Deep learning algorithms have been applied very successfully to various problems such as image recognition, language processing, mobile advertising, and autonomous driving. However, at the moment the reasons for their performance are not entirely understood. In particular, there is no full mathematical analysis for deep learning algorithms which explains their success. Roughly speaking, the field of deep learning can be divided into three subfields, deep supervised learning, deep unsupervised learning, and deep reinforcement learning. In the following we will focus on supervised learning, since algorithms in this subfield seem to be most accessible for a rigorous mathematical analysis. Loosely speaking, a typical situation that arises in deep supervised learning is the following (cf., e.g., [7]).

Despite existing work in machine learning inference serving, ease-of-use and cost efficiency remain challenges at large scales. Developers must manually search through thousands of model-variants -- versions of already-trained models that differ in hardware, resource footprints, latencies, costs, and accuracies -- to meet the diverse application requirements. Since requirements, query load, and applications themselves evolve over time, these decisions need to be made dynamically for each inference query to avoid excessive costs through naive autoscaling. To avoid navigating through the large and complex trade-off space of model-variants, developers often fix a variant across queries, and replicate it when load increases. However, given the diversity across variants and hardware platforms in the cloud, a lack of understanding of the trade-off space can incur significant costs to developers. This paper introduces INFaaS, a managed and model-less system for distributed inference serving, where developers simply specify the performance and accuracy requirements for their applications without needing to specify a specific model-variant for each query. INFaaS generates model-variants, and efficiently navigates the large trade-off space of model-variants on behalf of developers to meet application-specific objectives: (a) for each query, it selects a model, hardware architecture, and model optimizations, (b) it combines VM-level horizontal autoscaling with model-level autoscaling, where multiple, different model-variants are used to serve queries within each machine. By leveraging diverse variants and sharing hardware resources across models, INFaaS achieves 1.3x higher throughput, violates latency objectives 1.6x less often, and saves up to 21.6x in cost (8.5x on average) compared to state-of-the-art inference serving systems on AWS EC2.

While federated learning traditionally aims to train a single global model across decentralized local datasets, one model may not always be ideal for all participating clients. Here we propose an alternative, where each client only federates with other relevant clients to obtain a stronger model per client-specific objectives. To achieve this personalization, rather than computing a single model average with constant weights for the entire federation as in traditional FL, we efficiently calculate optimal weighted model combinations for each client, based on figuring out how much a client can benefit from another's model. We do not assume knowledge of any underlying data distributions or client similarities, and allow each client to optimize for arbitrary target distributions of interest, enabling greater flexibility for personalization. We evaluate and characterize our method on a variety of federated settings, datasets, and degrees of local data heterogeneity. Our method outperforms existing alternatives, while also enabling new features for personalized FL such as transfer outside of local data distributions.

Supervised learning is a typical problem setting for machine learning that approximates the relationship between the input and output based on a given sets of input and output data. The accuracy of the approximation can be increased using more input and output data to build the model; however, obtaining the appropriate output for the input can be costly. A classic example is the crossbreeding of plants. The environmental conditions (e.g., average monthly temperature, type and amount of fertilizer used, watering conditions, weather) are the input, and the specific properties of the crops are the output. In this case, the controllable variables are related to the fertilizer and watering conditions, but it would take several months to years to perform experiments under various conditions and determine the optimal fertilizer composition and watering conditions.

Over the last five years, the advanced LIGO and advanced Virgo detectors have completed three observing runs, reporting over 50 gravitational wave sources [3, 4]. Significant improvements in the sensitivity of the advanced LIGO and advanced Virgo detectors during the last three observing runs have increased the observable volume they can probe, thereby increasing the number of gravitational wave observations [4]. As these observatories continue to enhance their detection capabilities, and other detectors join the international array of gravitational wave detectors, it is expected that gravitational wave sources will be observed at a rate of several per day [4, 5]. An ever-increasing catalog of gravitational wave sources will enable systematic studies that will refine and advance our understanding of stellar evolution, cosmology, alternative theories and gravity, among others [6-11]. The combination of gravitational and electromagnetic waves, and cosmic neutrinos, will shed revolutionary insights into the nature of supranuclear matter in neutron stars [12-14] and the formation and evolution of black holes and neutron stars, providing new and detailed information about their astrophysical environments [15-18]. While all of these science goals are feasible in principle given the proven detection capabilities of astronomical observatories, it is equally true that established algorithms for the observation of multi-messenger sources, such as template matching and nearest neighbors, are compute-intensive and poorly scalable [19-23]. Furthermore, available computational resources will remain oversubscribed, and planned enhancements will be rapidly outstripped with the advent of next-generation detectors within the next couple of years [24, 25]. Thus, an urgent rethinking is critical if we are to realize the Multi-Messenger Astrophysics program in the big-data era [26-28]. To contend with these challenges, a number of researchers have been exploring the application of deep learning and GPU-accelerated computing.

Limited by the time complexity of querying k-hop neighbors in a graph database, most graph algorithms cannot be deployed online and execute millisecond-level inference. This problem dramatically limits the potential of applying graph algorithms in certain areas, such as financial fraud detection. Therefore, we propose Asynchronous Propagation Attention Network, an asynchronous continuous time dynamic graph algorithm for real-time temporal graph embedding. Traditional graph models usually execute two serial operations: first graph computation and then model inference. We decouple model inference and graph computation step so that the heavy graph query operations will not damage the speed of model inference. Extensive experiments demonstrate that the proposed method can achieve competitive performance and 8.7 times inference speed improvement in the meantime.

Spatial-temporal data forecasting of traffic flow is a challenging task because of complicated spatial dependencies and dynamical trends of temporal pattern between different roads. Existing frameworks typically utilize given spatial adjacency graph and sophisticated mechanisms for modeling spatial and temporal correlations. However, limited representations of given spatial graph structure with incomplete adjacent connections may restrict effective spatial-temporal dependencies learning of those models. To overcome those limitations, our paper proposes Spatial-Temporal Fusion Graph Neural Networks (STFGNN) for traffic flow forecasting. SFTGNN could effectively learn hidden spatial-temporal dependencies by a novel fusion operation of various spatial and temporal graphs, which is generated by a data-driven method. Meanwhile, by integrating this fusion graph module and a novel gated convolution module into a unified layer, SFTGNN could handle long sequences. Experimental results on several public traffic datasets demonstrate that our method achieves state-of-the-art performance consistently than other baselines.

Expressivity plays a fundamental role in evaluating deep neural networks, and it is closely related to understanding the limit of performance improvement. In this paper, we propose a three-pipeline training framework based on critical expressivity, including global model contraction, weight evolution, and link's weight rewiring. Specifically, we propose a pyramidal-like skeleton to overcome the saddle points that affect information transfer. Then we analyze the reason for the modularity (clustering) phenomenon in network topology and use it to rewire potential erroneous weighted links. We conduct numerical experiments on node classification and the results confirm that the proposed training framework leads to a significantly improved performance in terms of fast convergence and robustness to potential erroneous weighted links. The architecture design on GNNs, in turn, verifies the expressivity of GNNs from dynamics and topological space aspects and provides useful guidelines in designing more efficient neural networks.

The generation of stylish Chinese fonts is an important problem involved in many applications. Most of existing generation methods are based on the deep generative models, particularly, the generative adversarial networks (GAN) based models. However, these deep generative models may suffer from the mode collapse issue, which significantly degrades the diversity and quality of generated results. In this paper, we introduce a one-bit stroke encoding to capture the key mode information of Chinese characters and then incorporate it into CycleGAN, a popular deep generative model for Chinese font generation. As a result we propose an efficient method called StrokeGAN, mainly motivated by the observation that the stroke encoding contains amount of mode information of Chinese characters. In order to reconstruct the one-bit stroke encoding of the associated generated characters, we introduce a stroke-encoding reconstruction loss imposed on the discriminator. Equipped with such one-bit stroke encoding and stroke-encoding reconstruction loss, the mode collapse issue of CycleGAN can be significantly alleviated, with an improved preservation of strokes and diversity of generated characters. The effectiveness of StrokeGAN is demonstrated by a series of generation tasks over nine datasets with different fonts. The numerical results demonstrate that StrokeGAN generally outperforms the state-of-the-art methods in terms of content and recognition accuracies, as well as certain stroke error, and also generates more realistic characters.

To achieve high-levels of autonomy, modern robots require the ability to detect and recover from anomalies and failures with minimal human supervision. Multi-modal sensor signals could provide more information for such anomaly detection tasks; however, the fusion of high-dimensional and heterogeneous sensor modalities remains a challenging problem. We propose a deep learning neural network: supervised variational autoencoder (SVAE), for failure identification in unstructured and uncertain environments. Our model leverages the representational power of VAE to extract robust features from high-dimensional inputs for supervised learning tasks. The training objective unifies the generative model and the discriminative model, thus making the learning a one-stage procedure. Our experiments on real field robot data demonstrate superior failure identification performance than baseline methods, and that our model learns interpretable representations. Videos of our results are available on our website: https://sites.google.com/illinois.edu/supervised-vae .