The event sequence of many diverse systems is represented as a sequence of discrete events in a continuous space. Examples of such an event sequence are earthquake aftershock events, financial transactions, e-commerce transactions, social network activity of a user, and the user's web search pattern. Finding such an intricate pattern helps discover which event will occur in the future and when it will occur. A Hawkes process is a mathematical tool used for modeling such time series discrete events. Traditionally, the Hawkes process uses a critical component for modeling data as an intensity function with a parameterized kernel function. The Hawkes process's intensity function involves two components: the background intensity and the effect of events' history. However, such parameterized assumption can not capture future event characteristics using past events data precisely due to bias in modeling kernel function. This paper explores the recent advancement using novel deep learning-based methods to model kernel function to remove such parametrized kernel function. In the end, we will give potential future research directions to improve modeling using the Hawkes process.

Current modeling approaches for hydrological modeling often rely on either physics-based or data-science methods, including Machine Learning (ML) algorithms. While physics-based models tend to rigid structure resulting in unrealistic parameter values in certain instances, ML algorithms establish the input-output relationship while ignoring the constraints imposed by well-known physical processes. While there is a notion that the physics model enables better process understanding and ML algorithms exhibit better predictive skills, scientific knowledge that does not add to predictive ability may be deceptive. Hence, there is a need for a hybrid modeling approach to couple ML algorithms and physics-based models in a synergistic manner. Here we develop a Physics Informed Machine Learning (PIML) model that combines the process understanding of conceptual hydrological model with predictive abilities of state-of-the-art ML models. We apply the proposed model to predict the monthly time series of the target (streamflow) and intermediate variables (actual evapotranspiration) in the Narmada river basin in India. Our results show the capability of the PIML model to outperform a purely conceptual model ($abcd$ model) and ML algorithms while ensuring the physical consistency in outputs validated through water balance analysis. The systematic approach for combining conceptual model structure with ML algorithms could be used to improve the predictive accuracy of crucial hydrological processes important for flood risk assessment.

We present Multiscale Vision Transformers (MViT) for video and image recognition, by connecting the seminal idea of multiscale feature hierarchies with transformer models. Multiscale Transformers have several channel-resolution scale stages. Starting from the input resolution and a small channel dimension, the stages hierarchically expand the channel capacity while reducing the spatial resolution. This creates a multiscale pyramid of features with early layers operating at high spatial resolution to model simple low-level visual information, and deeper layers at spatially coarse, but complex, high-dimensional features. We evaluate this fundamental architectural prior for modeling the dense nature of visual signals for a variety of video recognition tasks where it outperforms concurrent vision transformers that rely on large scale external pre-training and are 5-10x more costly in computation and parameters. We further remove the temporal dimension and apply our model for image classification where it outperforms prior work on vision transformers. Code is available at: https://github.com/facebookresearch/SlowFast

In this paper, a novel cognitive architecture for action recognition is developed by applying layers of growing grid neural networks.Using these layers makes the system capable of automatically arranging its representational structure. In addition to the expansion of the neural map during the growth phase, the system is provided with a prior knowledge of the input space, which increases the processing speed of the learning phase. Apart from two layers of growing grid networks the architecture is composed of a preprocessing layer, an ordered vector representation layer and a one-layer supervised neural network. These layers are designed to solve the action recognition problem. The first-layer growing grid receives the input data of human actions and the neural map generates an action pattern vector representing each action sequence by connecting the elicited activation of the trained map. The pattern vectors are then sent to the ordered vector representation layer to build the time-invariant input vectors of key activations for the second-layer growing grid. The second-layer growing grid categorizes the input vectors to the corresponding action clusters/sub-clusters and finally the one-layer supervised neural network labels the shaped clusters with action labels. Three experiments using different datasets of actions show that the system is capable of learning to categorize the actions quickly and efficiently. The performance of the growing grid architecture is com-pared with the results from a system based on Self-Organizing Maps, showing that the growing grid architecture performs significantly superior on the action recognition tasks.

This paper explores the use of reinforcement learning (RL) models for autonomous racing. In contrast to passenger cars, where safety is the top priority, a racing car aims to minimize the lap-time. We frame the problem as a reinforcement learning task with a multidimensional input consisting of the vehicle telemetry, and a continuous action space. To find out which RL methods better solve the problem and whether the obtained models generalize to driving on unknown tracks, we put 10 variants of deep deterministic policy gradient (DDPG) to race in two experiments: i)~studying how RL methods learn to drive a racing car and ii)~studying how the learning scenario influences the capability of the models to generalize. Our studies show that models trained with RL are not only able to drive faster than the baseline open source handcrafted bots but also generalize to unknown tracks.

This paper presents a novel patch-based adversarial attack pipeline that trains adversarial patches on 3D human meshes. We sample triangular faces on a reference human mesh, and create an adversarial texture atlas over those faces. The adversarial texture is transferred to human meshes in various poses, which are rendered onto a collection of real-world background images. Contrary to the traditional patch-based adversarial attacks, where prior work attempts to fool trained object detectors using appended adversarial patches, this new form of attack is mapped into the 3D object world and back-propagated to the texture atlas through differentiable rendering. As such, the adversarial patch is trained under deformation consistent with real-world materials. In addition, and unlike existing adversarial patches, our new 3D adversarial patch is shown to fool state-of-the-art deep object detectors robustly under varying views, potentially leading to an attacking scheme that is persistently strong in the physical world.

Recommending medications for patients using electronic health records (EHRs) is a crucial data mining task for an intelligent healthcare system. It can assist doctors in making clinical decisions more efficiently. However, the inherent complexity of the EHR data renders it as a challenging task: (1) Multilevel structures: the EHR data typically contains multilevel structures which are closely related with the decision-making pathways, e.g., laboratory results lead to disease diagnoses, and then contribute to the prescribed medications; (2) Multiple sequences interactions: multiple sequences in EHR data are usually closely correlated with each other; (3) Abundant noise: lots of task-unrelated features or noise information within EHR data generally result in suboptimal performance. To tackle the above challenges, we propose a multilevel selective and interactive network (MeSIN) for medication recommendation. Specifically, MeSIN is designed with three components. First, an attentional selective module (ASM) is applied to assign flexible attention scores to different medical codes embeddings by their relevance to the recommended medications in every admission. Second, we incorporate a novel interactive long-short term memory network (InLSTM) to reinforce the interactions of multilevel medical sequences in EHR data with the help of the calibrated memory-augmented cell and an enhanced input gate. Finally, we employ a global selective fusion module (GSFM) to infuse the multi-sourced information embeddings into final patient representations for medications recommendation. To validate our method, extensive experiments have been conducted on a real-world clinical dataset. The results demonstrate a consistent superiority of our framework over several baselines and testify the effectiveness of our proposed approach.

Recently the vision transformer (ViT) architecture, where the backbone purely consists of self-attention mechanism, has achieved very promising performance in visual classification. However, the high performance of the original ViT heavily depends on pretraining using ultra large-scale datasets, and it significantly underperforms on ImageNet-1K if trained from scratch. This paper makes the efforts toward addressing this problem, by carefully considering the role of visual tokens. First, for classification head, existing ViT only exploits class token while entirely neglecting rich semantic information inherent in high-level visual tokens. Therefore, we propose a new classification paradigm, where the second-order, cross-covariance pooling of visual tokens is combined with class token for final classification. Meanwhile, a fast singular value power normalization is proposed for improving the second-order pooling. Second, the original ViT employs the naive embedding of fixed-size image patches, lacking the ability to model translation equivariance and locality. To alleviate this problem, we develop a light-weight, hierarchical module based on off-the-shelf convolutions for visual token embedding. The proposed architecture, which we call So-ViT, is thoroughly evaluated on ImageNet-1K. The results show our models, when trained from scratch, outperform the competing ViT variants, while being on par with or better than state-of-the-art CNN models. Code is available at https://github.com/jiangtaoxie/So-ViT

Click-Through Rate (CTR) prediction is a core task in nowadays commercial recommender systems. Feature crossing, as the mainline of research on CTR prediction, has shown a promising way to enhance predictive performance. Even though various models are able to learn feature interactions without manual feature engineering, they rarely attempt to individually learn representations for different feature structures. In particular, they mainly focus on the modeling of cross sparse features but neglect to specifically represent cross dense features. Motivated by this, we propose a novel Extreme Cross Network, abbreviated XCrossNet, which aims at learning dense and sparse feature interactions in an explicit manner. XCrossNet as a feature structure-oriented model leads to a more expressive representation and a more precise CTR prediction, which is not only explicit and interpretable, but also time-efficient and easy to implement.

We develop a simple and elegant method for lossless compression using latent variable models, which we call 'bits back with asymmetric numeral systems' (BB-ANS). The method involves interleaving encode and decode steps, and achieves an optimal rate when compressing batches of data. We demonstrate it firstly on the MNIST test set, showing that state-of-the-art lossless compression is possible using a small variational autoencoder (VAE) model. We then make use of a novel empirical insight, that fully convolutional generative models, trained on small images, are able to generalize to images of arbitrary size, and extend BB-ANS to hierarchical latent variable models, enabling state-of-the-art lossless compression of full-size colour images from the ImageNet dataset. We describe 'Craystack', a modular software framework which we have developed for rapid prototyping of compression using deep generative models.