In the absence of explicit regularization, Kernel "Ridgeless" Regression with nonlinear kernels has the potential to fit the training data perfectly. It has been observed empirically, however, that such interpolated solutions can still generalize well on test data. We isolate a phenomenon of implicit regularization for minimum-norm interpolated solutions which is due to a combination of high dimensionality of the input data, curvature of the kernel function, and favorable geometric properties of the data such as an eigenvalue decay of the empirical covariance and kernel matrices. In addition to deriving a data-dependent upper bound on the out-of-sample error, we present experimental evidence suggesting that the phenomenon occurs in the MNIST dataset.

With only a subset of its entries available, matrix completion (MC) amounts to recovering the unavailable entries by leveraging just the low-rank attribute of the matrix itself [1]. The relevant task arises in applications as diverse as image restoration [2], sensor networks [3], and recommender systems [4]. To save power for instance, only a fraction of sensors may collect and transmit measurements to a fusion center, where the available spatiotemporal data can be organized in a matrix format, and the unavailable ones can be eventually interpolated via MC [3]. Similarly, collaborative filtering of ratings given by users to a small number of items are stored in a sparse matrix, and the objective is to predict their ratings for the rest of the items [4]. Existing MC approaches rely on some form of rank minimization or low-rank matrix factorization. Specifically, [1] proves that when MC is formulated as the minimization of the nuclear norm subject to the constraint that the observed entries remain unchanged, exact recovery is possible under mild assumptions; see also [5] where reliable recovery from a few observations is established even in the presence of additive noise. Alternatively, [4] replaces the nuclear norm by a product of two low-rank factor matrices that are identified in order to recover the complete matrix. While the low-rank assumption can be sufficient for reliable recovery, prior information about the unknown matrix can be also accounted to improve the completion outcome. Forms of prior information can include sparsity [3], local smoothness [6], and interdependencies encoded by graphs [7]-[10].

This study evaluates the performances of CNN and LSTM for recognizing common charts patterns in a stock historical data. It presents two common patterns, the method used to build the training set, the neural networks architectures and the accuracies obtained.

We formulate simple assumptions, implying the Robbins-Monro conditions for the $Q$-learning algorithm with the local learning rate, depending on the number of visits of a particular state-action pair (local clock) and the number of iteration (global clock). It is assumed that the Markov decision process is communicating and the learning policy ensures the persistent exploration. The restrictions are imposed on the functional dependence of the learning rate on the local and global clocks. The result partially confirms the conjecture of Bradkte (1994).

Accurate on-device keyword spotting (KWS) with low false accept and false reject rate is crucial to customer experience for far-field voice control of conversational agents. It is particularly challenging to maintain low false reject rate in real world conditions where there is (a) ambient noise from external sources such as TV, household appliances, or other speech that is not directed at the device (b) imperfect cancellation of the audio playback from the device, resulting in residual echo, after being processed by the Acoustic Echo Cancellation (AEC) system. In this paper, we propose a data augmentation strategy to improve keyword spotting performance under these challenging conditions. The training set audio is artificially corrupted by mixing in music and TV/movie audio, at different signal to interference ratios. Our results show that we get around 30-45% relative reduction in false reject rates, at a range of false alarm rates, under audio playback from such devices.

Clustering is an extensive research area in data science. The aim of clustering is to discover groups and to identify interesting patterns in datasets. Crisp (hard) clustering considers that each data point belongs to one and only one cluster. However, it is inadequate as some data points may belong to several clusters, as is the case in text categorization. Thus, we need more flexible clustering. Fuzzy clustering methods, where each data point can belong to several clusters, are an interesting alternative. Yet, seeding iterative fuzzy algorithms to achieve high quality clustering is an issue. In this paper, we propose a new linear and efficient initialization algorithm MaxMin Linear to deal with this problem. Then, we validate our theoretical results through extensive experiments on a variety of numerical real-world and artificial datasets. We also test several validity indices, including a new validity index that we propose, Transformed Standardized Fuzzy Difference (TSFD).

In this work we study an application of machine learning to the construction industry and we use classical and modern machine learning methods to categorize images of building designs into three classes: Apartment building, Industrial building or Other. No real images are used, but only images extracted from Building Information Model (BIM) software, as these are used by the construction industry to store building designs. For this task, we compared four different methods: the first is based on classical machine learning, where Histogram of Oriented Gradients (HOG) was used for feature extraction and a Support Vector Machine (SVM) for classification; the other three methods are based on deep learning, covering common pre-trained networks as well as ones designed from scratch. To validate the accuracy of the models, a database of 240 images was used. The accuracy achieved is 57% for the HOG + SVM model, and above 89% for the neural networks.

Convolutional neural networks have recently achieved significant breakthroughs in various image classification tasks. However, they are computationally expensive,which can make their feasible mplementation on embedded and low-power devices difficult. In this paper convolutional neural network binarization is implemented on GPU-based platforms for real-time inference on resource constrained devices. In binarized networks, all weights and intermediate computations between layers are quantized to +1 and -1, allowing multiplications and additions to be replaced with bit-wise operations between 32-bit words. This representation completely eliminates the need for floating point multiplications and additions and decreases both the computational load and the memory footprint compared to a full-precision network implemented in floating point, making it well-suited for resource-constrained environments. We compare the performance of our implementation with an equivalent floating point implementation on one desktop and two embedded GPU platforms. Our implementation achieves a maximum speed up of 7. 4X with only 4.4% loss in accuracy compared to a reference implementation.

Spectral Mixture (SM) kernels form a powerful class of kernels for Gaussian processes, capable to discover patterns, extrapolate, and model negative co-variances. In SM kernels, spectral mixture components are linearly combined to construct a final flexible kernel. As a consequence SM kernels does not explicitly model correlations between components and dependencies related to time and phase delays between components, because only the auto-convolution of base components are used. To address these drawbacks we introduce Generalized Convolution Spectral Mixture (GCSM) kernels. We incorporate time and phase delay into the base spectral mixture and use cross-convolution between a base component and the complex conjugate of another base component to construct a complex-valued and positive definite kernel representing correlations between base components. In this way the total number of components in GCSM becomes quadratic. We perform a thorough comparative experimental analysis of GCSM on synthetic and real-life datasets. Results indicate the beneficial effect of the extra features of GCSM. This is illustrated in the problem of forecasting the long range trend of a river flow to monitor environment evolution, where GCSM is capable of discovering correlated patterns that SM cannot and improving patterns recognition ability of SM.

Professional sports are developing towards increasingly scientific training methods with increasing amounts of data being collected from laboratory tests, training sessions and competitions. In cycling, it is standard to equip bicycles with small computers recording data from sensors such as power-meters, in addition to heart-rate, speed, altitude etc. Recently, machine learning techniques have provided huge success in a wide variety of areas where large amounts of data ("big data") is available. In this paper, we perform a pilot experiment on machine learning to model physical response in elite cyclists. As a first experiment, we show that it is possible to train a LSTM machine learning algorithm to predict the heart-rate response of a cyclist during a training session. This work is a promising first step towards developing more elaborate models based on big data and machine learning to capture performance aspects of athletes.