Representation learning for time series using contrastive learning has emerged as a critical technique for improving the performance of downstream tasks. To advance this effective approach, we introduce CaTT (Contrast All The Time), a new approach to unsupervised contrastive learning for time series, which takes advantage of dynamics between temporally similar moments more efficiently and effectively than existing methods. CaTT departs from conventional time-series contrastive approaches that rely on data augmentations or selected views. Instead, it uses the full temporal dimension by contrasting all time steps in parallel. This is made possible by a scalable NT-pair formulation, which extends the classic N-pair loss across both batch and temporal dimensions, making the learning process end-to-end and more efficient. CaTT learns directly from the natural structure of temporal data, using repeated or adjacent time steps as implicit supervision, without the need for pair selection heuristics. We demonstrate that this approach produces superior embeddings which allow better performance in downstream tasks. Additionally, training is faster than other contrastive learning approaches, making it suitable for large-scale and real-world time series applications. The source code is publicly available at https://github.com/sfi-norwai/CaTT.

Constructing an accurate representation model of phenomena with fewer measurements is a fundamental challenge in the Internet of Things. Leveraging sparse sensing policies to select the most informative measurements is a prominent technique for addressing resource constraints. However, designing such sensing policies requires significant domain knowledge and involves manually fine-tuned heuristics that are task-specific and often non-adaptive. In this work, we propose reducing manual-engineering efforts in designing sensing policies by using an automated approach based on deep reinforcement learning. Guided by an uncertainty-aware prediction model, the sensors learn sensing behaviors autonomously by optimizing an application goal formulated in the reward function based on the measured peaks-over-threshold. We apply the proposed approach in two use cases of monitoring air quality and indoor noise and show the adaptability and transferability of the learned policies. Compared to conventional periodic sensing methods, our results achieve, on average, an increased detection in periods of interest by 78.5% and 357.3% while reducing energy expenditure by 14.3% and 7.6% for air quality and noise monitoring, respectively. Additionally, the resulting representation models are more credible, as measured by various metrics of probabilistic modeling.

Data augmentation helps neural networks generalize better by enlarging the training set, but it remains an open question how to effectively augment graph data to enhance the performance of GNNs (Graph Neural Networks). While most existing graph regularizers focus on manipulating graph topological structures by adding/removing edges, we offer a method to augment node features for better performance. We propose FLAG (Free Large-scale Adversarial Augmentation on Graphs), which iteratively augments node features with gradient-based adversarial perturbations during training. By making the model invariant to small fluctuations in input data, our method helps models generalize to out-of-distribution samples and boosts model performance at test time. FLAG is a general-purpose approach for graph data, which universally works in node classification, link prediction, and graph classification tasks. FLAG is also highly flexible and scalable, and is deployable with arbitrary GNN backbones and large-scale datasets. We demon-strate the efficacy and stability of our method through ex-tensive experiments and ablation studies. We also provide intuitive observations for a deeper understanding of our method. We open source our implementation at https://github.com/devnkong/FLAG.

Data poisoning---the process by which an attacker takes control of a model by making imperceptible changes to a subset of the training data---is an emerging threat in the context of neural networks. Existing attacks for data poisoning neural networks have relied on hand-crafted heuristics, because solving the poisoning problem directly via bilevel optimization is generally thought of as intractable for deep models. We propose MetaPoison, a first-order method that approximates the bilevel problem via meta-learning and crafts poisons that fool neural networks. MetaPoison is effective: it outperforms previous clean-label poisoning methods by a large margin. MetaPoison is robust: poisoned data made for one model transfer to a variety of victim models with unknown training settings and architectures. MetaPoison is general-purpose, it works not only in fine-tuning scenarios, but also for end-to-end training from scratch, which till now hasn't been feasible for clean-label attacks with deep nets. MetaPoison can achieve arbitrary adversary goals---like using poisons of one class to make a target image don the label of another arbitrarily chosen class. Finally, MetaPoison works in the real-world. We demonstrate for the first time successful data poisoning of models trained on the black-box Google Cloud AutoML API.

In this paper, we explore clean-label poisoning attacks on deep convolutional networks with access to neither the network’s output nor its architecture or parameters. Our goal is to ensure that after injecting the poisons into the training data, a model with unknown architecture and parameters trained on that data will misclassify the target image into a specific class. To achieve this goal, we generate multiple poison images from the base class by adding small perturbations which cause the poison images to trap the target image within their convex polytope in feature space. We also demonstrate that using Dropout during crafting of the poisons and enforcing this objective in multiple layers enhances transferability, enabling attacks against both the transfer learning and end-to-end training settings. We demonstrate transferable attack success rates of over 50% by poisoning only 1% of the training set.

Adversarial training, in which a network is trained on adversarial examples, is one of the few defenses against adversarial attacks that withstands strong attacks. Unfortunately, the high cost of generating strong adversarial examples makes standard adversarial training impractical on large-scale problems like ImageNet. We present an algorithm that eliminates the overhead cost of generating adversarial examples by recycling the gradient information computed when updating model parameters. Our "free" adversarial training algorithm achieves comparable robustness to PGD adversarial training on the CIFAR-10 and CIFAR-100 datasets at negligible additional cost compared to natural training, and can be 7 to 30 times faster than other strong adversarial training methods. Using a single workstation with 4 P100 GPUs and 2 days of runtime, we can train a robust model for the large-scale ImageNet classification task that maintains 40% accuracy against PGD attacks.

Neural network training relies on our ability to find "good" minimizers of highly non-convex loss functions. It is well known that certain network architecture designs (e.g., skip connections) produce loss functions that train easier, and well-chosen training parameters (batch size, learning rate, optimizer) produce minimizers that generalize better. However, the reasons for these differences, and their effect on the underlying loss landscape, is not well understood. In this paper, we explore the structure of neural loss functions, and the effect of loss landscapes on generalization, using a range of visualization methods. First, we introduce a simple "filter normalization" method that helps us visualize loss function curvature, and make meaningful side-by-side comparisons between loss functions. Then, using a variety of visualizations, we explore how network architecture affects the loss landscape, and how training parameters affect the shape of minimizers.

The alternating direction method of multipliers (ADMM) is commonly used for distributed model fitting problems, but its performance and reliability depend strongly on user-defined penalty parameters. We study distributed ADMM methods that boost performance by using different fine-tuned algorithm parameters on each worker node. We present a O(1/k) convergence rate for adaptive ADMM methods with node-specific parameters, and propose adaptive consensus ADMM (ACADMM), which automatically tunes parameters without user oversight.

Recent approaches to distributed model fitting rely heavily on consensus ADMM, where each node solves small sub-problems using only local data. We propose iterative methods that solve global sub-problems over an entire distributed dataset. This is possible using transpose reduction strategies that allow a single node to solve least-squares over massive datasets without putting all the data in one place. This results in simple iterative methods that avoid the expensive inner loops required for consensus methods. We analyze the convergence rates of the proposed schemes and demonstrate the efficiency of this approach by fitting linear classifiers and sparse linear models to large datasets using a distributed implementation with up to 20,000 cores in far less time than previous approaches.

With the growing importance of large network models and enormous training datasets, GPUs have become increasingly necessary to train neural networks. This is largely because conventional optimization algorithms rely on stochastic gradient methods that don’t scale well to large numbers of cores in a cluster setting. Furthermore, the convergence of all gradient methods, including batch methods, suffers from common problems like saturation effects, poor conditioning, and saddle points. This paper explores an unconventional training method that uses alternating direction methods and Bregman iteration to train networks without gradient descent steps. The proposed method reduces the network training problem to a sequence of minimization sub-steps that can each be solved globally in closed form. The proposed method is advantageous because it avoids many of the caveats that make gradient methods slow on highly non-convex problems. In addition, the method exhibits strong scaling in the distributed setting, yielding linear speedups even when split over thousands of cores.

The increasing complexity of deep learning architectures is resulting in training time requiring weeks or even months. This slow training is due in part to "vanishing gradients," in which the gradients used by back-propagation are extremely large for weights connecting deep layers (layers near the output layer), and extremely small for shallow layers (near the input layer), this results in slow learning in the shallow layers. Additionally, it has also been shown that in highly non-convex problems, such as deep neural networks, there is a proliferation of high-error low curvature saddle points, which slows down learning dramatically [1]. In this paper, we attempt to overcome the two above problems by proposing an optimization method for training deep neural networks which uses learning rates which are both specific to each layer in the network and adaptive to the curvature of the function, increasing the learning rate at low curvature points. This enables us to speed up learning in the shallow layers of the network and quickly escape high-error low curvature saddle points. We test our method on standard image classification datasets such as MNIST, CIFAR10 and ImageNet, and demonstrate that our method increases accuracy as well as reduces the required training time over standard algorithms.

Recent interest in the use of L_1 regularization in the use of value function approximation includes Petrik et al.’s introduction of L_1-Regularized Approximate Linear Programming (RALP). RALP is unique among L_1-regularized approaches in that it approximates the optimal value function using off-policy samples. Additionally, it produces policies which outperform those of previous methods, such as LSPI. RALP’s value function approximation quality is affected heavily by the choice of state-relevance weights in the objective function of the linear program, and by the distribution from which samples are drawn; however, there has been no discussion of these considerations in the previous literature. In this paper, we discuss and explain the effects of choices in the state-relevance weights and sampling distribution on approximation quality, using both theoretical and experimental illustrations. The results provide insight not only onto these effects, but also provide intuition into the types of MDPs which are especially well suited for approximation with RALP.

A recent surge in research in kernelized approaches to reinforcement learning has sought to bring the benefits of kernelized machine learning techniques to reinforcement learning. Kernelized reinforcement learning techniques are fairly new and different authors have approached the topic with different assumptions and goals. Neither a unifying view nor an understanding of the pros and cons of different approaches has yet emerged. In this paper, we offer a unifying view of the different approaches to kernelized value function approximation for reinforcement learning. We show that, except for different approaches to regularization, Kernelized LSTD (KLSTD) is equivalent to a modelbased approach that uses kernelized regression to find an approximate reward and transition model, and that Gaussian Process Temporal Difference learning (GPTD) returns a mean value function that is equivalent to these other approaches. We also discuss the relationship between our modelbased approach and the earlier Gaussian Processes in Reinforcement Learning (GPRL). Finally, we decompose the Bellman error into the sum of transition error and reward error terms, and demonstrate through experiments that this decomposition can be helpful in choosing regularization parameters.

We show that linear value-function approximation is equivalent to a form of linear model approximation. We then derive a relationship between the model-approximation error and the Bellman error, and show how this relationship can guide feature selection for model improvement and/or value-function improvement. We also show how these results give insight into the behavior of existing feature-selection algorithms.