Generalization bounds which assess the difference between the true risk and the empirical risk have been studied extensively. However, to obtain bounds, current techniques use strict assumptions such as a uniformly bounded or a Lipschitz loss function. To avoid these assumptions, in this paper, we propose a new generalization bound for Bayesian deep nets by exploiting the contractivity of the Log-Sobolev inequalities. Using these inequalities adds an additional loss-gradient norm term to the generalization bound, which is intuitively a surrogate of the model complexity. Empirically, we analyze the affect of this loss-gradient norm term using different deep nets.
Automatic identification of animal species by their vocalization is an important and challenging task. Although many kinds of audio monitoring system have been proposed in the literature, they suffer from several disadvantages such as non-trivial feature selection, accuracy degradation because of environmental noise or intensive local computation. In this paper, we propose a deep learning based acoustic classification framework for Wireless Acoustic Sensor Network (WASN). The proposed framework is based on cloud architecture which relaxes the computational burden on the wireless sensor node. To improve the recognition accuracy, we design a multi-view Convolution Neural Network (CNN) to extract the short-, middle-, and long-term dependencies in parallel. The evaluation on two real datasets shows that the proposed architecture can achieve high accuracy and outperforms traditional classification systems significantly when the environmental noise dominate the audio signal (low SNR). Moreover, we implement and deploy the proposed system on a testbed and analyse the system performance in real-world environments. Both simulation and real-world evaluation demonstrate the accuracy and robustness of the proposed acoustic classification system in distinguishing species of animals.
The success of neural networks in image classification has inspired various hardware implementations on embedded platforms such as Field Programmable Gate Arrays, embedded processors and Graphical Processing Units. These embedded platforms are constrained in terms of power, which is mainly consumed by the Multiply Accumulate operations and the memory accesses for weight fetching. Quantization and pruning have been proposed to address this issue. Though effective, these techniques do not take into account the underlying architecture of the embedded hardware. In this work, we propose PoET-BiN, a Look-Up Table based power efficient implementation on resource constrained embedded devices. A modified Decision Tree approach forms the backbone of the proposed implementation in the binary domain. A LUT access consumes far less power than the equivalent Multiply Accumulate operation it replaces, and the modified Decision Tree algorithm eliminates the need for memory accesses. We applied the PoET-BiN architecture to implement the classification layers of networks trained on MNIST, SVHN and CIFAR-10 datasets, with near state-of-the art results. The energy reduction for the classifier portion reaches up to six orders of magnitude compared to a floating point implementations and up to three orders of magnitude when compared to recent binary quantized neural networks.
Max-Pooling operations are a core component of deep learning architectures. In particular, they are part of most convolutional architectures used in machine vision, since pooling is a natural approach to pattern detection problems. However, these architectures are not well understood from a theoretical perspective. For example, we do not understand when they can be globally optimized, and what is the effect of over-parameterization on generalization. Here we perform a theoretical analysis of a convolutional max-pooling architecture, proving that it can be globally optimized, and can generalize well even for highly over-parameterized models. Our analysis focuses on a data generating distribution inspired by pattern detection problem, where a "discriminative" pattern needs to be detected among "spurious" patterns. We empirically validate that CNNs significantly outperform fully connected networks in our setting, as predicted by our theoretical results.
Recent studies have shown that deep learning models are vulnerable to specifically crafted adversarial inputs that are quasi-imperceptible to humans. In this letter, we propose a novel method to detect adversarial inputs, by augmenting the main classification network with multiple binary detectors (observer networks) which take inputs from the hidden layers of the original network (convolutional kernel outputs) and classify the input as clean or adversarial. During inference, the detectors are treated as a part of an ensemble network and the input is deemed adversarial if at least half of the detectors classify it as so. The proposed method addresses the trade-off between accuracy of classification on clean and adversarial samples, as the original classification network is not modified during the detection process. The use of multiple observer networks makes attacking the detection mechanism non-trivial even when the attacker is aware of the victim classifier. We achieve a 99.5% detection accuracy on the MNIST dataset and 97.5% on the CIFAR-10 dataset using the Fast Gradient Sign Attack in a semi-white box setup. The number of false positive detections is a mere 0.12% in the worst case scenario.
We investigate the challenge of multi-output learning, where the goal is to learn a vector-valued function based on a supervised data set. This includes a range of important problems in Machine Learning including multi-target regression, multi-class classification and multi-label classification. We begin our analysis by introducing the self-bounding Lipschitz condition for multi-output loss functions, which interpolates continuously between a classical Lipschitz condition and a multi-dimensional analogue of a smoothness condition. We then show that the self-bounding Lipschitz condition gives rise to optimistic bounds for multi-output learning, which are minimax optimal up to logarithmic factors. The proof exploits local Rademacher complexity combined with a powerful minoration inequality due to Srebro, Sridharan and Tewari. As an application we derive a state-of-the-art generalization bound for multi-class gradient boosting.
Targeted sentiment classification predicts the sentiment polarity on given target mentions in input texts. Dominant methods employ neural networks for encoding the input sentence and extracting relations between target mentions and their contexts. Recently, graph neural network has been investigated for integrating dependency syntax for the task, achieving the state-of-the-art results. However, existing methods do not consider dependency label information, which can be intuitively useful. To solve the problem, we investigate a novel relational graph attention network that integrates typed syntactic dependency information. Results on standard benchmarks show that our method can effectively leverage label information for improving targeted sentiment classification performances. Our final model significantly outperforms state-of-the-art syntax-based approaches.
The dominant text classification studies focus on training classifiers using textual instances only or introducing external knowledge (e.g., hand-craft features and domain expert knowledge). In contrast, some corpus-level statistical features, like word frequency and distribution, are not well exploited. Our work shows that such simple statistical information can enhance classification performance both efficiently and significantly compared with several baseline models. In this paper, we propose a classifier with gate mechanism named Adaptive Gate Attention model with Global Information (AGA+GI), in which the adaptive gate mechanism incorporates global statistical features into latent semantic features and the attention layer captures dependency relationship within the sentence. To alleviate the overfitting issue, we propose a novel Leaky Dropout mechanism to improve generalization ability and performance stability. Our experiments show that the proposed method can achieve better accuracy than CNN-based and RNN-based approaches without global information on several benchmarks.
We propose self-adaptive training---a new training algorithm that dynamically corrects problematic training labels by model predictions without incurring extra computational cost---to improve generalization of deep learning for potentially corrupted training data. This problem is crucial towards robustly learning from data that are corrupted by, e.g., label noises and out-of-distribution samples. The standard empirical risk minimization (ERM) for such data, however, may easily overfit noises and thus suffers from sub-optimal performance. In this paper, we observe that model predictions can substantially benefit the training process: self-adaptive training significantly improves generalization over ERM under various levels of noises, and mitigates the overfitting issue in both natural and adversarial training. We evaluate the error-capacity curve of self-adaptive training: the test error is monotonously decreasing w.r.t. model capacity. This is in sharp contrast to the recently-discovered double-descent phenomenon in ERM which might be a result of overfitting of noises. Experiments on CIFAR and ImageNet datasets verify the effectiveness of our approach in two applications: classification with label noise and selective classification. We release our code at https://github.com/LayneH/self-adaptive-training.
Deep learning image classifiers usually rely on huge training sets and their training process can be described as learning the similarities and differences among training images. But, images in large training sets are not usually studied from this perspective and fine-level similarities and differences among images is usually overlooked. This is due to lack of fast and efficient computational methods to analyze the contents of these datasets. Some studies aim to identify the influential and redundant training images, but such methods require a model that is already trained on the entire training set. Here, using image processing and numerical analysis tools we develop a practical and fast method to analyze the similarities in image classification datasets. We show that such analysis can provide valuable insights about the datasets and the classification task at hand, prior to training a model. Our method uses wavelet decomposition of images and other numerical analysis tools, with no need for a pre-trained model. Interestingly, the results we obtain corroborate the previous results in the literature that analyzed the similarities using pre-trained CNNs. We show that similar images in standard datasets (such as CIFAR) can be identified in a few seconds, a significant speed-up compared to alternative methods in the literature. By removing the computational speed obstacle, it becomes practical to gain new insights about the contents of datasets and the models trained on them. We show that similarities between training and testing images may provide insights about the generalization of models. Finally, we investigate the similarities between images in relation to decision boundaries of a trained model.
Fine-grained classification is a challenging problem, due to subtle differences among highly-confused categories. Most approaches address this difficulty by learning discriminative representation of individual input image. On the other hand, humans can effectively identify contrastive clues by comparing image pairs. Inspired by this fact, this paper proposes a simple but effective Attentive Pairwise Interaction Network (API-Net), which can progressively recognize a pair of fine-grained images by interaction. Specifically, API-Net first learns a mutual feature vector to capture semantic differences in the input pair. It then compares this mutual vector with individual vectors to generate gates for each input image. These distinct gate vectors inherit mutual context on semantic differences, which allow API-Net to attentively capture contrastive clues by pairwise interaction between two images. Additionally, we train API-Net in an end-to-end manner with a score ranking regularization, which can further generalize API-Net by taking feature priorities into account. We conduct extensive experiments on five popular benchmarks in fine-grained classification. API-Net outperforms the recent SOTA methods, i.e., CUB-200-2011 (90.0%), Aircraft(93.9%), Stanford Cars (95.3%), Stanford Dogs (90.3%), and NABirds (88.1%).
In recent years, spiking neural networks (SNNs) emerge as an alternative to deep neural networks (DNNs). SNNs present a higher computational efficiency using low-power neuromorphic hardware and require less labeled data for training using local and unsupervised learning rules such as spike timing-dependent plasticity (STDP). SNN have proven their effectiveness in image classification on simple datasets such as MNIST. However, to process natural images, a pre-processing step is required. Difference-of-Gaussians (DoG) filtering is typically used together with on-center/off-center coding, but it results in a loss of information that is detrimental to the classification performance. In this paper, we propose to use whitening as a pre-processing step before learning features with STDP. Experiments on CIFAR-10 show that whitening allows STDP to learn visual features that are closer to the ones learned with standard neural networks, with a significantly increased classification performance as compared to DoG filtering. We also propose an approximation of whitening as convolution kernels that is computationally cheaper to learn and more suited to be implemented on neuromorphic hardware. Experiments on CIFAR-10 show that it performs similarly to regular whitening. Cross-dataset experiments on CIFAR-10 and STL-10 also show that it is fairly stable across datasets, making it possible to learn a single whitening transformation to process different datasets.
The Receptive Field (RF) size has been one of the most important factors for One Dimensional Convolutional Neural Networks (1D-CNNs) on time series classification tasks. Large efforts have been taken to choose the appropriate size because it has a huge influence on the performance and differs significantly for each dataset. In this paper, we propose an Omni-Scale block (OS-block) for 1D-CNNs, where the kernel sizes are decided by a simple and universal rule. Particularly, it is a set of kernel sizes that can efficiently cover the best RF size across different datasets via consisting of multiple prime numbers according to the length of the time series. The experiment result shows that models with the OS-block can achieve a similar performance as models with the searched optimal RF size and due to the strong optimal RF size capture ability, simple 1D-CNN models with OS-block achieves the state-of-the-art performance on four time series benchmarks, including both univariate and multivariate data from multiple domains. Comprehensive analysis and discussions shed light on why the OS-block can capture optimal RF sizes across different datasets. Code available [https://github.com/Wensi-Tang/OS-CNN]
For some or all of the data instances a number of independent-world clustering issues suffer from incomplete data characterization due to losing or absent attributes. Typical clustering approaches cannot be applied directly to such data unless pre-processing by techniques like imputation or marginalization. We have overcome this drawback by utilizing a Sentenced Discrepancy Measure which we refer to as the Attribute Weighted Penalty based Discrepancy (AWPD). Using the AWPD measure, we modified the K-MEANS++ and Scalable K-MEANS++ for clustering algorithm and k Nearest Neighbor (kNN) for classification so as to make them directly applicable to datasets with non-existence attributes. We have presented a detailed theoretical analysis which shows that the new AWPD based K-MEANS++, Scalable K-MEANS++ and kNN algorithm merge into a local prime among the number of iterations is finite. We have reported in depth experiments on numerous benchmark datasets for various forms of Non-Existence showing that the projected clustering and classification techniques usually show better results in comparison to some of the renowned imputation methods that are generally used to process such insufficient data. This technique is designed to trace invaluable data to: directly apply our method on the datasets which have Non-Existence attributes and establish a method for detecting unstructured Non-Existence attributes with the best accuracy rate and minimum cost.
Sketchformer is a novel transformer-based representation for encoding free-hand sketches input in a vector form, i.e. as a sequence of strokes. Sketchformer effectively addresses multiple tasks: sketch classification, sketch based image retrieval (SBIR), and the reconstruction and interpolation of sketches. We report several variants exploring continuous and tokenized input representations, and contrast their performance. Our learned embedding, driven by a dictionary learning tokenization scheme, yields state of the art performance in classification and image retrieval tasks, when compared against baseline representations driven by LSTM sequence to sequence architectures: SketchRNN and derivatives. We show that sketch reconstruction and interpolation are improved significantly by the Sketchformer embedding for complex sketches with longer stroke sequences.
We propose \emph{MaxUp}, an embarrassingly simple, highly effective technique for improving the generalization performance of machine learning models, especially deep neural networks. The idea is to generate a set of augmented data with some random perturbations or transforms and minimize the maximum, or worst case loss over the augmented data. By doing so, we implicitly introduce a smoothness or robustness regularization against the random perturbations, and hence improve the generation performance. For example, in the case of Gaussian perturbation, \emph{MaxUp} is asymptotically equivalent to using the gradient norm of the loss as a penalty to encourage smoothness. We test \emph{MaxUp} on a range of tasks, including image classification, language modeling, and adversarial certification, on which \emph{MaxUp} consistently outperforms the existing best baseline methods, without introducing substantial computational overhead. In particular, we improve ImageNet classification from the state-of-the-art top-1 accuracy $85.5\%$ without extra data to $85.8\%$. Code will be released soon.
\textit{SummerTime} seeks to summarize globally time series signals and provides a fixed-length, robust summarization of the variable-length time series. Many classical machine learning methods for classification and regression depend on data instances with a fixed number of features. As a result, those methods cannot be directly applied to variable-length time series data. One common approach is to perform classification over a sliding window on the data and aggregate the decisions made at local sections of the time series in some way, through majority voting for classification or averaging for regression. The downside to this approach is that minority local information is lost in the voting process and averaging assumes that each time series measurement is equal in significance. Also, since time series can be of varying length, the quality of votes and averages could vary greatly in cases where there is a close voting tie or bimodal distribution of regression domain. Summarization conducted by the \textit{SummerTime} method will be a fixed-length feature vector which can be used in-place of the time series dataset for use with classical machine learning methods. We use Gaussian Mixture models (GMM) over small same-length disjoint windows in the time series to group local data into clusters. The time series' rate of membership for each cluster will be a feature in the summarization. The model is naturally capable of converging to an appropriate cluster count. We compare our results to state-of-the-art studies in physical activity classification and show high-quality improvement by classifying with only the summarization. Finally, we show that regression using the summarization can augment energy expenditure estimation, producing more robust and precise results.
Chest radiographs are among the most frequently acquired images in radiology and are often the subject of computer vision research. However, most of the models used to classify chest radiographs are derived from openly available deep neural networks, trained on large image-datasets. These datasets routinely differ from chest radiographs in that they are mostly color images and contain several possible image classes, while radiographs are greyscale images and often only contain fewer image classes. Therefore, very deep neural networks, which can represent more complex relationships in image-features, might not be required for the comparatively simpler task of classifying grayscale chest radiographs. We compared fifteen different architectures of artificial neural networks regarding training-time and performance on the openly available CheXpert dataset to identify the most suitable models for deep learning tasks on chest radiographs. We could show, that smaller networks such as ResNet-34, AlexNet or VGG-16 have the potential to classify chest radiographs as precisely as deeper neural networks such as DenseNet-201 or ResNet-151, while being less computationally demanding.
New generation head-mounted displays, such as VR and AR glasses, are coming into the market with already integrated eye tracking and are expected to enable novel ways of human-computer interaction in numerous applications. However, since eye movement properties contain biometric information, privacy concerns have to be handled properly. Privacy-preservation techniques such as differential privacy mechanisms have recently been applied to eye movement data obtained from such displays. Standard differential privacy mechanisms; however, are vulnerable due to temporal correlations between the eye movement observations. In this work, we propose a novel transform-coding based differential privacy mechanism to further adapt it to the statistics of eye movement feature data and compare various low-complexity methods. We extend the Fourier perturbation algorithm, which is a differential privacy mechanism, and correct a scaling mistake in its proof. Furthermore, we illustrate significant reductions in sample correlations in addition to query sensitivities, which provide the best utility-privacy trade-off in the eye tracking literature. Our results provide significantly high privacy without any essential loss in classification accuracies while hiding personal identifiers.
In multiclass classification, the goal is to learn how to predict a random label $Y$, valued in $\mathcal{Y}=\{1,\; \ldots,\; K \}$ with $K\geq 3$, based upon observing a r.v. $X$, taking its values in $\mathbb{R}^q$ with $q\geq 1$ say, by means of a classification rule $g:\mathbb{R}^q\to \mathcal{Y}$ with minimum probability of error $\mathbb{P}\{Y\neq g(X) \}$. However, in a wide variety of situations, the task targeted may be more ambitious, consisting in sorting all the possible label values $y$ that may be assigned to $X$ by decreasing order of the posterior probability $\eta_y(X)=\mathbb{P}\{Y=y \mid X \}$. This article is devoted to the analysis of this statistical learning problem, halfway between multiclass classification and posterior probability estimation (regression) and referred to as label ranking here. We highlight the fact that it can be viewed as a specific variant of ranking median regression (RMR), where, rather than observing a random permutation $\Sigma$ assigned to the input vector $X$ and drawn from a Bradley-Terry-Luce-Plackett model with conditional preference vector $(\eta_1(X),\; \ldots,\; \eta_K(X))$, the sole information available for training a label ranking rule is the label $Y$ ranked on top, namely $\Sigma^{-1}(1)$. Inspired by recent results in RMR, we prove that under appropriate noise conditions, the One-Versus-One (OVO) approach to multiclassification yields, as a by-product, an optimal ranking of the labels with overwhelming probability. Beyond theoretical guarantees, the relevance of the approach to label ranking promoted in this article is supported by experimental results.