Traffic classification associates packet streams with known application labels, which is vital for network security and network management. With the rise of NAT, port dynamics, and encrypted traffic, it is increasingly challenging to obtain unified traffic features for accurate classification. Many state-of-the-art traffic classifiers automatically extract features from the packet stream based on deep learning models such as convolution networks. Unfortunately, the compositional and causal relationships between packets are not well extracted in these deep learning models, which affects both prediction accuracy and generalization on different traffic types. In this paper, we present a chained graph model on the packet stream to keep the chained compositional sequence. Next, we propose CGNN, a graph neural network based traffic classification method, which builds a graph classifier over automatically extracted features over the chained graph. Extensive evaluation over real-world traffic data sets, including normal, encrypted and malicious labels, show that, CGNN improves the prediction accuracy by 23\% to 29\% for application classification, by 2\% to 37\% for malicious traffic classification, and reaches the same accuracy level for encrypted traffic classification. CGNN is quite robust in terms of the recall and precision metrics. We have extensively evaluated the parameter sensitivity of CGNN, which yields optimized parameters that are quite effective for traffic classification.

Deep reinforcement learning (DRL) has achieved super-human performance on complex video games (e.g., StarCraft II and Dota II). However, current DRL systems still suffer from challenges of multi-agent coordination, sparse rewards, stochastic environments, etc. In seeking to address these challenges, we employ a football video game, e.g., Google Research Football (GRF), as our testbed and develop an end-to-end learning-based AI system (denoted as TiKick) to complete this challenging task. In this work, we first generated a large replay dataset from the self-playing of single-agent experts, which are obtained from league training. We then developed a distributed learning system and new offline algorithms to learn a powerful multi-agent AI from the fixed single-agent dataset. To the best of our knowledge, Tikick is the first learning-based AI system that can take over the multi-agent Google Research Football full game, while previous work could either control a single agent or experiment on toy academic scenarios. Extensive experiments further show that our pre-trained model can accelerate the training process of the modern multi-agent algorithm and our method achieves state-of-the-art performances on various academic scenarios.

The principle of independent causal mechanisms (ICM) states that generative processes of real world data consist of independent modules which do not influence or inform each other. While this idea has led to fruitful developments in the field of causal inference, it is not widely-known in the NLP community. In this work, we argue that the causal direction of the data collection process bears nontrivial implications that can explain a number of published NLP findings, such as differences in semi-supervised learning (SSL) and domain adaptation (DA) performance across different settings. We categorize common NLP tasks according to their causal direction and empirically assay the validity of the ICM principle for text data using minimum description length. We conduct an extensive meta-analysis of over 100 published SSL and 30 DA studies, and find that the results are consistent with our expectations based on causal insights. This work presents the first attempt to analyze the ICM principle in NLP, and provides constructive suggestions for future modeling choices. Code available at https://github.com/zhijing-jin/icm4nlp

Pruning neural networks reduces inference time and memory costs. On standard hardware, these benefits will be especially prominent if coarse-grained structures, like feature maps, are pruned. We devise two novel saliency-based methods for second-order structured pruning (SOSP) which include correlations among all structures and layers. Our main method SOSP-H employs an innovative second-order approximation, which enables saliency evaluations by fast Hessian-vector products. We validate SOSP-H by comparing it to our second method SOSP-I that uses a well-established Hessian approximation, and to numerous state-of-the-art methods. While SOSP-H performs on par or better in terms of accuracy, it has clear advantages in terms of scalability and efficiency. This allowed us to scale SOSP-H to large-scale vision tasks, even though it captures correlations across all layers of the network. To underscore the global nature of our pruning methods, we evaluate their performance not only by removing structures from a pretrained network, but also by detecting architectural bottlenecks. We show that our algorithms allow to systematically reveal architectural bottlenecks, which we then remove to further increase the accuracy of the networks. Deep neural networks have consistently grown in size over the last years with increasing performance. However, this increase in size leads to slower inference, higher computational requirements and higher cost. To reduce the size of the networks without affecting their performance, a large number of pruning algorithms have been proposed (e.g., LeCun et al., 1990; Hassibi et al., 1993; Reed, 1993; Han et al., 2015; Blalock et al., 2020). Pruning can either be unstructured, i.e. removing individual weights, or structured, i.e. removing entire substructures like nodes or channels.

We introduce a framework for generating, organizing, and reasoning with computational knowledge. It is motivated by the observation that most problems in Computational Sciences and Engineering (CSE) can be described as that of completing (from data) a computational graph representing dependencies between functions and variables. Functions and variables may be known, unknown, or random. Data comes in the form of observations of distinct values of a finite number of subsets of the variables of the graph. The underlying problem combines a regression problem (approximating unknown functions) with a matrix completion problem (recovering unobserved variables in the data). Replacing unknown functions by Gaussian Processes (GPs) and conditioning on observed data provides a simple but efficient approach to completing such graphs. Since the proposed framework is highly expressive, it has a vast potential application scope. Since the completion process can be automatized, as one solves $\sqrt{\sqrt{2}+\sqrt{3}}$ on a pocket calculator without thinking about it, one could, with the proposed framework, solve a complex CSE problem by drawing a diagram. Compared to traditional kriging, the proposed framework can be used to recover unknown functions with much scarcer data by exploiting interdependencies between multiple functions and variables. The Computational Graph Completion (CGC) problem addressed by the proposed framework could therefore also be interpreted as a generalization of that of solving linear systems of equations to that of approximating unknown variables and functions with noisy, incomplete, and nonlinear dependencies. Numerous examples illustrate the flexibility, scope, efficacy, and robustness of the CGC framework and show how it can be used as a pathway to identifying simple solutions to classical CSE problems (digital twin modeling, dimension reduction, mode decomposition, etc.).

Self-attention, an architectural motif designed to model long-range interactions in sequential data, has driven numerous recent breakthroughs in natural language processing and beyond. This work provides a theoretical analysis of the inductive biases of self-attention modules, where our focus is to rigorously establish which functions and long-range dependencies self-attention blocks prefer to represent. Our main result shows that bounded-norm Transformer layers create sparse variables: they can represent sparse functions of the input sequence, with sample complexity scaling only logarithmically with the context length. Furthermore, we propose new experimental protocols to support this analysis and to guide the practice of training Transformers, built around the large body of work on provably learning sparse Boolean functions.

This paper aims to improve the explainability of Autoencoder's (AE) predictions by proposing two explanation methods based on the mean and epistemic uncertainty of log-likelihood estimate, which naturally arise from the probabilistic formulation of the AE called Bayesian Autoencoders (BAE). To quantitatively evaluate the performance of explanation methods, we test them in sensor network applications, and propose three metrics based on covariate shift of sensors : (1) G-mean of Spearman drift coefficients, (2) G-mean of sensitivity-specificity of explanation ranking and (3) sensor explanation quality index (SEQI) which combines the two aforementioned metrics. Surprisingly, we find that explanations of BAE's predictions suffer from high correlation resulting in misleading explanations. To alleviate this, a "Coalitional BAE" is proposed, which is inspired by agent-based system theory. Our comprehensive experiments on publicly available condition monitoring datasets demonstrate the improved quality of explanations using the Coalitional BAE.

Short Term Load Forecast (STLF) is necessary for effective scheduling, operation optimization trading, and decision-making for electricity consumers. Modern and efficient machine learning methods are recalled nowadays to manage complicated structural big datasets, which are characterized by having a nonlinear temporal dependence structure. We propose different statistical nonlinear models to manage these challenges of hard type datasets and forecast 15-min frequency electricity load up to 2-days ahead. We show that the Long-short Term Memory (LSTM) and the Gated Recurrent Unit (GRU) models applied to the production line of a chemical production facility outperform several other predictive models in terms of out-of-sample forecasting accuracy by the Diebold-Mariano (DM) test with several metrics. The predictive information is fundamental for the risk and production management of electricity consumers.

Neural Architecture Search (NAS) algorithms automate the task of finding optimal deep learning architectures given an initial search space of possible operations. Developing these search spaces is usually a manual affair with pre-optimized search spaces being more efficient, rather than searching from scratch. In this paper we present a new framework called Neural Architecture Type System (NeuralArTS) that categorizes the infinite set of network operations in a structured type system. We further demonstrate how NeuralArTS can be applied to convolutional layers and propose several future directions.

Object detection is a field in deep learning that is applied in several computer vision tasks, including text detection. There are mainly two types of object detection models: one-stage detectors and two-stages detectors. Some famous examples of one-stage detectors are SSD (Single Shot Multibox Detectors) and YOLO (You Only Look Once). These object detection models are very similar in the way they work. First, they take an image as input; then this image is passed through a set of convolutional layers. These layers are usually part of a larger network that was pre-trained on a large dataset of images.