Thereby, node features provide an inevitable alternative in the latter scenario. To that end, we propose Graph2Feat, which enables inductive link prediction by exploiting knowledge distillation (KD) through the Student-Teacher learning framework. In particular, Graph2Feat learns to match the representations of a lightweight student multi-layer perceptron (MLP) with a more expressive teacher GNN while learning to predict missing links based on the node features, thus attaining both GNN's expressiveness and MLP's fast inference. Furthermore, our approach is general; it is suitable for transductive and inductive link predictions on different types of graphs regardless of them being homogeneous or heterogeneous, directed or undirected. We carry out extensive experiments on seven real-world datasets including homogeneous and heterogeneous graphs. Our experiments demonstrate that Graph2Feat significantly outperforms SOTA methods in terms of AUC and average precision in homogeneous and heterogeneous graphs. Finally, Graph2Feat has the minimum inference time compared to the SOTA methods, and 100x acceleration compared to GNNs. The code and datasets are available on GitHub1.

Cellular coverage quality estimation has been a critical task for self-organized networks. In real-world scenarios, deep-learning-powered coverage quality estimation methods cannot scale up to large areas due to little ground truth can be provided during network design & optimization. In addition, they fall short in producing expressive embeddings to adequately capture the variations of the cells' configurations. To deal with this challenge, we formulate the task in a graph representation and so that we can apply state-of-the-art graph neural networks, that show exemplary performance. We propose a novel training framework that can both produce quality cell configuration embeddings for estimating multiple KPIs, while we show it is capable of generalising to large (area-wide) scenarios given very few labeled cells. We show that our framework yields comparable accuracy with models that have been trained using massively labeled samples.