Upcoming Events

Towards Interpretable and Controllable Representation Learning

Duke Computer Science Colloquium
Speaker Name
Xuezhe (Max) Ma
Location
LSRC D106
Date and Time
-
In this talk, I will first present our recent work on using probing methods to investigate and interpret two learning properties in deep neural networks: (i) laziness and (ii) targetedness. Second, I will present how to utilize these two properties to control the learned representations. I will conclude by laying out future research directions towards interpretable and controllable representation learning by establishing theoretical framework to formally link various neural architectures with the learned representations.

Fast and Flexible Data-Center Networks

Duke Electrical Computer Engineering Colloquium
Speaker Name
Muhammad Shahbaz
Location
Fitzpatrick Center, Schiciano Auditorium side B, room 1466
Date and Time
-
Exposing the right domain-specific abstractions and hardware primitives for networks can enable more flexibility with negligible (or zero) loss in performance. In this talk, I will present systems and approaches I developed, like Pisces and Elmo, that let network programmers modify switch behavior more flexibly using a high-level P4 language (e.g., to encode state inside packets) and new hardware primitives (i.e., to read, infer, and act on the encoded state), without compromising speed.

Building Data Science Platforms for Better Lives: A Pattern Mining Approach

Miscellaneous Talk
Speaker Name
Jian Pei
Location
Hock, 2nd floor CRTP Classroom
Date and Time
-
In this talk, I will demonstrate some of the challenges and opportunities of mining various patterns for data science platforms, starting from mining patterns itself to producing early prediction and providing better interpretation using patterns, as well as some other great potentials. Pattern-based data science platforms put medical and health science researchers, practitioners, service providers and users at the center of the loop, and provide novel tools to extend their perception and cognition capability for better lives.

Policy Compliance in Online Services

Duke Computer Science Colloquium
Speaker Name
Aastha Mehta
Location
LSRC D106
Date and Time
-
In this talk, I will describe Pacer, a compliance system designed to prevent indirect inference of sensitive data via side channels in shared network links in the Cloud. Pacer shapes the outbound traffic of a Cloud tenant to make it independent of the tenant's secrets. At the same time, Pacer does allow variations in the traffic shape based only on public (non-secret) aspects of the tenants' workloads, thus enabling efficient sharing of network resources and incurring moderate overhead.

Better Foundations for Secure Software: Minimize Trust and Verify It

Duke Computer Science/Electrical Computer Engineering Colloquium
Speaker Name
Shweta Shinde
Location
LSRC D106
Date and Time
-
In this talk, I present two key results from my work that shows a foundation approach to safeguard applications against large and potentially buggy software. First, I present a principled way of using hardware isolation to securely execute Linux applications while only trusting a few thousand lines of code. Second, I show the feasibility of full formal verification of the trusted code by proving guarantees over a large-subset, such as the file system interface. Finally, I will summarize my long-term vision for building the next generation of better, trusted, and verified secure hardware and software designs.

Auditing Outsourced Services

Duke Computer Science Colloquium
Speaker Name
Cheng Tan
Location
LSRC D106
Date and Time
-
This talk will present verifiable infrastructure, a framework that lets users audit outsourced applications and services. I will introduce two systems: Orochi and Cobra, which verify the execution of, respectively, untrusted servers and black-box databases. Orochi and Cobra introduce various techniques, including deduplicated re-execution, consistent ordering verification, GPU accelerated pruning, and others. Beyond these two systems, I will also discuss verifiable infrastructure more generally.

Securing the Web through Dynamic Program Monitoring

Duke Electrical Computer Engineering Colloquium
Speaker Name
Wei Meng
Location
Gross Hall 330
Date and Time
-
In this talk, I will share my experience on detecting and mitigating web attacks and abuses through dynamic program monitoring, which allows us to better analyze the runtime behavior of both client-side and server-side application code. Specifically, I will focus on two works that address two emerging threats against the end users and the service providers, respectively.

Cross-Layering in Future Wireless Networks: From Compact Full-duplex Radios to City-Scale Experimentation

Duke Electrical Computer Engineering Colloquium
Speaker Name
Tingjun Chen
Location
Hudson Hall 208
Date and Time
-
In this talk, I will first describe our design and optimization of compact single-antenna full-duplex radios which are suitable for hand-held devices. I will then present a novel approach to enable full-duplex operation in multi-antenna systems by repurposing beamforming degrees of freedom to achieve both wideband self-interference cancellation and improved data rate gain.

Gromov-Wasserstein Learning: A New Machine Learning Framework for Structured Data Analysis

Duke Computer Science/Biostatistics & Bioinformatics Colloquium
Speaker Name
Hongteng Xu
Location
LSRC D106
Date and Time
-
In this talk, I will introduce a novel machine learning framework called Gromov-Wasserstein Learning (GWL) — a new systematic solution I proposed for structured data analysis. First, I will introduce the theoretical fundamentals of GWL and link it to learning tasks from structured data. Next, I will describe the optimization algorithms in the GWL, analyzing their convergence, computational complexity, and scalability in detail. Finally, I will show that the GWL unifies graph matching, partitioning, and representation into the same algorithmic framework, which outperforms existing methods on PPI network analysis and molecule clustering and classification.

Towards Language Technology for Everyone

Duke Computer Science Colloquium
Speaker Name
Antonios Anastasopoulos
Location
LSRC D106
Date and Time
-
In this talk, I will focus on a sample of research threads towards the goal of building NLP that serves everyone. First, I'll describe work that is tailored to creating NLP tools for endangered language documentation, with neural models that take advantage of additional signals (in this case translations) in multi-source and multi-task settings. Second, I'll show how similarities across languages can be leveraged for building more accurate morphological inflection systems in numerous under-resourced languages, along with data augmentation through hallucination. Last, I will suggest several directions for future research.

Enabling Future-Proof Telemetry for Networked Systems

Duke Electrical Computer Engineering Colloquium
Speaker Name
Alan (Zaoxing) Liu
Location
Gross Hall 318
Date and Time
-
In this talk, I will present my research that focuses on building telemetry systems that are future-proof for current and new telemetry tasks, diverse workloads, and heterogeneous platforms. I will discuss the efficient algorithms and implementations that realize this future-proof vision in network monitoring for hardware and software platforms.

Causal Inference & Graphical Models: From Missing Data to the Future of Artificial Intelligence

Duke Computer Science Colloquium
Speaker Name
Karthika Mohan
Location
LSRC D106
Date and Time
-
The remarkable progress in AI and machine learning owe much to the availability of massive amounts of data, and where there is data, there is missingness. I address these deficiencies by using a graphical representation called "Missingness Graph" which portrays the causal mechanisms responsible for missingness. Viewing the missing data problem from a causal perspective has ushered in several notable surprises.

PhD Defense

Ph. D. Defense
Speaker Name
Yuan Deng
Location
LSRC D344
Date and Time
-

Reuse-Centric Programming System Support of Machine Learning

Duke Computer Science Colloquium
Speaker Name
Hui Guan
Location
LSRC D106
Date and Time
-

Modern machine learning, especially deep learning, faces a fundamental question: how to create models that efficiently deliver reliable predictions to meet the requirements of diverse applications running on various systems. This talk will introduce reuse-centric optimization, a novel direction for addressing the fundamental question. Reuse-centric optimization centers around harnessing reuse opportunities for enhancing computing efficiency. It generalizes the principle to a higher level and a larger scope through a synergy between programming systems and machine learning algorithms.

Distributive Justice for Machine Learning: An Interdisciplinary Perspective on Defining, Measuring, and Mitigating Algorithmic Unfairness

Duke Computer Science Colloquium
Speaker Name
Hoda Heidari
Location
LSRC D106
Date and Time
-
In this talk, I will illustrate how we can bring together tools and methods from computer science, economics, and political philosophy to define, measure, and mitigate algorithmic unfairness. In particular, I will address two key questions: Given the decision-making context, how should we define fairness as the equality of some notion of benefit or harm across socially salient groups? How can we measure unfairness (both at the individual and group level) and bound it in a computationally efficient manner?

Cold-Start Universal Information Extraction

Duke Computer Science Colloquium
Speaker Name
Lifu Huang
Location
LSRC D106
Date and Time
-
In this talk, I will introduce a new information extraction paradigm - Cold-Start Universal Information Extraction, which aims to create the next generation of information access where machines can automatically discover accurate, concise, and trustworthy information embedded in data of any form without requiring any human effort.

Evaluating Robustness of Neural Networks

Duke Computer Science/Electrical Computer Engineering Colloquium
Speaker Name
Lily Weng
Location
LSRC D106
Date and Time
-

The robustness of neural networks to adversarial examples has received great attention due to security implications. Despite various attack approaches to crafting visually imperceptible adversarial examples, little has been developed towards a comprehensive measure of robustness. In this talk, I'll present a series of our work on robustness evaluation and certification, including the first robustness score CLEVER, efficient certification algorithms Fast-Lin, CROWN, CNN-Cert, and probabilistic robustness verification algorithm PROVEN.

Putting Ethical AI to the Vote

Triangle Computer Science Distinguished Lecturer Series
Speaker Name
Ariel Procaccia
Location
LSRC D106
Date and Time
-

I will present the 'virtual democracy' framework for the design of ethical AI. In a nutshell, the framework consists of three steps: first, collect preferences from voters on example dilemmas; second, learn models of their preferences, which generalize to any (previously unseen) dilemma; and third, at runtime, predict the voters' preferences on the current dilemma, and aggregate these virtual 'votes' using a voting rule to reach a decision.