Restricted Boltzmann machines (RBMs) and their variants are usually trained by contrastive divergence (CD) learning, but the training procedure is an unsupervised learning approach, without any guidances of background knowledge. To enhance expression ability traditional RBMs, in this paper, we propose pairwise constraints (PCs) RBM with Gaussian visible units (pcGRBM) model, which guided PCs process encoding conducted under these guidances. The encoded hidden layer features pcGRBM. Then,...

The instability is shown in the existing methods of representation learning based on Euclidean distance under a broad set conditions. Furthermore, scarcity and high cost labels prompt us to explore more expressive which depends as few possible. To address above issues, small-perturbation ideology firstly introduced model probability distribution. positive information (SPI) only depend two each cluster used stimulate distribution then variant models are proposed fine-tune expected Restricted...

Reinforcement Learning (RL) has been used to implement the perception-action cycle of multi-input-multi-output (MIMO) cognitive radar (CR). This allows for adaptive optimization radar's beampattern, which is guided by information from echoes and an appropriate reward signal. However, present approaches rely on a greedy DOA estimation select candidate angles, means that decisions are primarily based detection results last pulse. If system misses target, it can be time-consuming recapture it,...

In this letter, a low-rank and sparse representation classifier with spectral consistency constraint (LRSRC-SCC) is proposed. Different from the SRC that represents samples individually, LRSRC-SCC reconstructs jointly able to capture local global structures simultaneously. proposed classifier, an adaptive imposed on both terms so as better reveal data structure enhance its discriminative power. addition, alternating direction method introduced solve underlying minimization problem, in which,...

Representing belief information is a fundamental problem in the field of revision. The AGM framework uses deductively closed set formulas, known as theory, to represent an agent, because rational agent should satisfy properties similar theory. However, iterated revision setting, DP conditional beliefs like (φ | ψ) such information, which not natural, are formulas and logical connections between them cannot be characterized clearly. In this paper, we propose novel logic system for...

In order to properly regulate iterated belief revision, Darwiche and Pearl (1997) model revision as revising epistemic states by propositions. An state in their sense consists of a set conditional beliefs. Although the denotation an can be indirectly captured total preorder on worlds, it is unclear how directly capture structure terms beliefs contains. this paper, we first provide axiomatic characterisation for using nine rules about beliefs, then argue that last two are too strong should...

Dimensionality reduction is a fundamental and important research topic in the field of machine learning. This paper focuses on dimensionality technique that exploits semi-supervising information form pairwise constraints; specifically, these constraints specify whether two instances belong to same class or not. We propose dual linear methods accomplish under setting. These overcome difficulty maximizing between-class difference minimizing within-class at time, by transforming original data...

In the area of network intrusion detection, harmful cyber-attacks are usually rare when compared to normal samples, which makes feature extraction and detection difficult. Increasing amount anomalous samples through data generation can effectively relieve class imbalance, thus significantly improve classifier's performance. Variational Autoencoder (VAE) is an effective model that assumes latent variables satisfy one prior distribution creates new by decoding sampled from distribution. This...

Aiming at the power plant energy consumption and gas balance influenced serious with affluent fluctuate frequently of byproduct system in an iron steel industry, which is very difficult to be modeled using mechanism modeling, a forecast trend sequence supply HP-ENN model was established based on characteristics self-provided utilization properties HP filter, Elman neural network. The prediction results practical production data show that proposed HP-Elman method sample A 48, 60 points...

Lightweight networks with simplified structure and fewer parameters ensure fast computation response time on mobile devices, making them well-suited for finger vein recognition tasks. However, these are limited by their parameter settings, leading to a lack of network generalization ability. Furthermore, they primarily focus extracting semantic features, it challenging learn topological features such as the number connectivity veins within image. Inspired remarkable performance topology in...