A5100335187- Advanced Image and Video Retrieval Techniques
- Image Retrieval and Classification Techniques
- Video Analysis and Summarization
- Rough Sets and Fuzzy Logic
- Multimodal Machine Learning Applications
- Bayesian Methods and Mixture Models
- Adversarial Robustness in Machine Learning
- Domain Adaptation and Few-Shot Learning
- Neural Networks and Applications
- Video Surveillance and Tracking Methods
- Data Management and Algorithms
- Anomaly Detection Techniques and Applications
- Advanced Neural Network Applications
- Cognitive Computing and Networks
- Topic Modeling
- Face and Expression Recognition
- Algorithms and Data Compression
- Optical measurement and interference techniques
- Web Data Mining and Analysis
- Generative Adversarial Networks and Image Synthesis
- Constraint Satisfaction and Optimization
- Natural Language Processing Techniques
- Gaussian Processes and Bayesian Inference
- Digital Media Forensic Detection
- Image Processing Techniques and Applications
Shanghai Micro Satellite Engineering Center
2024
Tsinghua University
2014-2023
Sichuan University
2023
Donghua University
2022
Individual Differences
2020
Shanghai Institute of Measurement and Testing Technology
2020
Weatherford College
2020
University of Shanghai for Science and Technology
2018
Intelligent Health (United Kingdom)
2009-2016
The decomposition-based multiobjective evolutionary algorithms (MOEAs) generally make use of aggregation functions to decompose a optimization problem into multiple single-objective problems. However, due the nature contour lines for adopted functions, they usually fail preserve diversity in high-dimensional objective space even by using diverse weight vectors. To address this problem, we propose maintain desired solutions their process explicitly exploiting perpendicular distance from...
Multicomponent therapeutics offer bright prospects for the control of complex diseases in a synergistic manner. However, finding ways to screen combinations from numerous pharmacological agents is still an ongoing challenge. In this work, we proposed first time “network target”-based paradigm instead traditional "single target"-based virtual screening and established algorithm termed NIMS (Network target-based Identification Synergy) prioritize agent high throughput way. treats...
This paper conducts a formal study of the shot boundary detection problem. First, general framework techniques is proposed. Three critical techniques, i.e., representation visual content, construction continuity signal and classification values, are identified formulated in perspective pattern recognition. Meanwhile, major challenges to identified. Second, comprehensive review existing approaches conducted. The representative categorized compared according their roles framework. Based on...
Nowadays, cameras equipped with AI systems can capture and analyze images to detect people automatically. However, the system make mistakes when receiving deliberately designed patterns in real world, i.e., physical adversarial examples. Prior works have shown that it is possible print patches on clothes evade DNN-based person detectors. these examples could catastrophic drops attack success rate viewing angle (i.e., camera's towards object) changes. To perform a multi-angle attack, we...
Traditional relation extraction methods require pre-specified relations and relation-specific human-tagged examples. Bootstrapping systems significantly reduce the number of training examples, but they usually apply heuristic-based to combine a set strict hard rules, which limit ability generalize thus generate low recall. Furthermore, existing bootstrapping do not perform open information (Open IE), can identify various types without requiring pre-specifications. In this paper, we propose...
As a special topic in computer vision, fine-grained visual categorization (FGVC) has been attracting growing attention these years. Different with traditional image classification tasks which objects have large inter-class variation, the concepts datasets, such as hundreds of bird species, often very similar semantics. Due to similarity, it is difficult classify without locating really discriminative features, therefore becomes more important for algorithm make full use part information...
In this paper, we demonstrate that the essentials of image classification and retrieval are same, since both tasks could be tackled by measuring similarity between images. To end, propose ONE (Online Nearest-neighbor Estimation), a unified algorithm for retrieval. is surprisingly simple, which only involves manual object definition, regional description nearest-neighbor search. We take advantage PCA PQ approximation GPU parallelization to scale our up large-scale Experimental results verify...
Interpretability of deep neural networks (DNNs) is essential since it enables users to understand the overall strengths and weaknesses models, conveys an understanding how models will behave in future, diagnose correct potential problems. However, challenging reason about what a DNN actually does due its opaque or black-box nature. To address this issue, we propose novel technique improve interpretability DNNs by leveraging rich semantic information embedded human descriptions. By...
Interpretability of a deep neural network aims to explain the rationale behind its decisions and enable users understand intelligent agents, which has become an important issue due importance in practical applications. To address this issue, we develop Distillation Guided Routing method, is flexible framework interpret by identifying critical data routing paths analyzing functional processing behavior corresponding layers. Specifically, propose discover nodes on during inferring prediction...
Abstract The explosive growth in data volume and the availability of cheap computing resources have sparked increasing interest Big learning, an emerging subfield that studies scalable machine learning algorithms, systems applications with Data. Bayesian methods represent one important class statistical for substantial recent developments on adaptive, flexible learning. This article provides a survey advances methods, termed Learning, including non-parametric adaptively inferring model...
Adversarial attacks can easily fool object recognition systems based on deep neural networks (DNNs). Although many defense methods have been proposed in recent years, most of them still be adaptively evaded. One reason for the weak adversarial robustness may that DNNs are only supervised by category labels and do not part-based inductive bias like process humans. Inspired a well-known theory cognitive psychology – recognition-by-components, we propose novel model ROCK (Recognizing Object...
Cognitive informatics is a transdisciplinary enquiry of computer science, information sciences, cognitive and intelligence science that investigates the internal processing mechanisms processes brain natural intelligence, as well their engineering applications in computing. computing an emerging paradigm intelligent methodologies systems based on implements computational by autonomous inferences perceptions mimicking brain. This article presents set collective perspectives computing,...
The contemporary wonder of sciences and engineering has recently refocused on the beginning point of: how brain processes internal external information autonomously cognitively rather than imperatively like conventional computers. Cognitive Informatics (CI) is a transdisciplinary enquiry computer science, sciences, cognitive intelligence science that investigates processing mechanisms natural intelligence, as well their applications in computing. This paper reports set eight position...
Freeform optics, due to the more general surface geometry that offers high degrees of design freedom control light propagation, has already been widely used in both nonimaging optics and imaging optics. With recent advances fabrication freeform one remaining challenges is how accurately measure optical surfaces, especially those included refractive To meet this imperative need, for first time, we believe, present an effective simultaneous multisurface measurement method Instead using a...
Though both quantity and quality of semantic concept detection in video are continuously improving, it still remains unclear how to exploit these detected concepts as indices search, given a specific query. In this paper, we tackle problem propose search framework which operates like searching text documents. Noteworthy for its adoption the well-founded principles, first selects few related query, by employing tf-idf scheme, called c-tf-idf, measure informativeness These selected form...
Abstract Background Drug combination therapy is commonly used in clinical practice. Many methods including Bliss independence method have been proposed for drug design based on simulations models or experiments. Although can help to solve the problem when there are only a small number of combinations, as combinations increases, it may not be scalable. Exploration system structure becomes important reduce complexity problem. Results In this paper, we deduced mathematical model which simplify...
Min-wise hash is a widely-used hashing method for scalable similarity search in terms of Jaccard similarity, while practice it necessary to compute many such functions certain precision, leading expensive computational cost. In this paper, we introduce an effective method, i.e. the min-max which significantly reduces time by half, yet has provably slightly smaller variance estimating pair wise similarity. addition, estimator only contains equality checking, thus especially suitable...
By highlighting important features that contribute to model prediction, visual saliency is used as a natural form interpret the working mechanism of deep neural networks. Numerous methods have been proposed achieve better results. However, we find previous are not reliable enough provide meaningful interpretation through simple sanity check: required explain output non-maximum prediction classes, which usually ground-truth classes. For example, let an image "dog" given wrong class label...