In this work, we benchmark a variety of single- and multi-task graph neural network (GNN) models against lower-bar and higher-bar traditional machine learning approaches employing human engineered molecular features. We consider four GNN variants -- Graph Convolutional Network (GCN), Graph Attention Network (GAT), Message Passing Neural Network (MPNN), and Attentive Fingerprint (AttentiveFP). So far deep learning models have been primarily benchmarked using lower-bar traditional models solely based on fingerprints, while more realistic benchmarks employing fingerprints, whole-molecule descriptors and predictions from other related endpoints (e.g., LogD7.4) appear to be scarce for industrial ADME datasets. In addition to time-split test sets based on Genentech data, this study benefits from the availability of measurements from an external chemical space (Roche data). We identify GAT as a promising approach to implementing deep learning models. While all GNN models significantly outperform lower-bar benchmark traditional models solely based on fingerprints, only GATs seem to offer a small but consistent improvement over higher-bar benchmark traditional models. Finally, the accuracy of in vitro assays from different laboratories predicting the same experimental endpoints appears to be comparable with the accuracy of GAT single-task models, suggesting that most of the observed error from the models is a function of the experimental error propagation.
The goal of meta-learning is to learn to adapt to a new task with only a few labeled examples. To tackle this problem in NLP, we propose $\textit{in-context tuning}$, which recasts adaptation and prediction as a simple sequence prediction problem: to form the input sequence, we concatenate the task instruction, the labeled examples, and the target input to predict; to meta-train the model to learn from in-context examples, we fine-tune a pre-trained language model (LM) to predict the target label from the input sequences on a collection of tasks. We benchmark our method on two collections of text classification tasks: LAMA and BinaryClfs. Compared to first-order MAML which adapts the model with gradient descent, our method better leverages the inductive bias of LMs to perform pattern matching, and outperforms MAML by an absolute $6\%$ AUC ROC score on BinaryClfs, with increasing advantage w.r.t. model size. Compared to non-fine-tuned in-context learning (i.e. prompting a raw LM), in-context tuning directly learns to learn from in-context examples. On BinaryClfs, in-context tuning improves the average AUC-ROC score by an absolute $10\%$, and reduces the variance with respect to example ordering by 6x and example choices by 2x.
Consistency training is a popular method to improve deep learning models in computer vision and natural language processing. Graph neural networks (GNNs) have achieved remarkable performance in a variety of network science learning tasks, but to date no work has studied the effect of consistency training on large-scale graph problems. GNNs scale to large graphs by minibatch training and subsample node neighbors to deal with high degree nodes. We utilize the randomness inherent in the subsampling of neighbors and introduce a novel consistency training method to improve accuracy. For a target node we generate different neighborhood expansions, and distill the knowledge of the average of the predictions to the GNN. Our method approximates the expected prediction of the possible neighborhood samples and practically only requires a few samples. We demonstrate that our training method outperforms standard GNN training in several different settings, and yields the largest gains when label rates are low.
We aim to identify how different components in the KD pipeline affect the resulting performance and how much the optimal KD pipeline varies across different datasets/tasks, such as the data augmentation policy, the loss function, and the intermediate representation for transferring the knowledge between teacher and student. To tease apart their effects, we propose Distiller, a meta KD framework that systematically combines a broad range of techniques across different stages of the KD pipeline, which enables us to quantify each component's contribution. Within Distiller, we unify commonly used objectives for distillation of intermediate representations under a universal mutual information (MI) objective and propose a class of MI-$\alpha$ objective functions with better bias/variance trade-off for estimating the MI between the teacher and the student. On a diverse set of NLP datasets, the best Distiller configurations are identified via large-scale hyperparameter optimization. Our experiments reveal the following: 1) the approach used to distill the intermediate representations is the most important factor in KD performance, 2) among different objectives for intermediate distillation, MI-$\alpha$ performs the best, and 3) data augmentation provides a large boost for small training datasets or small student networks. Moreover, we find that different datasets/tasks prefer different KD algorithms, and thus propose a simple AutoDistiller algorithm that can recommend a good KD pipeline for a new dataset.
Many real-world graphs involve different types of nodes and relations between nodes, being heterogeneous by nature. The representation learning of heterogeneous graphs (HGs) embeds the rich structure and semantics of such graphs into a low-dimensional space and facilitates various data mining tasks, such as node classification, node clustering, and link prediction. In this paper, we propose a self-supervised method that learns HG representations by relying on knowledge exchange and discovery among different HG structural semantics (meta-paths). Specifically, by maximizing the mutual information of meta-path representations, we promote meta-path information fusion and consensus, and ensure that globally shared semantics are encoded. By extensive experiments on node classification, node clustering, and link prediction tasks, we show that the proposed self-supervision both outperforms and improves competing methods by 1% and up to 10% for all tasks.
Course selection is challenging for students in higher educational institutions. Existing course recommendation systems make relevant suggestions to the students and help them in exploring the available courses. The recommended courses can influence students' choice of degree program, future employment, and even their socioeconomic status. This paper focuses on identifying and alleviating biases that might be present in a course recommender system. We strive to promote balanced opportunities with our suggestions to all groups of students. At the same time, we need to make recommendations of good quality to all protected groups. We formulate our approach as a multi-objective optimization problem and study the trade-offs between equal opportunity and quality. We evaluate our methods using both real-world and synthetic datasets. The results indicate that we can considerably improve fairness regarding equality of opportunity, but we will introduce some quality loss. Out of the four methods we tested, GHC-Inc and GHC-Tabu are the best performing ones with different advantageous characteristics.
Collaborative recommendation approaches based on nearest-neighbors are still highly popular today due to their simplicity, their efficiency, and their ability to produce accurate and personalized recommendations. This chapter offers a comprehensive survey of neighborhood-based methods for the item recommendation problem. It presents the main characteristics and benefits of such methods, describes key design choices for implementing a neighborhood-based recommender system, and gives practical information on how to make these choices. A broad range of methods is covered in the chapter, including traditional algorithms like k-nearest neighbors as well as advanced approaches based on matrix factorization, sparse coding and random walks.
Graph Neural Networks (GNNs) bring the power of deep representation learning to graph and relational data and achieve state-of-the-art performance in many applications. GNNs compute node representations by taking into account the topology of the node's ego-network and the features of the ego-network's nodes. When the nodes do not have high-quality features, GNNs learn an embedding layer to compute node embeddings and use them as input features. However, the size of the embedding layer is linear to the product of the number of nodes in the graph and the dimensionality of the embedding and does not scale to big data and graphs with hundreds of millions of nodes. To reduce the memory associated with this embedding layer, hashing-based approaches, commonly used in applications like NLP and recommender systems, can potentially be used. However, a direct application of these ideas fails to exploit the fact that in many real-world graphs, nodes that are topologically close will tend to be related to each other (homophily) and as such their representations will be similar. In this work, we present approaches that take advantage of the nodes' position in the graph to dramatically reduce the memory required, with minimal if any degradation in the quality of the resulting GNN model. Our approaches decompose a node's embedding into two components: a position-specific component and a node-specific component. The position-specific component models homophily and the node-specific component models the node-to-node variation. Extensive experiments using different datasets and GNN models show that our methods are able to reduce the memory requirements by 88% to 97% while achieving, in nearly all cases, better classification accuracy than other competing approaches, including the full embeddings.
This paper aims to unify spatial dependency and temporal dependency in a non-Euclidean space while capturing the inner spatial-temporal dependencies for spatial-temporal graph data. For spatial-temporal attribute entities with topological structure, the space-time is consecutive and unified while each node's current status is influenced by its neighbors' past states over variant periods of each neighbor. Most spatial-temporal neural networks study spatial dependency and temporal correlation separately in processing, gravely impaired the space-time continuum, and ignore the fact that the neighbors' temporal dependency period for a node can be delayed and dynamic. To model this actual condition, we propose TraverseNet, a novel spatial-temporal graph neural network, viewing space and time as an inseparable whole, to mine spatial-temporal graphs while exploiting the evolving spatial-temporal dependencies for each node via message traverse mechanisms. Experiments with ablation and parameter studies have validated the effectiveness of the proposed TraverseNets, and the detailed implementation can be found from https://github.com/nnzhan/TraverseNet.
Graph neural networks (GNNs) constitute a class of deep learning methods for graph data. They have wide applications in chemistry and biology, such as molecular property prediction, reaction prediction and drug-target interaction prediction. Despite the interest, GNN-based modeling is challenging as it requires graph data pre-processing and modeling in addition to programming and deep learning. Here we present DGL-LifeSci, an open-source package for deep learning on graphs in life science. DGL-LifeSci is a python toolkit based on RDKit, PyTorch and Deep Graph Library (DGL). DGL-LifeSci allows GNN-based modeling on custom datasets for molecular property prediction, reaction prediction and molecule generation. With its command-line interfaces, users can perform modeling without any background in programming and deep learning. We test the command-line interfaces using standard benchmarks MoleculeNet, USPTO, and ZINC. Compared with previous implementations, DGL-LifeSci achieves a speed up by up to 6x. For modeling flexibility, DGL-LifeSci provides well-optimized modules for various stages of the modeling pipeline. In addition, DGL-LifeSci provides pre-trained models for reproducing the test experiment results and applying models without training. The code is distributed under an Apache-2.0 License and is freely accessible at https://github.com/awslabs/dgl-lifesci.