We explore self-supervised models that can be potentially deployed on mobile devices to learn general purpose audio representations. Specifically, we propose methods that exploit the temporal context in the spectrogram domain. One method estimates the temporal gap between two short audio segments extracted at random from the same audio clip. The other methods are inspired by Word2Vec, a popular technique used to learn word embeddings, and aim at reconstructing a temporal spectrogram slice from past and future slices or, alternatively, at reconstructing the context of surrounding slices from the current slice. We focus our evaluation on small encoder architectures, which can be potentially run on mobile devices during both inference (re-using a common learned representation across multiple downstream tasks) and training (capturing the true data distribution without compromising users’ privacy when combined with federated learning). We evaluate the quality of the embeddings produced by the self-supervised learning models, and show that they can be re-used for a variety of downstream tasks, and for some tasks even approach the performance of fully supervised models of similar size.
In this paper, we propose a new estimation methodology based on a projected non-linear conjugate gradient (PNCG) algorithm for the destructive negative binomial cure rate model. We show that the PNCG algorithm can simultaneously maximize all model parameters even though the likelihood surface is flat with respect to some model parameters, where, in such a scenario, a profile likelihood approach has been proposed in the literature. We compare the performance of the PNCG algorithm with the well studied expectation maximization (EM) algorithm and show, in particular, that the PNCG algorithm results in more precise and accurate estimates of cure rates. We further show that the PNCG algorithm is computationally less expensive when compared to the EM algorithm with profile likelihood. Finally, we apply the proposed PNCG algorithm on a well-known melanoma data.
Considering the wide application of network embedding methods in graph data mining, inspired by the adversarial attack in deep learning, this paper proposes a Genetic Algorithm (GA) based Euclidean Distance Attack strategy (EDA) to attack the network embedding, so as to prevent certain structural information from being discovered. EDA focuses on disturbing the Euclidean distance between a pair of nodes in the embedding space as much as possible through minimal modifications of the network structure. Since a large number of downstream network algorithms, such as community detection and node classification, rely on the Euclidean distance between nodes to evaluate the similarity between them in the embedding space, EDA can be considered as a universal attack on a variety of network algorithms. Different from traditional supervised attack strategies, EDA does not need labeling information, and, to the best of our knowledge, is the first unsupervised network embedding attack method. We take DeepWalk as the base embedding method to develop the EDA. Experiments with a set of real networks demonstrate that the proposed EDA method can significantly reduce the performance of DeepWalk-based networking algorithms, i.e., community detection and node classification, outperforming several heuristic attack strategies. We also show that EDA also works well on attacking the network algorithms based on other common network embedding methods such as High-Order Proximity preserved Embedding (HOPE) and non-embedding-based network algorithms such as Label Propagation Algorithm (LPA) and Eigenvectors of Matrices (EM). The results indicate a strong transferability of the EDA method.
We propose an algorithm named best-scored random forest for binary classification problems. The terminology ‘best-scored’ means to select the one with the best empirical performance out of a certain number of purely random tree candidates as each single tree in the forest. In this way, the resulting forest can be more accurate than the original purely random forest. From the theoretical perspective, within the framework of regularized empirical risk minimization penalized on the number of splits, we establish almost optimal convergence rates for the proposed best-scored random trees under certain conditions which can be extended to the best-scored random forest. In addition, we present a counterexample to illustrate that in order to ensure the consistency of the forest, every dimension must have the chance to be split. In the numerical experiments, for the sake of efficiency, we employ an adaptive random splitting criterion. Comparative experiments with other state-of-art classification methods demonstrate the accuracy of our best-scored random forest.
We analyze the ordinal structure of long-range dependent time series. To this end, we use so called ordinal patterns which describe the relative position of consecutive data points. We provide two estimators for the probabilities of ordinal patterns and prove limit theorems in different settings, namely stationarity and (less restrictive) stationary increments. In the second setting, we encounter a Rosenblatt distribution in the limit. We prove more general limit theorems for functions with Hermite rank 1 and 2. We derive the limit distribution for an estimation of the Hurst parameter $H$ if it is higher than 3/4. Thus, our theorems complement results for lower values of $H$ which can be found in the literature. Finally, we provide some simulations that illustrate our theoretical results.
Example weighting algorithm is an effective solution to the training bias problem. However, typical methods are usually limited to human knowledge and require laborious tuning of hyperparameters. In this study, we propose a novel example weighting framework called Learning to Auto Weight (LAW), which can learn weighting policy from data adaptively based on reinforcement learning (RL). To shrink the huge searching space in a complete training process, we divide the training procedure consisting of numerous iterations into a small number of stages, and then search a low-deformational continuous vector as action, which determines the weight of each sample. To make training more efficient, we make an innovative design of the reward to remove randomness during the RL process. Experimental results demonstrate the superiority of weighting policy explored by LAW over standard training pipeline. Especially, compared with baselines, LAW can find a better weighting schedule which achieves higher accuracy in the origin CIFAR dataset, and over 10% higher in accuracy on the contaminated CIFAR dataset with 30% label noises. Our code will be released soon.
Machine learning tasks such as optimizing the hyper-parameters of a model for a new dataset or few-shot learning can be vastly accelerated if they are not done from scratch for every new dataset, but carry over findings from previous runs. Meta-learning makes use of features of a whole dataset such as its number of instances, its number of predictors, the means of the predictors etc., so called meta-features, dataset summary statistics or simply dataset characteristics, which so far have been hand-crafted, often specifically for the task at hand. More recently, unsupervised dataset encoding models based on variational auto-encoders have been successful in learning such characteristics for the special case when all datasets follow the same schema, but not beyond. In this paper we design a novel model, Dataset2Vec, that is able to characterize datasets with a latent feature vector based on batches and thus is able to generalize beyond datasets having the same schema to arbitrary (tabular) datasets. To do so, we employ auxiliary learning tasks on batches of datasets, esp. to distinguish batches from different datasets. We show empirically that the meta-features collected from batches of similar datasets are concentrated within a small area in the latent space, hence preserving similarity. We also show that using the dataset characteristics learned by Dataset2Vec in a state-of-the-art hyper-parameter optimization model outperforms the hand-crafted meta-features that have been used in the hyper-parameter optimization literature so far. As a result, we advance the current state-of-the-art results for hyper-parameter optimization.
Deep neural networks whose parameters are distributed according to typical initialization schemes exhibit undesirable properties that can emerge as the number of layers increases. These issues include a vanishing dependency on the input and a concentration on restrictive families of functions including constant functions. We address these problems by considering the limit of infinite total depth and examine the conditions under which we achieve convergence to well-behaved continuous-time processes. Doing so we establish the connection between infinitely deep residual networks and solutions to stochastic differential equations, i.e. diffusion processes. We show that deep neural networks satisfying such connection don’t suffer from the mentioned pathologies and analyze the SDE limits to shed light on their behavior.
For a broad variety of critical applications, it is essential to know how confident a classification prediction is. In this paper, we discuss the drawbacks of softmax to calculate class probabilities and to handle uncertainty in Bayesian neural networks. We introduce a new kind of prediction layer called radial prediction layer (RPL) to overcome these issues. In contrast to the softmax classification, RPL is based on the open-world assumption. Therefore, the class prediction probabilities are much more meaningful to assess the uncertainty concerning the novelty of the input. We show that neural networks with RPLs can be learned in the same way as neural networks using softmax. On a 2D toy data set (spiral data), we demonstrate the fundamental principles and advantages. On the real-world ImageNet data set, we show that the open-world properties are beneficially fulfilled. Additionally, we show that RPLs are less sensible to adversarial attacks on the MNIST data set. Due to its features, we expect RPL to be beneficial in a broad variety of applications, especially in critical environments, such as medicine or autonomous driving.
Complex structures are typical in machine learning. Tailoring learning algorithms for every structure requires an effort that may be saved by defining a generic learning procedure adaptive to any complex structure. In this paper, we propose to map any complex structure onto a generic form, called serialization, over which we can apply any sequence-based density estimator. We then show how to transfer the learned density back onto the space of original structures. To expose the learning procedure to the structural particularities of the original structures, we take care that the serializations reflect accurately the structures’ properties. Enumerating all serializations is infeasible. We propose an effective way to sample representative serializations from the complete set of serializations which preserves the statistics of the complete set. Our method is competitive or better than state of the art learning algorithms that have been specifically designed for given structures. In addition, since the serialization involves sampling from a combinatorial process it provides considerable protection from overfitting, which we clearly demonstrate on a number of experiments.
We propose a remarkably general variance-reduced method suitable for solving regularized empirical risk minimization problems with either a large number of training examples, or a large model dimension, or both. In special cases, our method reduces to several known and previously thought to be unrelated methods, such as {\tt SAGA}, {\tt LSVRG}, {\tt JacSketch}, {\tt SEGA} and {\tt ISEGA}, and their arbitrary sampling and proximal generalizations. However, we also highlight a large number of new specific algorithms with interesting properties. We provide a single theorem establishing linear convergence of the method under smoothness and quasi strong convexity assumptions. With this theorem we recover best-known and sometimes improved rates for known methods arising in special cases. As a by-product, we provide the first unified method and theory for stochastic gradient and stochastic coordinate descent type methods.
Graph or network representations are an important foundation for data mining and machine learning tasks in relational data. Many tools of network analysis, like centrality measures, information ranking, or cluster detection rest on the assumption that links capture direct influence, and that paths represent possible indirect influence. This assumption is invalidated in time-stamped network data capturing, e.g., dynamic social networks, biological sequences or financial transactions. In such data, for two time-stamped links (A,B) and (B,C) the chronological ordering and timing determines whether a causal path from node A via B to C exists. A number of works has shown that for that reason network analysis cannot be directly applied to time-stamped network data. Existing methods to address this issue require statistics on causal paths, which is computationally challenging for big data sets. Addressing this problem, we develop an efficient algorithm to count causal paths in time-stamped network data. Applying it to empirical data, we show that our method is more efficient than a baseline method implemented in an OpenSource data analytics package. Our method works efficiently for different values of the maximum time difference between consecutive links of a causal path and supports streaming scenarios. With it, we are closing a gap that hinders an efficient analysis of big time series data on complex networks.
Over-parameterized deep neural networks trained by simple first-order methods are known to be able to fit any labeling of data. When the training dataset contains a fraction of noisy labels, can neural networks be resistant to over-fitting and still generalize on the true distribution? Inspired by recent theoretical work that established connections between over-parameterized neural networks and neural tangent kernel (NTK), we propose two simple regularization methods for this purpose: (i) regularization by the distance between the network parameters to initialization, and (ii) adding a trainable auxiliary variable to the network output for each training example. Theoretically, both methods are related to kernel ridge regression with respect to the NTK, and we prove their generalization guarantee on the true data distribution despite being trained using noisy labels. The generalization bound is independent of the network size, and only depends on the training inputs and true labels (instead of noisy labels) as well as the noise level in the labels. Empirical results verify the effectiveness of these methods on noisily labeled datasets.