Financial time-series forecasting has long been a challenging problem because of the inherently noisy and stochastic nature market. In high-frequency trading, for trading purposes is even more task, since an automated inference system required to be both accurate fast. this paper, we propose neural network layer architecture that incorporates idea bilinear projection as well attention mechanism enables detect focus on crucial temporal information. The resulting highly interpretable, given...

The traditional multilayer perceptron (MLP) using a McCulloch-Pitts neuron model is inherently limited to set of neuronal activities, i.e., linear weighted sum followed by nonlinear thresholding step. Previously, generalized operational (GOP) was proposed extend the conventional defining diverse activities imitate biological neurons. Together with GOP, progressive (POP) algorithm optimize predefined template multiple homogeneous layers in layerwise manner. In this paper, we propose an...

Nowadays, with the availability of massive amount trade data collected, dynamics financial markets pose both a challenge and an opportunity for high frequency traders. In order to take advantage rapid, subtle movement assets in High Frequency Trading (HFT), automatic algorithm analyze detect patterns price change based on transaction records must be available. The multichannel, time-series representation naturally suggests tensor-based learning algorithms. this work, we investigate...

Compressive Learning is an emerging topic that combines signal acquisition via compressive sensing and machine learning to perform inference tasks directly on a small number of measurements. Many data modalities naturally have multi-dimensional or tensorial format, with each dimension tensor mode representing different features such as the spatial temporal information in video sequences spectral hyperspectral images. However, existing frameworks, component utilizes either random learned...

Face verification is a prominent biometric technique for identity authentication that has been used extensively in several security applications. In practice, face often performed along with other visual surveillance tasks the computing device. Thus, ability to share computation and reuse information already extracted analysis can greatly help reduce load on devices. this study, we propose utilize knowledge transfer approach problem by building heterogeneous neural network architecture of...

Deep Learning models have become dominant in tackling financial time-series analysis problems, overturning conventional machine learning and statistical methods. Most often, a model trained for one market or security cannot be directly applied to another due differences inherent the conditions. In addition, as evolves over time, it is necessary update existing train new ones when data made available. This scenario, which most forecasting applications, naturally raises following research...

Learning to rank is an essential component in information retrieval system. The state-of-the-art ranking systems are often based on ensemble of classifiers, such as Random Forest or LambdaMART, which aggregates the outputs produced by thousands classifiers. storage and computation requirement model usually very high, imposing a significant operating cost To tackle this problem, we propose algorithm that adaptively learns single heterogeneous feedforward network architecture, composing...

Financial time-series forecasting is one of the most challenging domains in field analysis. This mostly due to highly non-stationary and noisy nature financial data. With progressive efforts community design specialized neural networks incorporating prior domain knowledge, many analysis problems have been successfully tackled. The temporal attention mechanism a layer that recently gained popular-ity its ability focus on important events. In this paper, we propose based ideas multi-head...

In this paper, we propose 2D-Attention (2DA), a generic attention formulation for sequence data, which acts as complementary computation block that can detect and focus on relevant sources of information the given learning objective. The proposed module is incorporated into recently Neural Bag Feature (NBoF) model to enhance its capacity. Since 2DA plug-in layer, injecting it different stages NBoF results in 2DA-NBoF architectures, each possesses unique interpretation. We conducted extensive...

The visit patterns of insects to specific flowers at times during the diurnal cycle and across season play important roles in pollination biology. Thus, ability automatically detect visitors occurring video sequences greatly reduces manual human efforts needed collect such data. Data-dependent approaches, as supervised machine learning algorithms, have become core component several automation systems. In this paper, we describe a flower visitor detection system using deep Convolutional...

Financial time-series analysis and forecasting have been extensively studied over the past decades, yet still remain as a very challenging research topic. Since financial market is inherently noisy stochastic, majority of interests are non-stationary, often obtained from different modalities. This property presents great challenges can significantly affect performance subsequent analysis/forecasting steps. Recently, Temporal Attention augmented Bilinear Layer (TABL) has shown performances in...

Recently, the Multilinear Compressive Learning (MCL) framework was proposed to efficiently optimize sensing and learning steps when working with multidimensional signals, i.e. tensors. In in general, MCL particular, number of compressed measurements captured by a compressive device characterizes storage requirement or bandwidth for transmission. This number, however, does not completely characterize performance system. this paper, we analyze relationship between input signal resolution, MCL....

Material is one of the intrinsic features objects, and consequently material recognition plays an important role in image understanding. The same may have various shapes appearance, while keeping physical characteristic. This brings great challenges for recognition. Besides suitable features, a powerful classifier also can improve overall performance. Due to limitations classical linear neurons, used all shallow deep neural networks, such as CNN, we propose apply generalized operational...

Deep Learning models have become dominant in tackling financial time-series analysis problems, overturning conventional machine learning and statistical methods. Most often, a model trained for one market or security cannot be directly applied to another due differences inherent the conditions. In addition, as evolves over time, it is necessary update existing train new ones when data made available. This scenario, which most forecasting applications, naturally raises following research...

Multilinear compressive learning (MCL) is an efficient signal acquisition and paradigm for multidimensional signals. The level of compression affects the detection or classification performance MCL model, with higher rates often associated lower inference accuracy. However, are more amenable to a wider range applications, especially those that require low operating bandwidth minimal energy consumption such as Internet Things (IoT) applications. Many communication protocols provide support...

Progressive Neural Network Learning is a class of algorithms that incrementally construct the network's topology and optimize its parameters based on training data. While this approach exempts users from manual task designing validating multiple network topologies, it often requires an enormous number computations. In paper, we propose to speed up process by exploiting subsets data at each incremental step. Three different sampling strategies for selecting samples according criteria are...