Modern recommendation systems rely on real-valued embeddings of categorical features. Increasing the dimension of embedding vectors improves model accuracy but comes at a high cost to model size. We introduce a multi-layer embedding training (MLET) architecture that trains embeddings via a sequence of linear layers to derive superior embedding accuracy vs. model size trade-off. Our approach is fundamentally based on the ability of factorized linear layers to produce superior embeddings to that of a single linear layer. We focus on the analysis and implementation of a two-layer scheme. Harnessing the recent results in dynamics of backpropagation in linear neural networks, we explain the ability to get superior multi-layer embeddings via their tendency to have lower effective rank. We show that substantial advantages are obtained in the regime where the width of the hidden layer is much larger than that of the final embedding (d). Crucially, at conclusion of training, we convert the two-layer solution into a single-layer one: as a result, the inference-time model size scales as d. We prototype the MLET scheme within Facebook's PyTorch-based open-source Deep Learning Recommendation Model. We show that it allows reducing d by 4-8X, with a corresponding improvement in memory footprint, at given model accuracy. The experiments are run on two publicly available click-through-rate prediction benchmarks (Criteo-Kaggle and Avazu). The runtime cost of MLET is 25%, on average.

We argue that the vulnerability of model parameters is of crucial value to the study of model robustness and generalization but little research has been devoted to understanding this matter. In this work, we propose an indicator to measure the robustness of neural network parameters by exploiting their vulnerability via parameter corruption. The proposed indicator describes the maximum loss variation in the non-trivial worst-case scenario under parameter corruption. For practical purposes, we give a gradient-based estimation, which is far more effective than random corruption trials that can hardly induce the worst accuracy degradation. Equipped with theoretical support and empirical validation, we are able to systematically investigate the robustness of different model parameters and reveal vulnerability of deep neural networks that has been rarely paid attention to before. Moreover, we can enhance the models accordingly with the proposed adversarial corruption-resistant training, which not only improves the parameter robustness but also translates into accuracy elevation.

We survey in this article the connections between Machine Learning and Control Theory. Control Theory provide useful concepts and tools for Machine Learning. Conversely Machine Learning can be used to solve large control problems. In the first part of the paper, we develop the connections between reinforcement learning and Markov Decision Processes, which are discrete time control problems. In the second part, we review the concept of supervised learning and the relation with static optimization. Deep learning which extends supervised learning, can be viewed as a control problem. In the third part, we present the links between stochastic gradient descent and mean-field theory. Conversely, in the fourth and fifth parts, we review machine learning approaches to stochastic control problems, and focus on the deterministic case, to explain, more easily, the numerical algorithms.

We propose a hierarchical learning strategy aimed at generating sparse representations and associated models for large noisy datasets. The hierarchy follows from approximation spaces identified at successively finer scales. For promoting model generalization at each scale, we also introduce a novel, projection based penalty operator across multiple dimension, using permutation operators for incorporating proximity and ordering information. The paper presents a detailed analysis of approximation properties in the reconstruction Reproducing Kernel Hilbert Spaces (RKHS) with emphasis on optimality and consistency of predictions and behavior of error functionals associated with the produced sparse representations. Results show the performance of the approach as a data reduction and modeling strategy on both synthetic (univariate and multivariate) and real datasets (time series). The sparse model for the test datasets, generated by the presented approach, is also shown to efficiently reconstruct the underlying process and preserve generalizability.

Recurrent neural networks (RNN) such as long-short-term memory (LSTM) networks are essential in a multitude of daily live tasks such as speech, language, video, and multimodal learning. The shift from cloud to edge computation intensifies the need to contain the growth of RNN parameters. Current research on RNN shows that despite the performance obtained on convolutional neural networks (CNN), keeping a good performance in compressed RNNs is still a challenge. Most of the literature on compression focuses on CNNs using matrix product (MPO) operator tensor trains. However, matrix product state (MPS) tensor trains have more attractive features than MPOs, in terms of storage reduction and computing time at inference. We show that MPS tensor trains should be at the forefront of LSTM network compression through a theoretical analysis and practical experiments on NLP task.

We propose a novel interactive learning framework which we refer to as Interactive Attention Learning (IAL), in which the human supervisors interactively manipulate the allocated attentions, to correct the model's behavior by updating the attention-generating network. However, such a model is prone to overfitting due to scarcity of human annotations, and requires costly retraining. Moreover, it is almost infeasible for the human annotators to examine attentions on tons of instances and features. We tackle these challenges by proposing a sample-efficient attention mechanism and a cost-effective reranking algorithm for instances and features. First, we propose Neural Attention Process (NAP), which is an attention generator that can update its behavior by incorporating new attention-level supervisions without any retraining. Secondly, we propose an algorithm which prioritizes the instances and the features by their negative impacts, such that the model can yield large improvements with minimal human feedback. We validate IAL on various time-series datasets from multiple domains (healthcare, real-estate, and computer vision) on which it significantly outperforms baselines with conventional attention mechanisms, or without cost-effective reranking, with substantially less retraining and human-model interaction cost.

We consider a distributed system, consisting of a heterogeneous set of devices, ranging from low-end to high-end. These devices have different profiles, e.g., different energy budgets, or different hardware specifications, determining their capabilities on performing certain learning tasks. We propose the first approach that enables distributed learning in such a heterogeneous system. Applying our approach, each device employs a neural network (NN) with a topology that fits its capabilities; however, part of these NNs share the same topology, so that their parameters can be jointly learned. This differs from current approaches, such as federated learning, which require all devices to employ the same NN, enforcing a trade-off between achievable accuracy and computational overhead of training. We evaluate heterogeneous distributed learning for reinforcement learning (RL) and observe that it greatly improves the achievable reward on more powerful devices, compared to current approaches, while still maintaining a high reward on the weaker devices. We also explore supervised learning, observing similar gains.

Global Navigation Satellite System typically performs poorly in urban environments when there is no line-of-sight between the devices and the satellites, and thus alternative localization methods are often required. We present a simple yet effective method for localization based on pathloss. In our approach, the user to be localized reports the received signal strength from a set of base stations with known locations. For each base station we have a good approximation of the pathloss at each location in the map, provided by RadioUNet, an efficient deep learning-based simulator of pathloss functions in urban environment, akin to ray-tracing. Using the approximations of the pathloss functions of all base stations and the reported signal strengths, we are able to extract a very accurate approximation of the location of the user.

Disentangled representation learning has recently attracted a significant amount of attention, particularly in the field of image representation learning. However, learning the disentangled representations behind a graph remains largely unexplored, especially for the attributed graph with both node and edge features. Disentanglement learning for graph generation has substantial new challenges including 1) the lack of graph deconvolution operations to jointly decode node and edge attributes; and 2) the difficulty in enforcing the disentanglement among latent factors that respectively influence: i) only nodes, ii) only edges, and iii) joint patterns between them. To address these challenges, we propose a new disentanglement enhancement framework for deep generative models for attributed graphs. In particular, a novel variational objective is proposed to disentangle the above three types of latent factors, with novel architecture for node and edge deconvolutions. Moreover, within each type, individual-factor-wise disentanglement is further enhanced, which is shown to be a generalization of the existing framework for images. Qualitative and quantitative experiments on both synthetic and real-world datasets demonstrate the effectiveness of the proposed model and its extensions.

Deep learning models often raise privacy concerns as they leak information about their training data. This enables an adversary to determine whether a data point was in a model's training set by conducting a membership inference attack (MIA). Prior work has conjectured that regularization techniques, which combat overfitting, may also mitigate the leakage. While many regularization mechanisms exist, their effectiveness against MIAs has not been studied systematically, and the resulting privacy properties are not well understood. We explore the lower bound for information leakage that practical attacks can achieve. First, we evaluate the effectiveness of 8 mechanisms in mitigating two recent MIAs, on three standard image classification tasks. We find that certain mechanisms, such as label smoothing, may inadvertently help MIAs. Second, we investigate the potential of improving the resilience to MIAs by combining complementary mechanisms. Finally, we quantify the opportunity of future MIAs to compromise privacy by designing a white-box `distance-to-confident' (DtC) metric, based on adversarial sample crafting. Our metric reveals that, even when existing MIAs fail, the training samples may remain distinguishable from test samples. This suggests that regularization mechanisms can provide a false sense of privacy, even when they appear effective against existing MIAs.