作者：
Wang, YingxuCai, TonyZatarain, OmarUniv Calgary
Int Inst Cognit Informat & Cognit Comp ICIC Dept Elect & Comp Engn Schulich Sch Engn 2500 Univ Dr NW Calgary AB T2N 1N4 Canada Univ Calgary
Int Inst Cognit Informat & Cognit Comp ICIC Dept Elect & Comp Engn Hotchkiss Brain Inst 2500 Univ Dr NW Calgary AB T2N 1N4 Canada
A key challenge to sequencelearning for video comprehension is objects detection and localization in dynamic and real-time environment. This paper presents two methodological approaches to autonomous and generic obje...
详细信息
ISBN:
(纸本)9781728114194
A key challenge to sequencelearning for video comprehension is objects detection and localization in dynamic and real-time environment. This paper presents two methodological approaches to autonomous and generic object detection and localization in video sequences. Algorithms for both facial and non-facial object localization, as well as their integration, are developed. A set of experiments and case studies for practical video image processing is demonstrated for sequencelearning. This work paves a way to sequencelearning towards enhanced computer and robot vision technologies in applications of self-driving cars and real-time facial recognition.
sequencelearning is one of the hard challenges to current machine learning technologies and deep neural network technologies. This paper presents a literature survey and analysis on a variety of neural networks towar...
详细信息
ISBN:
(纸本)9781538633601
sequencelearning is one of the hard challenges to current machine learning technologies and deep neural network technologies. This paper presents a literature survey and analysis on a variety of neural networks towards sequencelearning. The conceptual models, methodologies, mathematical models and usages of classic neural networks and their learning capabilities are contrasted. Advantages and disadvantages of neural networks for sequencelearning are formally analyzed. The state-of-the-art, theoretical problems and technical constraints of existing methodologies are reviewed. The needs for understanding temporal sequences by unsupervised or intensive-training-free learning theories and technologies are elaborated.
sequencelearning from real-time videos is one of the hard challenges to current machine learning technologies and classic neural networks. Since existing supervised learning technologies are heavily dependent on inte...
详细信息
ISBN:
(纸本)9781728114194
sequencelearning from real-time videos is one of the hard challenges to current machine learning technologies and classic neural networks. Since existing supervised learning technologies are heavily dependent on intensive data and prior training, new methodologies for learning temporal sequences by unsupervised learning theories and technologies are yet to be developed. This paper presents the design and implementation of a novel Differential Neural Network (VNN) for unsupervised sequencelearning. The methodology is developed with a set of fundamental theories and enabling technologies for solving the problems of visual object recognition, motion detection, and visual semantic analysis in video sequence. A set of experiments on VNN for sequencelearning is demonstrated. This work has not only led to a theoretical breakthrough to novel machine sequencelearning, but also applicable to a wide range of challenging problems in computational intelligence and the AI industry.
暂无评论