Date(s) - 13 Oct 2017
1:10 PM - 2:00 PM
3043 ECpE Building Addition
Speaker: Robi Polikar, Professor and Department Head of Electrical and Computer Engineering at Rowan University
Abstract: Part one of seminar: An increasing number of real-world applications are associated with streaming data drawn from drifting and nonstationary distributions that change over time. These applications demand new algorithms that can learn and adapt to such changes, also known as concept drift. Proper characterization of such data with existing approaches typically requires substantial amount of labeled instances, which may be difficult, expensive or even impractical to obtain. Such a scenario is also related to the problem known as verification latency, where the labels of the training data are not available until much later than the data itself – or in extreme verification latency that we discuss in this talk – they may never be available. In the first half of this lecture, we will introduce COMPOSE, a density tracking framework for learning from nonstationary streaming data, where labels are never available for the streaming data after initialization. We will discuss the algorithm in detail, as well as its results and performances on real-world datasets as well as several carefully constructed synthetic datasets, which demonstrate the ability of the algorithm to learn under this setting. Furthermore, we also demonstrate that COMPOSE is competitive even with a well-established, fully supervised, nonstationary learning algorithms that receive labeled data in every batch. Like all algorithms, COMPOSE makes certain assumptions such as the “limited-drift” assumption, where it assumes that any class distribution at two consecutive time-steps changes very little, i.e., the drift is gradual. In addition to such cases as abrupt drift, COMPOSE also cannot address special cases such as introduction of a new class or significant overlap among existing classes, as such scenarios cannot be learned without additional labeled data. Scenarios that provide occasional or periodic limited labeled data are not uncommon, however, for which many of COMPOSE’s restrictions can be lifted. In the second part of this talk, I will briefly introduce a couple alternate versions of COMPOSE as proof-of-concept algorithms that can identify the instances whose labels – if available – would be most beneficial, and then combine those instances with unlabeled data to actively learn from streaming nonstationary data, even when the distribution of the data experiences abrupt changes.
Part two of seminar: If there is one class common to all engineering programs, it is perhaps the Senior / Capstone Design – and for good reason: to provide students with an opportunity to put their cumulative knowledge and skills to actual good use, and demonstrate that they can indeed solve a real-world problems, often in a team setting. Not surprisingly, such a class in indeed required by ABET for all engineering programs seeking accreditation. The format and content of such classes are – by and large – quite similar across engineering programs: students propose a project or choose from a list of pre-approved projects; work on their design to solve the underlying problem; and present their findings through oral or written reports. The actual implementations of this time-tested and honored practice, however, are often riddled with a variety of challenges and short comings. If the projects are pre-selected by faculty, they are often limited in number and scope to be within the expertise of one (or few) instructor(s) responsible for the class. If the students pick the projects, there is a wide spread in the technical rigor of the projects. While all senior design courses claim that the projects involve “real-world” problems, those problems are often simple applications of known, well-established, or currently popular approaches, rather than being truly unsolved real challenges. Furthermore, since most engineering programs within a college run their senior design courses within their departments, all students within a project are all from the same department, and hence there is little or no opportunity for students to form and work in truly multidisciplinary teams. Finally, as the name implies, senior design is – by its very nature – limited to seniors in the program, and preclude other students from participating in solving real-world problems. At Rowan University’s College of Engineering, we take a different approach in providing our students with real-world design experience, while addressing all of the aforementioned issues associated with senior design. We challenge the notion that capstone design can only be during the final year of the program, or can only run within a department. In fact, our program does not even include a class called senior or capstone design. Instead, we have an eight-semester sequence, called Engineering Clinics, in which students work on increasingly sophisticated – and yes, real world – projects, while being intermingled with other students from across the College of Engineering’s six programs. During the four semesters of junior and senior years, students are required to work on not just real-world problems, but “unsolved” problems, which are often presented by industry and/or drawn from active research projects of the faculty. Not only any given project may have students from across the college – depending on the expertise needed by the project – juniors and seniors (and even graduate students) are often participate in any given project, providing both horizontal and vertical integration. Students may propose their own projects, but the same rigor of “unsolved real-world problem” threshold is applied to those projects as well. Students are required to participate in at least one “out-of-discipline” project experience, and perhaps more interestingly, all – not just a few – engineering faculty are involved in the clinics. In Fall 2017, ECE students had over 160 projects to choose from. In this talk, I will introduce our Engineering Clinics framework, describe in detail the specific implementation of this framework, and how it has contributed to the research outputs of the faculty, our job placement numbers, as well as the overall success of our program and the College.
Bio: Robi Polikar is a Professor and Department Head of Electrical and Computer Engineering at Rowan University, in Glassboro, NJ. He received his B.Sc. degree in electronics and communications engineering from Istanbul Technical University in 1993, and his M.Sc and Ph.D. degrees, both comajors in electrical engineering and biomedical engineering, from Iowa State in 1995 and 2000, respectively. His current research interests include ensemble systems, incremental and nonstationary learning, and various applications of machine learning in bioinformatics and biomedical engineering. He has served as an Associate Editor for IEEE Transactions on Neural Networks and Learning Systems and Springer’s Evolving Systems. He is also an ABET evaluator for Engineering Accreditation Commission.
Download event reminder: iCal