November 19, 2018, 12:00 PM
Professor Ran Canetti: From Soft Classifers to Hard Decisions: How Fair Can We Be?
[Monday, November 19, 2018 at 12:00 PM in Wegmans Hall 1400] As society is charging ahead with the prevalent use of (often opaque) algorithms in decision-making processes regarding people, it has become critical to study the societal implications of such algorithms. This study is fascinating, in that it forces us to mathematically formulate and quantify concepts such as fairness and social justice, which have traditionally been thought of in qualitative and subjective terms. In particular, this study is teasing out inherent inconsistencies in traditional thinking and helping us clarify our social goals. We consider one slice of this study: The need to balance the errors across societal groups in algorithmic decision-making based on imperfect information. Here some widely used statistical measures error-balance turn out to be contradictory (Kleinberg et al. '16, Chouldechova '17). We propose a way to - at least partially - evade this bound. Specifically, we observe that if the algorithm is allowed to `defer' on some decisions (that is, the algorithm can avoid making some decisions and hand them off to a separate process which is presumably more expensive and more precise), then the errors in the non-deferred decisions will be balanced across the considered societal groups by all statistical measures currently under consideration. We further demonstrate concrete ways for using deferrals in the natural context of post-processing ``soft classifiers that output a numerical score into a binary decision. We evaluate our algorithms using the COMPAS data set from 2016 and consider potential societal implications of deferrals.
November 26, 2018, 12:00 PM
Rob Shearer: The Promise of AR and New Challenges for Silicon
[Monday, November 26, 2018 at 12:00 PM in Wegmans Hall 1400] A look at how AR & VR are evolving and some possible applications and use cases. Then well review some of the key technology challenges that need to be overcome before widespread deployment of AR & VR will happen and the breakthroughs in silicon that are required to enable this future. We need substantial advancements beyond the state of the art to meet the silicon performance/power targets.
November 30, 2018, 01:00 PM
Guoyu Lu: TBD
[Friday, November 30, 2018 at 1:00 PM in Wegmans Hall 2506] TBD
December 3, 2018, 12:00 PM
Professor Dongyoon Lee: Fuzzing and Securing the Server-Side Event-Driven Architecture
[Monday, December 03, 2018 at 12:00 PM in Wegmans Hall 1400] The software development community is adopting the Event-Driven Architecture (EDA) to provide scalable web services, most prominently through Node.js. Though the EDA scales well, it comes with two inherent risks: concurrency errors and Event Handler Poisoning (EHP) Denial of Service attacks. Just as thread-based programs can have concurrency errors between unordered threads, event-driven programs may have them between unordered events. When an EDA-based server multiplexes many clients onto few threads, a blocked thread (EHP) renders the whole server unresponsive. In this talk, I present Node.fz and Node.cure to address these problems. First, Node.fz provides a schedule fuzzing test tool that randomly perturbs the execution of a Node.js program, allowing Node.js developers to explore a variety of possible schedules during testing. Second, Node.cure proposes First-Class Timeouts, which incorporates timeouts at the EDA framework level, defending Node.js applications against all known EHP attacks.
[Tuesday, December 04, 2018 at 10:00 AM in Wegmans Hall 2506] The rise of AI assistant systems, including Google Home, Apple Siri, and Amazon Echo, brings the urgent need for increased and deeper understanding of users. In this talk, I will present algorithmic and statistical methods for analyzing audio-visual human behavior, particularly focusing on emotional and social signals inferred from speech and facial expressions. These methods can provide emotional intelligence to AI systems. However, developing automatic emotion recognition systems is challenging since emotional expressions are complex, dynamic, inherently multimodal, and are entangled with other factors of modulation (e.g. speech generation and emphasis). I will present several algorithms to address these fundamental challenges in emotion recognition: (i) cross-modal modeling methods that capture and control for interactions between individual facial regions and speech using the Minimum Description Length (MDL) principle-based segmentation; (ii) localization and prediction of events with salient emotional behaviors using a max-margin optimization and dynamic programming; and (iii) temporal modeling methods to learn co-occurrence patterns between emotional behaviors and emotion label noise. These algorithms have enabled advancements in the modeling of audio-visual emotion recognition systems and increased the understanding of the underlying dynamic and multimodal structure of affective communication (e.g., cross-modal interaction, temporal structure, and inherent perceptual ambiguity).
December 10, 2018, 12:00 PM
Professor Zhong Shao: CertiKOS: From Hacker-Resistant OS to Certified Heterogeneous Systems
[Monday, December 10, 2018 at 12:00 PM in Wegmans Hall 1400] Computer science is being transformed by new exciting "vertical" research themes such as IoTs, self-driving cars, blockchains, and robotics. These new themes often require complex computing platforms involving not only heterogeneous hardware and software components but also physical, human, and social entities. The successful deployment of these technologies would critically depend on our ability to establish strong properties such as safety, security, resilience, resource efficiency, fairness, and privacy. In this talk, I present a novel compositional approach for building certified heterogeneous systems that aim to provide many such trustworthiness guarantees. We advocate abstraction over a particularly rich class of specification and present new methodologies and tools for formally specifying, programming, verifying, and composing abstraction layers. Using these new technologies, we have successfully developed the CertiKOS certified OS kernel and verified its contextual functional correctness in the Coq proof assistant. This is the world's first proof of functional correctness of a complete, general-purpose concurrent OS kernel with fine-grained locking. We show how to extend our base kernel with new features such as virtualization, interrupts and device drivers, and end-to-end information flow security, and how to quickly adapt existing verified layers to build new certified kernels for modern heterogeneous platforms.
January 18, 2019, 01:00 PM
Yu Kong: TBD
[Friday, January 18, 2019 at 1:00 PM in Wegmans Hall 2506] TBD