Researchers have made substantial progress towards developing Artificial Intelligence (AI) systems capable of a human-level performance at a narrow range of tasks. However, there do not yet exist machines capable of the kinds of flexible and efficient learning that humans exhibit. To overcome this gap, I present research exploring how data from intelligent tutoring systems can be leveraged to reverse engineer human capabilities and build more human-like models of learning.
The research of computer vision was motivated by a dream of making an intelligent machine that is able to see like our human beings: to automatically analyze and understand massive visual inputs. With the explosive growing of computing power, this dream evolves to many exciting emerging applications, such as intelligent robots, autonomous vehicles, intelligent video surveillances, computer-aided doctors, etc. A core component in these applications is visual recognition (including object classification, detection and localization).
Through deep learning, deep neural networks have produced state-of-the-art results in a number of different areas of machine learning, including computer vision, natural language processing, robotics and reinforcement learning. I will summarize three projects on better understanding deep neural networks and improving their performance. First I will describe our sustained effort to study how much deep neural networks know about the images they classify.
Interactive Learning (IL) is a prominent machine learning paradigm that explores how intelligent autonomous systems learn to make improved sequential decisions through multiple rounds of interactions with the real world. IL-based systems have a wide range of applications including robotics, health-care, and marketing. However, for these systems to meet the practical requirements of the related application domains, there is a perennial need for developing efficient algorithms.
The state of network security today is quite abysmal. Security breaches and downtime of critical infrastructures continue to be the norm rather than the exception, despite the dramatic rise in spending on network security. Attackers today can easily leverage a distributed and programmable infrastructure of compromised machines (or botnets) to launch large-scale and sophisticated attacks. In contrast, the defenders of our critical infrastructures are crippled as they rely on fixed capacity, inflexible, and expensive hardware appliances.
Double spending is a fundamental problem which should be addressed in all cryptocurrency systems, including Bitcoin. While Bitcoin is equipped with inherent security mechanisms to thwart double spending, it requires transactions to be confirmed in a very slow speed, which is impractical in many applications. In this talk, I will discuss the existing solutions and propose new solutions to address the double spending problem for fast transactions in Bitcoin. In particular, I will focus on potential attacks, pitfalls, and defending mechanisms in addressing the double spending problem.
Emerging Internet of Thing (IoT) platforms provide a centralized solution to integrate heterogeneous IoT devices and deploy applications for home automation. However, new privacy threats are also introduced since platforms may fail to protect the collected data due to a number of general or domain-specific reasons, e.g., remote attacks, insider attacks, improper data release, flawed access control, malware, etc.
The last half-decade ushered in a new era of vision research. Computer vision now works on real images, in natural environments, solving hard problems. But the technology is far from ubiquitous and many researchers are most concerned with getting the best performance on a handful of datasets. This hyper-focus on accuracy has largely turned vision into a numbers game and research tends toward complex, finely-tuned systems that are brittle and impractical in the real world.