Decomposition and replacement: Spatial knowledge distillation for monocular depth estimation

Knowledge distillation has become a key technique for making smart and light-weight networks through model compression and transfer learning. Unlike previous methods that applied knowledge distillation to the classification task, we propose to exploit the decomposition-and-replacement based distillation scheme for depth estimation from a single RGB color image. To do this, Laplacian pyramid-based knowledge distillation is firstly presented in this paper. The key idea of the proposed method is to transfer the rich knowledge of the scene depth, which is well encoded through the teacher network, to the student network in a structured way by decomposing it into the global context and local details. This is fairly desirable for the student network to restore the depth layout more accurately with limited resources. Moreover, we also propose a new guidance concept for knowledge distillation, so-called ReplaceBlock, which replaces blocks randomly selected in the decoded feature of the student network with those of the teacher network. Our ReplaceBlock gives a smoothing effect in learning the feature distribution of the teacher network by considering the spatial contiguity in the feature space. This process is also helpful to clearly restore the depth layout without the significant computational cost. Based on various experimental results on benchmark datasets, the effectiveness of our distillation scheme for monocular depth estimation is demonstrated in details. The code and model are publicly available at : https://github.com/tjqansthd/Lap_Rep_KD_Depth.     

Learn decision trees with deep visual primitives

In this paper, we strive to propose a self-interpretable framework, termed PrimitiveTree, that incorporates deep visual primitives condensed from deep features with a conventional decision tree, bridging the gap between deep features extracted from deep neural networks (DNNs) and trees’ transparent decision-making processes. Specifically, we utilize a codebook, which embeds the continuous deep features into a finite discrete space (deep visual primitives) to distill the most common semantic information. The decision tree adopts the spatial location information and the mapped primitives to present the decision-making process of the deep features in a tree hierarchy. Moreover, the trained interpretable PrimitiveTree can inversely explain the constituents of the deep features, highlighting the most critical and semantic-rich image patches attributing to the final predictions of the given DNN. Extensive experiments and visualization results validate the effectiveness and interpretability of our method.     

Analyzing dissipation effects on the non-classicality witnessing by a qubit via the entangled state representation

ABSTRACT

In this paper, we study detection of the state non-classicality for a quantum harmonic oscillator by a qubit in the presence of dissipation effects. We show that dissipation can enhance the effectiveness of the method in case of using the corrected form of the related nonclassicality witness. Such an improvement is attributed to the fact that dissipation leads to probing a surface, instead of a curve, of the complex plane for non-classicality condition on normally-ordered characteristic function.     

Applied data science in patient-centric healthcare: Adaptive analytic systems for empowering physicians and patients

We define the emerging research field of applied data science as the knowledge discovery process in which analytic systems are designed and evaluated to improve the daily practices of domain experts. We investigate adaptive analytic systems as a novel research perspective of the three intertwining aspects within the knowledge discovery process in healthcare: domain and data understanding for physician- and patient-centric healthcare, data preprocessing and modelling using natural language processing and (big) data analytic techniques, and model evaluation and knowledge deployment through information infrastructures. We align these knowledge discovery aspects with the design science research steps of problem investigation, treatment design, and treatment validation, respectively. We note that the adaptive component in healthcare system prototypes may translate to data-driven personalisation aspects including personalised medicine. We explore how applied data science for patient-centric healthcare can thus empower physicians and patients to more effectively and efficiently improve healthcare. We propose meta-algorithmic modelling as a solution-oriented design science research framework in alignment with the knowledge discovery process to address the three key dilemmas in the emerging “post-algorithmic era” of data science: depth versus breadth, selection versus configuration, and accuracy versus transparency.     

基于伪布尔模型和启发式算法求解无容量设施选址问题

采用伪布尔模型和启发式算法来求解无容量设施选址问题。首先给出了问题的伪布尔(pseudo-Boolean)表示，然后基于Khumawala规则对问题进行预处理，最后提出两种启发式分支准则来求解问题。实验结果表明所提算法简单有效。     

Examining the role of regional culture and geographical distances on the representation of unfamiliar foods in a continental-size country

This study provides an experimental-exploratory investigation about the role of regional culture and Euclidean distances on the consumers’ representation of edible insects in Brazil, a country with an extensive geographical surface. Seven hundred and eighty participants were recruited on the streets of eight cities from different Brazilian states: Manaus in Amazonas; Porto Velho in Rondônia; Macapá in Amapá; Cuiabá in Mato Grosso; Aracaju in Sergipe; Rio de Janeiro in Rio de Janeiro; Campinas in São Paulo; and Santa Maria in Rio Grande do Sul. These participating cities were considered from their cultural identity differences and geographical distances. Through a continual restricted word association task, participants were instructed to promptly verbalize the first five terms that came to their minds when stimulated with the expression “food made with edible insects”. Following, they had to score the valence of each term they produced. The dictionaries produced in each city were compared and classified into groups using the Ellegård’s index. Each group presented distinct ways of expression and attitude with respect to the inductive expression. Basically, Brazil was divided into two main groups according to their representation of edible insects: one consisted by the cities situated near the shore of the Atlantic Ocean, which present a cultural formation influenced by the European immigrants; and the other comprised the cities from the continental region that have strong cultural influence from the Amerindians. Thus, the cultural formation was more decisive to explain the similar representations among the cities than their geographical proximity. Given that, to effectively introduce a novel food in a country with varied regional culture, the marketing strategy should be focused on the values and beliefs of their culture subgroups instead of a single strategy for the whole country.     

Hierarchical learning using deep optimum-path forest

Bag-of-Visual Words (BoVW) and deep learning techniques have been widely used in several domains, which include computer-assisted medical diagnoses. In this work, we are interested in developing tools for the automatic identification of Parkinson’s disease using machine learning and the concept of BoVW. The proposed approach concerns a hierarchical-based learning technique to design visual dictionaries through the Deep Optimum-Path Forest classifier. The proposed method was evaluated in six datasets derived from data collected from individuals when performing handwriting exams. Experimental results showed the potential of the technique, with robust achievements.     

Exploiting Bayesian networks for fault isolation: A diagnostic case study of diesel fuel injection system

Fault isolation is known to be a challenging problem in machinery troubleshooting. It is not only because the isolation of multiple faults contains considerable number of uncertainties due to the strong correlation and coupling between different faults, but often massive prior knowledge is needed as well. This paper presents a Bayesian network-based approach for fault isolation in the presence of the uncertainties. Various faults and symptoms are parameterized using state variables, or the so-called nodes in Bayesian networks (BNs). Probabilistically causality between a fault and a symptom and its quantization are described respectively by a directed edge and conditional probability. To reduce the qualitative and quantitative knowledge needed, particular considerations are given to the simplification of Bayesian networks structures and conditional probability expressions using rough sets and noisy-OR/MAX model, respectively. By adopting the simplified approach, symptoms under multiple-fault are decoupled into the ones under every single fault, while the quantity of the conditional probabilities is simplified into the linear form of the faults quantity. Prior knowledge needed in Bayesian network-based diagnostic model is reduced significantly, which decreases the complexity in establishing and applying this diagnosis model. The computational efficiency is improved accordingly in the simplified BN model, after eliminating the redundant symptoms. The fault isolation methodology is illustrated through an example of diesel engine fuel injection system to verify the developed model.     

面向多源数据融合的稀疏表示目标跟踪

传统的基于稀疏表示的目标跟踪方法主要利用目标的灰度特征构建稀疏表示模型。由于灰度特征对光照变化敏感，这会影响目标跟踪在复杂场景下的鲁棒性。基于多源数据融合的目标跟踪可以明显提升目标跟踪鲁棒性，但如何有效融合不同维度，不同类型的多源目标特征成为基于多源数据融合的目标跟踪所要解决的关键问题。提出了一个基于目标状态以及灰度特征的稀疏表示目标跟踪方法。所提出的方法可通过基于核函数表示的稀疏表示模型，在探究目标状态以及灰度特征相关性的基础上，将两种不同维度的特征进行有效融合，提升目标跟踪在复杂场景下的鲁棒性。