The recent success of deep learning is driving a trend towards structurally complex computer vision models that combine feature extraction with predictive elements into integrated pipelines. While some of these models have achieved breakthrough results in applications like object recognition, they are difficult to design and tune, impeding progress. We feel that visual analysis can be a powerful tool to aid iterative development of deep model pipelines. Building on feature evaluation work in the computer vision community, we introduce ML-o-scope, an interactive visualization system for exploratory analysis of convolutional neural networks, a prominent type of pipelined model. We present ML-o-scope’s time-lapse engine that provides views into model dynamics during training, and evaluate the system as a support for tuning large scale object-classification pipelines.

关键词：深度机器学习；深度学习；卷积神经网络；可视化

查看摘要 索取原文[11页]

Digital hardware frontend specification is defined by two extremes - RTL specification and HLS specification. RTL specification produces highly optimized designs, but requires extremely verbose low level specification on the part of the designer. On the other hand, HLS specification requires much less verbose high level specification from the designer, but does not produce well optimized designs. In this thesis, I present a middle ground that captures the pros of these two extremes - a system in which the designer specifies a basic datapath and uses a tool to automatically apply optimizations to that basic datapath.

关键词：RTL；HLS；数据通路

查看摘要 索取原文[19页]

We begin this thesis with the development of a system that is capable of categorizing objects and landmarks in an efficient and distributed manner.In the second part of the thesis we extend our distributed object categorization framework to address the problem of automatic human activity detection and categorization.In most realistic scenarios that require situation awareness, objects and people occur in cluttered scenes and exhibit immense variability in their appearance, shape and pose. In the final part of this thesis, we analyze the interplay between computer vision tasks such as segmentation and categorization and present joint frameworks that significantly improve the performance of each task.

关键词：态势感知；局部感知；全局干涉；网络安全；智能相机

查看摘要 索取原文[79页]

In this capstone project, we demonstrate a design for combining spike sorting and compressed sensing algorithms in a DSP chip that can perform data compression by achieving accurate data reconstruction with injected noise. The resulting circuit architecture is implemented in a 32nm CMOS process.

关键词：生物信号；信号分类；压缩传感；ASIC

查看摘要 索取原文[23页]

This paper describes a natural language processing system that generates patterns from labeled training data and uses them to extract chemical reactions from PubMed. To train and validate our system, we create a dataset using BRENDA, the BRaunschweig ENzyme DAtabase, with 4387 labeled sentences. Our system achieves a recall of 0.82 and a precision of 0.88 via cross validation. On a selection of 600,000 PubMed abstracts, our system extracts almost 20% of existing reactions in BRENDA as well as many that are novel.

关键词：自然语言处理；化学反应；生物学；学术文献

查看摘要 索取原文[16页]

Model search is a crucial component of data analytics pipelines, and this laborious process of choosing an appropriate learning algorithm and tuning its parameters remains a major obstacle in the widespread adoption of machine learning techniques. Recent efforts aiming to automate this process have assumed model training itself to be a black-box, thus limiting the effectiveness of such approaches on large-scale problems. In this work, we build upon these recent efforts. By inspecting the inner workings of model training and framing model search as bandit-like resource allocation problem, we present an integrated distributed system for model search that targets large-scale learning applications. We study the impact of our approach on a variety of datasets and demonstrate that our system, named GHOSTFACE, solves the model search problem with comparable accuracy as basic strategies but an order of magnitude faster. We further demonstrate that GHOSTFACE can scale to models trained on terabytes of data across hundreds of machines.

关键词：模型训练；模型检索；集散系统；可伸缩

查看摘要 索取原文[13页]

We propose a new segment-based registration system for aerial images of the same scene taken at different times, from different view points, and/or by different sensors. We introduce a quantitative characterization of the registration difficulty for a given pair of images. Targeting high registration difficulty input, we exploit on linear edges in images. In the first step of our registration process, we detect line segments in each image. Next we conduct a merging step on the detected line segments. Finally, using the merged line segments as input, we generate possible hypothesis transformations by choosing three segments in each image. Our collinearity score metric for the transformations balances considerations of angular and perpendicular distances.

关键词：航拍图像配准；线段识别；特征识别；线段合并

查看摘要 索取原文[34页]

In this work, a comprehensive surface mount technology (SMT) for rigid and silicon dioxide based ICs onto flexible and glycol-modified Polyethylene Terephthalate screen-printed interconnects is developed. Here, the SMT solution addresses the following challenges: low thermal budget of Polyethylene Terephthalate, non-solderability of the screen-printed structures, aluminum coating of the IC, and small-sized pitches of the IC. Besides, various reliability aspects of the assembled electrical end product are investigated.

关键词：表面安装技术；电气互联；集成电路；柔性屏幕；印刷

查看摘要 索取原文[34页]

In this work, a temperature sensor based on 4H-SiC pn diode is theoretically and experimentally investigated. The device is capable of stable operation in a temperature range from 20 °C up to 600 °C. In the forward biased region, the forward voltage of the 4H-SiC pn diode shows linear dependence on temperature at a constant current. This dependence is utilized to sense temperature variations and the proposed device achieves a sensitivity of 3.5 mV/°C. These results indicate that an integrated circuit compatible temperature sensor based on 4H-SiC pn diode is a promising technology for harsh environment sensing applications.

关键词：温度传感器；半导体二极管；PN结；温度传感器； 恶劣环境

查看摘要 索取原文[27页]

This paper focuses on CPS modeling, which is an essential activity in CPS design, with multiple challenges. In particular, stakeholders lack a systematic framework and guidelines to help them choose among the many available modeling languages and tools. We propose such a framework in this paper. Our framework consists of three elements: viewpoints, which capture the stakeholders' interests and concerns; concrete languages and tools, among which the stakeholders must make a selection when defining their CPS design environments; and abstract, mathematical formalisms, which are the "semantic glue" linking the two worlds. As part of the framework, we survey various formalisms, languages, and tools and explain how they are related. We also provide examples of viewpoints and discuss how they are related to formalisms.

关键词：赛博物理系统；微分代数方程； 常微分方程；EOO；HDLs

查看摘要 索取原文[7页]