Computer Engineering Commons^™

Neuron Clustering For Mitigating Catastrophic Forgetting In Supervised And Reinforcement Learning, Benjamin Frederick Goodrich

Doctoral Dissertations

Neural networks have had many great successes in recent years, particularly with the advent of deep learning and many novel training techniques. One issue that has affected neural networks and prevented them from performing well in more realistic online environments is that of catastrophic forgetting. Catastrophic forgetting affects supervised learning systems when input samples are temporally correlated or are non-stationary. However, most real-world problems are non-stationary in nature, resulting in prolonged periods of time separating inputs drawn from different regions of the input space.

Reinforcement learning represents a worst-case scenario when it comes to precipitating catastrophic forgetting in neural networks. …

Design And Implementation Of An Economy Plane For The Internet, Xinming Chen

Doctoral Dissertations

The Internet has been very successful in supporting many network applications. As the diversity of uses for the Internet has increased, many protocols and services have been developed by the industry and the research community. However, many of them failed to get deployed in the Internet. One challenge of deploying these novel ideas in operational network is that the network providers need to be involved in the process. Many novel network protocols and services, like multicast and end-to-end QoS, need the support from network providers. However, since network providers are typically driven by business reasons, if they can not get …

Energy-Efficient Content Delivery Networks, Vimal Mathew

Doctoral Dissertations

Internet-scale distributed systems such as content delivery networks (CDNs) operate hundreds of thousands of servers deployed in thousands of data center locations around the globe. Since the energy costs of operating such a large IT infrastructure are a significant fraction of the total operating costs, we argue for redesigning them to incorporate energy optimization as a first-order principle. We focus on CDNs and demonstrate techniques to save energy while meeting client-perceived service level agreements (SLAs) and minimizing impact on hardware reliability. Servers deployed at individual data centers can be switched off at low load to save energy. We show that …

Threat Analysis, Countermeaures And Design Strategies For Secure Computation In Nanometer Cmos Regime, Raghavan Kumar

Doctoral Dissertations

Advancements in CMOS technologies have led to an era of Internet Of Things (IOT), where the devices have the ability to communicate with each other apart from their computational power. As more and more sensitive data is processed by embedded devices, the trend towards lightweight and efficient cryptographic primitives has gained significant momentum. Achieving a perfect security in silicon is extremely difficult, as the traditional cryptographic implementations are vulnerable to various active and passive attacks. There is also a threat in the form of "hardware Trojans" inserted into the supply chain by the untrusted third-party manufacturers for economic incentives. Apart …

Computational Framework For Small Animal Spect Imaging: Simulation And Reconstruction, Sang Hyeb Lee

Doctoral Dissertations

Small animal Single Photon Emission Computed Tomography (SPECT) has been an invaluable asset in biomedical science since this non-invasive imaging technique allows the longitudinal studies of animal models of human diseases. However, the image degradation caused by non-stationary collimator-detector response and single photon emitting nature of SPECT makes it difficult to provide a quantitative measure of 3D radio-pharmaceutical distribution inside the patient. Moreover, this problem exacerbates when an intra-peritoneal X-ray contrast agent is injected into a mouse for low-energy radiotracers.

In this dissertation, we design and develop a complete computational framework for the entire SPECT scan procedure from the radio-pharmaceutical …

Learning Parameterized Skills, Bruno Castro Da Silva

Doctoral Dissertations

One of the defining characteristics of human intelligence is the ability to acquire and refine skills. Skills are behaviors for solving problems that an agent encounters often—sometimes in different contexts and situations—throughout its lifetime. Identifying important problems that recur and retaining their solutions as skills allows agents to more rapidly solve novel problems by adjusting and combining their existing skills. In this thesis we introduce a general framework for learning reusable parameterized skills. Reusable skills are parameterized procedures that—given a description of a problem to be solved—produce appropriate behaviors or policies. They can be sequentially and hierarchically combined with other …