We present the T emporal Graph Benchmark (TGB), a collection of challenging and diverse benchmark datasets for realistic, reproducible, and robust evaluation of machine learning models on temporal graphs.

Dynamic Link Prediction (DLP) addresses the prediction of future links in evolving networks. However, accurately portraying the performance of DLP algorithms poses challenges that might impede progress in the field. Importantly, common evaluation pipelines usually calculate ranking or binary classification metrics, where the scores of observed interactions (positives) are compared with those of randomly generated ones (negatives). However, a single metric is not sufficient to fully capture the differences between DLP algorithms, and is prone to overly optimistic performance evaluation. Instead, an in-depth evaluation should reflect performance variations across different nodes, edges, and time segments. In this work, we contribute tools to perform such a comprehensive evaluation. (1) We propose Birth-Death diagrams, a simple but powerful visualization technique that illustrates the effect of time-based train-test splitting on the difficulty of DLP on a given dataset. (2) We describe an exhaustive taxonomy of negative sampling methods that can be used at evaluation time. (3) We carry out an empirical study of the effect of the different negative sampling strategies. Our comparison between heuristics and state-of-the-art memory-based methods on various real-world datasets confirms a strong effect of using different negative sampling strategies on the test Area Under the Curve (AUC). Moreover, we conduct a visual exploration of the prediction, with additional insights on which different types of errors are prominent over time.

This paper considers the problem of evaluating an autonomous system's competency in performing a task, particularly when working in dynamic and uncertain environments. The inherent opacity of machine learning models, from the perspective of the user, often described as a `black box', poses a challenge. To overcome this, we propose using a measure called the Surprise index, which leverages available measurement data to quantify whether the dynamic system performs as expected. We show that the surprise index can be computed in closed form for dynamic systems when observed evidence in a probabilistic model if the joint distribution for that evidence follows a multivariate Gaussian marginal distribution. We then apply it to a nonlinear spacecraft maneuver problem, where actions are chosen by a reinforcement learning agent and show it can indicate how well the trajectory follows the required orbit.

Effective quantification of uncertainty is an essential and still missing step towards a greater adoption of deep-learning approaches in different applications, including mission-critical ones. In particular, investigations on the predictive uncertainty of deep-learning models describing non-linear dynamical systems are very limited to date. This paper is aimed at filling this gap and presents preliminary results on uncertainty quantification for system identification with neural state-space models. We frame the learning problem in a Bayesian probabilistic setting and obtain posterior distributions for the neural network's weights and outputs through approximate inference techniques. Based on the posterior, we construct credible intervals on the outputs and define a surprise index which can effectively diagnose usage of the model in a potentially dangerous out-of-distribution regime, where predictions cannot be trusted.

Probabilistic graphical models, such as Bayesian networks, are intuitive and theoretically sound tools for modeling uncertainty. A major problem with applying Bayesian networks in practice is that it is hard to judge whether a model fits well a case that it is supposed to solve. One way of expressing a possible dissonance between a model and a case is the {\em surprise index}, proposed by Habbema, which expresses the degree of surprise by the evidence given the model. While this measure reflects the intuition that the probability of a case should be judged in the context of a model, it is computationally intractable. In this paper, we propose an efficient way of approximating the surprise index.

Probabilistic graphical models, such as Bayesian networks, are intuitive and theoretically sound tools for modeling uncertainty. A major problem with applying Bayesian networks in practice is that it is hard to judge whether a model fits well a case that it is supposed to solve. One way of expressing a possible dissonance between a model and a case is the {\em surprise index}, proposed by Habbema, which expresses the degree of surprise by the evidence given the model. While this measure reflects the intuition that the probability of a case should be judged in the context of a model, it is computationally intractable. In this paper, we propose an efficient way of approximating the surprise index.