We propose Cormorant, a rotationally covariant neural network architecture for learning the behavior and properties of complex many-body physical systems. We apply these networks to molecular systems with two goals: learning atomic potential energy surfaces for use in Molecular Dynamics simulations, and learning ground state properties of molecules calculated by Density Functional Theory. Some of the key features of our network are that (a) each neuron explicitly corresponds to a subset of atoms; (b) the activation of each neuron is covariant to rotations, ensuring that overall the network is fully rotationally invariant. Furthermore, the non-linearity in our network is based upon tensor products and the Clebsch-Gordan decomposition, allowing the network to operate entirely in Fourier space. Cormorant significantly outperforms competing algorithms in learning molecular Potential Energy Surfaces from conformational geometries in the MD-17 dataset, and is competitive with other methods at learning geometric, energetic, electronic, and thermodynamic properties of molecules on the GDB-9 dataset.

Detection of collectively routing fleets of vehicles in future urban systems may become important for the management of traffic, as such routing may destabilize urban networks leading to deterioration of driving conditions. Accordingly, in this paper we discuss the question whether it is possible to determine the flow of fleet vehicles on all routes given the fleet size and behaviour as well as the combined total flow of fleet and non-fleet vehicles on every route. We prove that the answer to this Inverse Fleet Assignment Problem is 'yes' for myopic fleet strategies which are more 'selfish' than 'altruistic', and 'no' otherwise, under mild assumptions on route/link performance functions. To reach these conclusions we introduce the forward fleet assignment operator and study its properties, proving that it is invertible for 'bad' objectives of fleet controllers. We also discuss the challenges of implementing myopic fleet routing in the real world and compare it to Stackelberg and Nash routing. Finally, we show that optimal Stackelberg fleet routing could involve highly variable mixed strategies in some scenarios, which would likely cause chaos in the traffic network.

Boulders form from a variety of geological processes, which their size, shape, and orientation may help us better understand. Furthermore, they represent potential hazards to spacecraft landing that need to be characterized. However, mapping individual boulders across vast areas is extremely labor-intensive, often limiting the extent over which they are characterized and the statistical robustness of obtained boulder morphometrics. To automate boulder characterization, we use an instance segmentation neural network, Mask R-CNN, to detect and outline boulders in high-resolution satellite images. Our neural network, BoulderNet, was trained from a dataset of > 33,000 boulders in > 750 image tiles from Earth, the Moon, and Mars. BoulderNet not only correctly detects the majority of boulders in images, but it identifies the outline of boulders with high fidelity, achieving average precision and recall values of 72% and 64% relative to manually digitized boulders from the test dataset, when only detections with intersection-over-union ratios > 50% are considered valid. These values are similar to those obtained by human mappers. On Earth, equivalent boulder diameters, aspect ratios, and orientations extracted from predictions were benchmarked against ground measurements and yield values within 15%, 0.20, and 20 degrees of their ground-truth values, respectively. BoulderNet achieves better boulder detection and characterization performance relative to existing methods, providing a versatile open-source tool to characterize entire boulder fields on planetary surfaces.

Undoubtedly, Location-based Social Networks (LBSNs) provide an interesting source of geo-located data that we have previously used to obtain patterns of the dynamics of crowds throughout urban areas. According to our previous results, activity in LBSNs reflects the real activity in the city. Therefore, unexpected behaviors in the social media activity are a trustful evidence of unexpected changes of the activity in the city. In this paper we introduce a hybrid solution to early detect these changes based on applying a combination of two approaches, the use of entropy analysis and clustering techniques, on the data gathered from LBSNs. In particular, we have performed our experiments over a data set collected from Instagram for seven months in New York City, obtaining promising results. The uninterrupted growth in both number of users and activity of Online Social Networks (OSNs) can be attributed to the parallel increase of the smartphone penetration rate. These mobile devices allow a quick interaction with OSNs and make it easy for subscribers to share their ideas, thoughts, photos, messages, etc. All these posts automatically include the subscriber's location when using GPS-enabled devices, especially when interacting with Location-based Social Networks (LBSNs), i.e. location-centred OSNs. These networks focus their activity on sharing experiences at the right place and time they are happening (Foursquare, Twitter, Instagram, etc.). Consequently, LBSNs provide a very attractive source of geo-located data that, in the smart city field, may be an interesting alternative to the traditional video sources to monitor human activity in urban areas. On the one hand, the infrastructure costs are really low. Instead of having complex video-surveillance networks, which need to be deployed and maintained, citizens are the ones in charge of buying, maintaining and connecting their mobile devices. Besides, due to the ubiquity of the LBSNs, the area under analysis may be easily changed without any extra investment. On the other hand, traditional video-surveillance systems need to be monitored by human staff, who may be supported by complex algorithms for video frames analysis.

In the same way that the computer vision (CV) and natural language processing (NLP) communities have developed self-supervised methods, reinforcement learning (RL) can be cast as a self-supervised problem: learning to reach any goal, without requiring human-specified rewards or labels. However, actually building a self-supervised foundation for RL faces some important challenges. Building on prior contrastive approaches to this RL problem, we conduct careful ablation experiments and discover that a shallow and wide architecture, combined with careful weight initialization and data augmentation, can significantly boost the performance of these contrastive RL approaches on challenging simulated benchmarks. Additionally, we demonstrate that, with these design decisions, contrastive approaches can solve real-world robotic manipulation tasks, with tasks being specified by a single goal image provided after training.

We present a unified hard-constraint framework for solving geometrically complex PDEs with neural networks, where the most commonly used Dirichlet, Neumann, and Robin boundary conditions (BCs) are considered. Specifically, we first introduce the "extra fields" from the mixed finite element method to reformulate the PDEs so as to equivalently transform the three types of BCs into linear equations. Based on the reformulation, we derive the general solutions of the BCs analytically, which are employed to construct an ansatz that automatically satisfies the BCs. With such a framework, we can train the neural networks without adding extra loss terms and thus efficiently handle geometrically complex PDEs, alleviating the unbalanced competition between the loss terms corresponding to the BCs and PDEs. We theoretically demonstrate that the "extra fields" can stabilize the training process. Experimental results on real-world geometrically complex PDEs showcase the effectiveness of our method compared with state-of-the-art baselines.

In this paper, we present a novel approach for the design of leaky-wave holographic antennas that generates OAM-carrying electromagnetic waves by combining Flat Optics (FO) and machine learning (ML) techniques. To improve the performance of our system, we use a machine learning technique to discover a mathematical function that can effectively control the entire radiation pattern, i.e., decrease the side lobe level (SLL) while simultaneously increasing the central null depth of the radiation pattern. Precise tuning of the parameters of the impedance equation based on holographic theory is necessary to achieve optimal results in a variety of scenarios. In this research, we applied machine learning to determine the approximate values of the parameters. We can determine the optimal values for each parameter, resulting in the desired radiation pattern, using a total of 77,000 generated datasets. Furthermore, the use of ML not only saves time, but also yields more precise and accurate results than manual parameter tuning and conventional optimization methods.

Predicting the binding sites of the target proteins plays a fundamental role in drug discovery. Most existing deep-learning methods consider a protein as a 3D image by spatially clustering its atoms into voxels and then feed the voxelized protein into a 3D CNN for prediction. However, the CNN-based methods encounter several critical issues: 1) defective in representing irregular protein structures; 2) sensitive to rotations; 3) insufficient to characterize the protein surface; 4) unaware of data distribution shift. To address the above issues, this work proposes EquiPocket, an E(3)-equivariant Graph Neural Network (GNN) for binding site prediction. In particular, EquiPocket consists of three modules: the first one to extract local geometric information for each surface atom, the second one to model both the chemical and spatial structure of the protein, and the last one to capture the geometry of the surface via equivariant message passing over the surface atoms. We further propose a dense attention output layer to better alleviate the data distribution shift effect incurred by the variable protein size. Extensive experiments on several representative benchmarks demonstrate the superiority of our framework to the state-of-the-art methods.

The promising research on Artificial Intelligence usages in suicide prevention has principal gaps, including black box methodologies, inadequate outcome measures, and scarce research on non-verbal inputs, such as social media images (despite their popularity today, in our digital era). This study addresses these gaps and combines theory-driven and bottom-up strategies to construct a hybrid and interpretable prediction model of valid suicide risk from images. The lead hypothesis was that images contain valuable information about emotions and interpersonal relationships, two central concepts in suicide-related treatments and theories. The dataset included 177,220 images by 841 Facebook users who completed a gold-standard suicide scale. The images were represented with CLIP, a state-of-the-art algorithm, which was utilized, unconventionally, to extract predefined features that served as inputs to a simple logistic-regression prediction model (in contrast to complex neural networks). The features addressed basic and theory-driven visual elements using everyday language (e.g., bright photo, photo of sad people). The results of the hybrid model (that integrated theory-driven and bottom-up methods) indicated high prediction performance that surpassed common bottom-up algorithms, thus providing a first proof that images (alone) can be leveraged to predict validated suicide risk. Corresponding with the lead hypothesis, at-risk users had images with increased negative emotions and decreased belonginess. The results are discussed in the context of non-verbal warning signs of suicide. Notably, the study illustrates the advantages of hybrid models in such complicated tasks and provides simple and flexible prediction strategies that could be utilized to develop real-life monitoring tools of suicide.

Digital technology will play a significant role in achieving sustainable human development worldwide. In 2015, United Nations Member States set 17 goals, The Sustainable Development Goals (SDGs), to provide a road map for the achievement of Earth’s peace and human prosperity by 2030. SDG 3 as one of the goals which is aimed at ensuring healthy lives and promoting well-being for all at all ages, will greatly benefit from the implementation of digital technology. With over a billion people, Africa can be better positioned to surmount its health challenges -especially regarding maternal and child health, infectious and non-communicable disease- using digital technology including artificial intelligence.Artificial intelligence (AI) is defined as the automation of activities associated with human thinking such as decision-making, problem-solving and learning.1AI was first used in medicine in the 1970s when medical expert systems – based on Bayesian statistics and decision theory – diagnosed and recommended treatments for glaucoma and infectious disease.2 Progress in Bayesian networks, artificial neural networks, and hybrid intelligent systems in the late 1990s has scaled up bioinformatics research; thereby expanding uptake of Medical Artificial Intelligence (MAI).7 Global investment in MAI is projected to hit about $6.6 Billion by 2021 as it is anticipated that AI implementations in healthcare can help save $150 Billion in costs by 2026.8At present, a more meaningful applicat...