Artificial intelligence (AI) systems hold great promise as decision-support tools, but we must be able to identify and understand their inevitable mistakes if they are to fulfill this potential. This is particularly true in domains where the decisions are high-stakes, such as law, medicine, and the military. In this Perspective, we describe the particular challenges for AI decision support posed in military coalition operations. These include having to deal with limited, low-quality data, which inevitably compromises AI performance. We suggest that these problems can be mitigated by taking steps that allow rapid trust calibration so that decision makers understand the AI system's limitations and likely failures and can calibrate their trust in its outputs appropriately. We propose that AI services can achieve this by being both interpretable and uncertainty-aware. Creating such AI systems poses various technical and human factors challenges. We review these challenges and recommend directions for future research.

Heterogeneous information networks (HINs) are ubiquitous in real-world applications. Due to the heterogeneity in HINs, the typed edges may not fully align with each other. In order to capture the semantic subtlety, we propose the concept of aspects with each aspect being a unit representing one underlying semantic facet. Meanwhile, network embedding has emerged as a powerful method for learning network representation, where the learned embedding can be used as features in various downstream applications. Therefore, we are motivated to propose a novel embedding learning framework—AspEm—to preserve the semantic information in HINs based on multiple aspects. Instead of preserving information of the network in one semantic space, AspEm encapsulates information regarding each aspect individually. In order to select aspects for embedding purpose, we further devise a solution for AspEm based on dataset-wide statistics. To corroborate the efficacy of AspEm, we conducted experiments on two real-words datasets with two types of applications—classification and link prediction. Experiment results demonstrate that AspEm can outperform baseline network embedding learning methods by considering multiple aspects, where the aspects can be selected from the given HIN in an unsupervised manner.

Non-Bayesian social learning theory provides a framework that models distributed inference for a group of agents interacting over a network. Agents iteratively form and communicate beliefs about an unknown state of the world with their neighbors using a learning rule. Existing approaches assume agents have access to precise statistical models (in the form of likelihoods) for the state of the world. However in many situations, such models must be learned from finite data. We propose a social learning rule that takes into account uncertainty in the statistical models using second-order probabilities. Therefore, beliefs derived from uncertain models are sensitive to the amount of past evidence collected for each hypothesis. These beliefs characterize whether or not the hypotheses are consistent with the true state of the world. We explicitly show the dependency of the generated beliefs with respect to the amount of prior evidence. Furthermore, as the amount of prior evidence goes to infinity, learning occurs and is consistent with traditional social learning theory.

We consider a sensor network focused on target localization, where sensors measure the signal strength emitted from the target. Each measurement is quantized to one bit and sent to the fusion center. A general attack is considered at some sensors that attempts to cause the fusion center to produce an inaccurate estimation of the target location. The attack is a combination of man-in-the-middle, hacking, and spoofing attacks that can effectively change both signals going into and coming out of the sensor nodes in a realistic manner. We show that the essential effect of attacks is to alter the naive estimate of the distance between the target and each attacked sensor, which ignores the existence of attacks, to a different extent, giving rise to a geometric inconsistency among the attacked and unattacked sensors. With the help of two secure sensors, a class of detectors are proposed to detect the attacked sensors by scrutinizing the existence of the geometric inconsistency. We show that the false alarm and miss probabilities of the proposed detectors decrease exponentially as the number of measurement samples increases, which implies that with sufficient measurement samples, the proposed detectors can identify the attacked and unattacked sensors with any required accuracy. Numerical results show that compared to the cases where all sensors are employed without detecting attacks or only the secure sensors are employed, the localization performance can be significantly improved if we employ the secure sensors and the sensors which are declared as unattacked by the proposed detector.

Deterministic neural nets have been shown to learn effective predictors on a wide range of machine learning problems. However, as the standard approach is to train the network to minimize a prediction loss, the resultant model remains ignorant to its prediction confidence. Orthogonally to Bayesian neural nets that indirectly infer prediction uncertainty through weight uncertainties, we propose explicit modeling of the same using the theory of subjective logic. By placing a Dirichlet distribution on the class probabilities, we treat predictions of a neural net as subjective opinions and learn the function that collects the evidence leading to these opinions by a deterministic neural net from data. The resultant predictor for a multi-class classification problem is another Dirichlet distribution whose parameters are set by the continuous output of a neural net. We provide a preliminary analysis on how the peculiarities of our new loss function drive improved uncertainty estimation. We observe that our method achieves unprecedented success on detection of out-of-distribution queries and endurance against adversarial perturbations.

Despite the remarkable success in a broad set of sensing applications, state-of-the-art deep learning techniques struggle with complex reasoning tasks across a distributed set of sensors. Unlike recognizing transient complex activities (e.g., human activities such as walking or running) from a single sensor, detecting more complex events with larger spatial and temporal dependencies across multiple sensors is extremely difficult, e.g., utilizing a hospital's sensor network to detect whether a nurse is following a sanitary protocol as they traverse from patient to patient. Training a more complicated model requires a larger amount of data-which is unrealistic considering complex events rarely happen in nature. Moreover, neural networks struggle with reasoning about serial, aperiodic events separated by large quantities in the spatial-temporal dimensions. We propose Neuroplex, a neural-symbolic framework that learns to perform complex reasoning on raw sensory data with the help of high-level, injected human knowledge. Neuroplex decomposes the entire complex learning space into explicit perception and reasoning layers, i.e., by maintaining neural networks to perform low-level perception tasks and neurally reconstructed reasoning models to perform high-level, explainable reasoning. After training the neurally reconstructed reasoning model using human knowledge, Neuroplex allows effective end-to-end training of perception models with an additional semantic loss using only sparse, high-level annotations. Our experiments and evaluation show that Neuroplex is capable of learning to efficiently and effectively detect complex events-which cannot be handled by state-of-the-art neural network models. During the training, Neuroplex not only reduces data annotation requirements by 100x, but also significantly speeds up the learning process for complex event detection by 4x.