The increasing popularity of deep learning models has created new opportunities for developing artificial intelligence–based recommender systems. Designing recommender systems using deep neural networks (DNNs) requires careful architecture design, and further optimization demands extensive co-design efforts on jointly optimizing model architecture and hardware. Design automation, such as Automated Machine Learning (AutoML), is necessary to fully exploit the potential of recommender model design, including model choices and model–hardware co-design strategies. We introduce a novel paradigm that utilizes weight sharing to explore abundant solution spaces. Our paradigm creates a large supernet to search for optimal architectures and co-design strategies to address the challenges of data multimodality and heterogeneity in the recommendation domain. From a model perspective, the supernet includes a variety of operators, dense connectivity, and dimension search options. From a co-design perspective, it encompasses versatile Processing-In-Memory (PIM) configurations to produce hardware-efficient models. Our solution space’s scale, heterogeneity, and complexity pose several challenges, which we address by proposing various techniques for training and evaluating the supernet. Our crafted models show promising results on three Click-Through Rate (CTR) prediction benchmarks, outperforming both manually designed and AutoML-crafted models with state-of-the-art performance when focusing solely on architecture search. From a co-design perspective, we achieve 2× floating-point operations efficiency, 1.8× energy efficiency, and 1.5× performance improvements in recommender models.

The rise of deep neural networks offers new opportunities in optimizing recommender systems. However, optimizing recommender systems using deep neural networks requires delicate architecture fabrication. We propose NASRec, a paradigm that trains a single supernet and efficiently produces abundant models/sub-architectures by weight sharing. To overcome the data multi-modality and architecture heterogeneity challenges in the recommendation domain, NASRec establishes a large supernet (i.e., search space) to search the full architectures. The supernet incorporates versatile choice of operators and dense connectivity to minimize human efforts for finding priors. The scale and heterogeneity in NASRec impose several challenges, such as training inefficiency, operator-imbalance, and degraded rank correlation. We tackle these challenges by proposing single-operator any-connection sampling, operator-balancing interaction modules, and post-training fine-tuning. Our crafted models, NASRecNet, show promising results on three Click-Through Rates (CTR) prediction benchmarks, indicating that NASRec outperforms both manually designed models and existing NAS methods with state-of-the-art performance. Our work is publicly available here.

Dense subgraph detection is a fundamental building block for a variety of applications. Most of the existing methods aim to discover dense subgraphs within either a single network or a multi-view network while ignoring the informative node dependencies across multiple layers of networks in a complex system. To date, it largely remains a daunting task to detect dense subgraphs on multi-layered networks. In this paper, we formulate the problem of dense subgraph detection on multi-layered networks based on cross-layer consistency principle. We further propose a novel algorithm DESTINE based on projected gradient descent with the following advantages. First, armed with the cross-layer dependencies, DESTINE is able to detect significantly more accurate and meaningful dense subgraphs at each layer. Second, it scales linearly w.r.t. the number of links in the multi-layered network. Extensive experiments demonstrate the efficacy of the proposed DESTINE algorithm in various cases.

Ranking on networks plays an important role in many high-impact applications, including recommender systems, social network analysis, bioinformatics and many more. In the age of big data, a recent trend is to address the variety aspect of network ranking. Among others, two representative lines of research include (1) heterogeneous information network with different types of nodes and edges, and (2) network of networks with edges at different resolutions. In this paper, we propose a new network model named Network of Heterogeneous Information Networks (NeoHIN for short) that is capable of simultaneously modeling both different types of nodes/edges, and different edge resolutions. We further propose two new ranking algorithms on NeoHIN based on the cross-domain consistency principle. Experiments on synthetic and real-world networks show that our proposed algorithms are (1) effective, which outperform other existing methods, and (2) efficient, without additional time cost per iteration to their counterparts.

The widespread application of deep learning has changed the landscape of computation in data centers. In particular, personalized recommendation for content ranking is now largely accomplished using deep neural networks. However, despite their importance and the amount of compute cycles they consume, relatively little research attention has been devoted to recommendation systems. To facilitate research and advance the understanding of these workloads, this paper presents a set of real-world, production-scale DNNs for personalized recommendation coupled with relevant performance metrics for evaluation. In addition to releasing a set of open-source workloads, we conduct in-depth analysis that underpins future system design and optimization for at-scale recommendation: Inference latency varies by 60% across three Intel server generations, batching and co-location of inference jobs can drastically improve latency-bounded throughput, and diversity across recommendation models leads to different optimization strategies.

Finding node associations across different networks is the cornerstone behind a wealth of high-impact data mining applications. Traditional approaches are often, explicitly or implicitly, built upon the linearity and/or consistency assumptions. On the other hand, the recent network embedding based methods promise a natural way to handle the non-linearity, yet they could suffer from the disparate node embedding space of different networks. In this paper, we address these limitations and tackle cross-network node associations from a new angle, i.e., cross-network transformation. We ask a generic question: Given two different networks, how can we transform one network to another? We propose an end-to-end model that learns a composition of nonlinear operations so that one network can be transformed to another in a hierarchical manner. The proposed model bears three distinctive advantages. First (composite transformation), it goes beyond the linearity/consistency assumptions and performs the cross-network transformation through a composition of nonlinear computations. Second (representation power), it can learn the transformation of both network structures and node attributes at different resolutions while identifying the cross-network node associations. Third (generality), it can be applied to various tasks, including network alignment, recommendation, cross-layer dependency inference. Extensive experiments on different tasks validate and verify the effectiveness of the proposed model.

We focus on a mobile caching system using erasure codes to cache content in mobile devices, which enter and depart a fixed area according to Poisson process. Due to the high mobility of devices, cached content is lost and to be repaired by device-to-device (D2D) communication. We consider the limited communication range and repair process with multiple contacts among mobile devices. We adopt a coded repair scheme which the repair process runs periodically, and derive analytical expressions of the average repair delay. The derived expressions are then used to evaluate repair delay using different erasure codes and file size. The results show that maximum distance separable codes can yield lower average repair delay compared to regenerating codes for small size of file. We further find that increasing the speed of mobile devices can reduce the average repair delay.

Machine learning sits at the core of many essential products and services at Facebook. This paper describes the hardware and software infrastructure that supports machine learning at global scale. Facebook's machine learning workloads are extremely diverse: services require many different types of models in practice. This diversity has implications at all layers in the system stack. In addition, a sizable fraction of all data stored at Facebook flows through machine learning pipelines, presenting significant challenges in delivering data to high-performance distributed training flows. Computational requirements are also intense, leveraging both GPU and CPU platforms for training and abundant CPU capacity for real-time inference. Addressing these and other emerging challenges continues to require diverse efforts that span machine learning algorithms, software, and hardware design.

Recently several methods have been proposed to learn from data that are represented as sets of multidimensional vectors. Such algorithms usually suffer from the high demand of computational resources, making them impractical on large-scale problems. We propose to solve this problem by condensing i.e. reducing the sizes of the sets while maintaining the learning performance. Three methods are examined and evaluated with a wide spectrum of set learning algorithms on several large-scale image data sets. We discover that k-Means can successfully achieve the goal of condensing. In many cases, k-Means condensing can improve the algorithms' speed, space requirements, and surprisingly, learning performances simultaneously.

We introduce a new discriminative learning method for image classification. We assume that the images are represented by unordered, multi-dimensional, finite sets of feature vectors, and that these sets might have different cardinality. This allows us to use consistent nonparametric divergence estimators to define new kernels over these sets, and then apply them in kernel classifiers. Our numerical results demonstrate that in many cases this approach can outperform state-of-the-art competitors on both simulated and challenging real-world datasets.

MOTIVATION: Knowledge of the subcellular location of a protein is crucial for understanding its functions. The subcellular pattern of a protein is typically represented as the set of cellular components in which it is located, and an important task is to determine this set from microscope images. In this article, we address this classification problem using confocal immunofluorescence images from the Human Protein Atlas (HPA) project. The HPA contains images of cells stained for many proteins; each is also stained for three reference components, but there are many other components that are invisible. Given one such cell, the task is to classify the pattern type of the stained protein. We first randomly select local image regions within the cells, and then extract various carefully designed features from these regions. This region-based approach enables us to explicitly study the relationship between proteins and different cell components, as well as the interactions between these components. To achieve these two goals, we propose two discriminative models that extend logistic regression with structured latent variables. The first model allows the same protein pattern class to be expressed differently according to the underlying components in different regions. The second model further captures the spatial dependencies between the components within the same cell so that we can better infer these components. To learn these models, we propose a fast approximate algorithm for inference, and then use gradient-based methods to maximize the data likelihood. RESULTS: In the experiments, we show that the proposed models help improve the classification accuracies on synthetic data and real cellular images. The best overall accuracy we report in this article for classifying 942 proteins into 13 classes of patterns is about 84.6%, which to our knowledge is the best so far. In addition, the dependencies learned are consistent with prior knowledge of cell organization. AVAILABILITY: http://murphylab.web.cmu.edu/software/.

Matrix factorization methods are extremely useful in many data mining tasks, yet their performances are often degraded by outliers. In this paper, we propose a novel robust matrix factorization algorithm that is insensitive to outliers. We directly formulate robust factorization as a matrix approximation problem with constraints on the rank of the matrix and the cardinality of the outlier set. Then, unlike existing methods that resort to convex relaxations, we solve this problem directly and efficiently. In addition, structural knowledge about the outliers can be incorporated to find outliers more effectively. We applied this method in anomaly detection tasks on various data sets. Empirical results show that this new algorithm is effective in robust modeling and anomaly detection, and our direct solution achieves superior performance over the state-of-the-art methods based on the L1-norm and the nuclear norm of matrices.

An important task in exploring and analyzing real-world data sets is to detect unusual and interesting phenomena. In this paper, we study the group anomaly detection problem. Unlike traditional anomaly detection research that focuses on data points, our goal is to discover anomalous aggregated behaviors of groups of points. For this purpose, we propose the Flexible Genre Model (FGM). FGM is designed to characterize data groups at both the point level and the group level so as to detect various types of group anomalies. We evaluate the effectiveness of FGM on both synthetic and real data sets including images and turbulence data, and show that it is superior to existing approaches in detecting group anomalies.

Real-world relational data are seldom stationary, yet traditional collaborative filtering algorithms generally rely on this assumption. Motivated by our sales prediction problem, we propose a factor-based algorithm that is able to take time into account. By introducing additional factors for time, we formalize this problem as a tensor factorization with a special constraint on the time dimension. Further, we provide a fully Bayesian treatment to avoid tuning parameters and achieve automatic model complexity control. To learn the model we develop an efficient sampling procedure that is capable of analyzing large-scale data sets. This new algorithm, called Bayesian Probabilistic Tensor Factorization (BPTF), is evaluated on several real-world problems including sales prediction and movie recommendation. Empirical results demonstrate the superiority of our temporal model.

Pattern variation is a major factor that affects the performance of recognition systems. In this paper, a novel manifold tangent modeling method called discriminant additive tangent spaces (DATS) is proposed for invariant pattern recognition. In DATS, intra-class variations for traditional tangent learning are called positive tangent samples. In addition, extra-class variations are introduced as negative tangent samples. We use log-odds to measure the significance of samples being positive or negative, and then directly characterizes this log-odds using generalized additive models (GAM). This model is estimated to maximally discriminate positive and negative samples. Besides, since traditional GAM fitting algorithm can not handle the high dimensional data in visual recognition tasks, we also present an efficient, sparse solution for GAM estimation. The resulting DATS is a nonparametric discriminant model based on quite weak prior hypotheses, hence it can depict various pattern variations effectively. Experiments demonstrate the effectiveness of our method in several recognition tasks.

Graph-based inference plays an important role in many mining and learning tasks. Among all the solvers for this problem, belief propagation (BP) provides a general and efficient way to derive approximate solutions. However, for large scale graphs the computational cost of BP is still demanding. In this paper, we propose a multilevel algorithm to accelerate belief propagation on Markov Random Fields (MRF). First, we coarsen the original graph to get a smaller one. Then, BP is applied on the new graph to get a coarse result. Finally the coarse solution is efficiently refined back to derive the original solution. Unlike traditional multi- resolution approaches, our method features adaptive coarsening and efficient refinement. The above process can be recursively applied to reduce the computational cost remarkably. We theoretically justify the feasibility of our method on Gaussian MRFs, and empirically show that it is also effectual on discrete MRFs. The effectiveness of our method is verified in experiments on various inference tasks.