We show that a single softmax neural net with minimal changes can beat the uncertainty predictions of Deep Ensembles and other more complex single-forward-pass uncertainty approaches. Softmax neural nets cannot capture epistemic uncertainty reliably because for OoD points they extrapolate arbitrarily and suffer from feature collapse. This results in arbitrary softmax entropies for OoD points which can have high entropy, low, or anything in between. We study why, and show that with the right inductive biases, softmax neural nets trained with maximum likelihood reliably capture epistemic uncertainty through the feature-space density. This density is obtained using Gaussian Discriminant Analysis, but it cannot disentangle uncertainties. We show that it is necessary to combine this density with the softmax entropy to disentangle aleatoric and epistemic uncertainty -- crucial e.g. for active learning. We examine the quality of epistemic uncertainty on active learning and OoD detection, where we obtain SOTA ~0.98 AUROC on CIFAR-10 vs SVHN.

This paper investigates the motion planning of autonomous dynamical systems modeled by Markov decision processes (MDP) with unknown transition probabilities over continuous state and action spaces. Linear temporal logic (LTL) is used to specify high-level tasks over infinite horizon, which can be converted into a limit deterministic generalized B\"uchi automaton (LDGBA) with several accepting sets. The novelty is to design an embedded product MDP (EP-MDP) between the LDGBA and the MDP by incorporating a synchronous tracking-frontier function to record unvisited accepting sets of the automaton, and to facilitate the satisfaction of the accepting conditions. The proposed LDGBA-based reward shaping and discounting schemes for the model-free reinforcement learning (RL) only depend on the EP-MDP states and can overcome the issues of sparse rewards. Rigorous analysis shows that any RL method that optimizes the expected discounted return is guaranteed to find an optimal policy whose traces maximize the satisfaction probability. A modular deep deterministic policy gradient (DDPG) is then developed to generate such policies over continuous state and action spaces. The performance of our framework is evaluated via an array of OpenAI gym environments.

We consider image completion from the perspective of amortized inference in an image generative model. We leverage recent state of the art variational auto-encoder architectures that have been shown to produce photo-realistic natural images at non-trivial resolutions. Through amortized inference in such a model we can train neural artifacts that produce diverse, realistic image completions even when the vast majority of an image is missing. We demonstrate superior sample quality and diversity compared to prior art on the CIFAR-10 and FFHQ-256 datasets. We conclude by describing and demonstrating an application that requires an in-painting model with the capabilities ours exhibits: the use of Bayesian optimal experimental design to select the most informative sequence of small field of view x-rays for chest pathology detection.

Recent approaches to goal recognition have leveraged planning landmarks to achieve high-accuracy with low runtime cost. These approaches, however, lack a probabilistic interpretation. Furthermore, while most probabilistic models to goal recognition assume that the recognizer has access to a prior probability representing, for example, an agent's preferences, virtually no goal recognition approach actually uses the prior in practice, simply assuming a uniform prior. In this paper, we provide a model to both extend landmark-based goal recognition with a probabilistic interpretation and allow the estimation of such prior probability and its usage to compute posterior probabilities after repeated interactions of observed agents. We empirically show that our model can not only recognize goals effectively but also successfully infer the correct prior probability distribution representing an agent's preferences.

This paper introduces a procedure for testing the identifiability of Bayesian models for causal inference. Although the do-calculus is sound and complete given a causal graph, many practical assumptions cannot be expressed in terms of graph structure alone, such as the assumptions required by instrumental variable designs, regression discontinuity designs, and within-subjects designs. We present simulation-based identifiability (SBI), a fully automated identification test based on a particle optimization scheme with simulated observations. This approach expresses causal assumptions as priors over functions in a structural causal model, including flexible priors using Gaussian processes. We prove that SBI is asymptotically sound and complete, and produces practical finite-sample bounds. We also show empirically that SBI agrees with known results in graph-based identification as well as with widely-held intuitions for designs in which graph-based methods are inconclusive.

Causal structure discovery from observational data is fundamental to the causal understanding of autonomous systems such as medical decision support systems, advertising campaigns and self-driving cars. This is essential to solve well-known causal decision making and prediction problems associated with those real-world applications. Recently, recursive causal discovery algorithms have gained particular attention among the research community due to their ability to provide good results by using Conditional Independent (CI) tests in smaller sub-problems. However, each of such algorithms needs a refinement function to remove undesired causal relations of the discovered graphs. Notably, with the increase of the problem size, the computation cost (i.e., the number of CI-tests) of the refinement function makes an algorithm expensive to deploy in practice. This paper proposes a generic causal structure refinement strategy that can locate the undesired relations with a small number of CI-tests, thus speeding up the algorithm for large and complex problems. We theoretically prove the correctness of our algorithm. We then empirically evaluate its performance against the state-of-the-art algorithms in terms of solution quality and completion time in synthetic and real datasets.

Real world applications such as economics and policy making often involve solving multi-agent games with two unique features: (1) The agents are inherently asymmetric and partitioned into leaders and followers; (2) The agents have different reward functions, thus the game is general-sum. The majority of existing results in this field focuses on either symmetric solution concepts (e.g. Nash equilibrium) or zero-sum games. It remains vastly open how to learn the Stackelberg equilibrium -- an asymmetric analog of the Nash equilibrium -- in general-sum games efficiently from samples. This paper initiates the theoretical study of sample-efficient learning of the Stackelberg equilibrium in two-player turn-based general-sum games. We identify a fundamental gap between the exact value of the Stackelberg equilibrium and its estimated version using finite samples, which can not be closed information-theoretically regardless of the algorithm. We then establish a positive result on sample-efficient learning of Stackelberg equilibrium with value optimal up to the gap identified above. We show that our sample complexity is tight with matching upper and lower bounds. Finally, we extend our learning results to the setting where the follower plays in a Markov Decision Process (MDP), and the setting where the leader and the follower act simultaneously.

Bayesian neural networks (BNNs) are making significant progress in many research areas where decision making needs to be accompanied by uncertainty estimation. Being able to quantify uncertainty while making decisions is essential for understanding when the model is over-/under-confident, and hence BNNs are attracting interest in safety-critical applications, such as autonomous driving, healthcare and robotics. Nevertheless, BNNs have not been as widely used in industrial practice, mainly because of their increased memory and compute costs. In this work, we investigate quantisation of BNNs by compressing 32-bit floating-point weights and activations to their integer counterparts, that has already been successful in reducing the compute demand in standard pointwise neural networks. We study three types of quantised BNNs, we evaluate them under a wide range of different settings, and we empirically demonstrate that an uniform quantisation scheme applied to BNNs does not substantially decrease their quality of uncertainty estimation.

The softmax policy gradient (PG) method, which performs gradient ascent under softmax policy parameterization, is arguably one of the de facto implementations of policy optimization in modern reinforcement learning. For $\gamma$-discounted infinite-horizon tabular Markov decision processes (MDPs), remarkable progress has recently been achieved towards establishing global convergence of softmax PG methods in finding a near-optimal policy. However, prior results fall short of delineating clear dependencies of convergence rates on salient parameters such as the cardinality of the state space $\mathcal{S}$ and the effective horizon $\frac{1}{1-\gamma}$, both of which could be excessively large. In this paper, we deliver a pessimistic message regarding the iteration complexity of softmax PG methods, despite assuming access to exact gradient computation. Specifically, we demonstrate that softmax PG methods can take exponential time -- in terms of $|\mathcal{S}|$ and $\frac{1}{1-\gamma}$ -- to converge, even in the presence of a benign policy initialization and an initial state distribution amenable to exploration. This is accomplished by characterizing the algorithmic dynamics over a carefully-constructed MDP containing only three actions. Our exponential lower bound hints at the necessity of carefully adjusting update rules or enforcing proper regularization in accelerating PG methods.

Updating observations of a signal due to the delays in the measurement process is a common problem in signal processing, with prominent examples in a wide range of fields. An important example of this problem is the nowcasting of COVID-19 mortality: given a stream of reported counts of daily deaths, can we correct for the delays in reporting to paint an accurate picture of the present, with uncertainty? Without this correction, raw data will often mislead by suggesting an improving situation. We present a flexible approach using a latent Gaussian process that is capable of describing the changing auto-correlation structure present in the reporting time-delay surface. This approach also yields robust estimates of uncertainty for the estimated nowcasted numbers of deaths. We test assumptions in model specification such as the choice of kernel or hyper priors, and evaluate model performance on a challenging real dataset from Brazil. Our experiments show that Gaussian process nowcasting performs favourably against both comparable methods, and a small sample of expert human predictions. Our approach has substantial practical utility in disease modelling -- by applying our approach to COVID-19 mortality data from Brazil, where reporting delays are large, we can make informative predictions on important epidemiological quantities such as the current effective reproduction number.