CMMRS will include lectures from 8 professors from the University of Maryland, Cornell University, and the Max Planck Institutes in Computer Science. The full list of lectures and topics is yet to be finalized. Lectures will cover a diverse range of topics.
Tapomayukh Bhattacharjee, Cornell University
Towards Robotic Caregiving: Building robots that work alongside human stakeholders
How do we build robots that can assist people with mobility limitations with activities of daily living? To successfully perform these activities, a robot needs to be able to physically interact with humans and objects in unstructured human environments. Through these two talks, I will first give you a brief high-level introduction to the field of robotics and then dive deeper into the sub-field of robotic caregiving that my lab works on. I will cover various projects in my lab that showcase fundamental advances in the field of robotic caregiving. Specifically, I will show you how we can build caregiving robots to perform activities of daily living such as feeding and bed-bathing. Both tasks require a robot to reason about the safety of complex physical interactions with humans in the presence of uncertainties due to perception and planning in cluttered assistive environments as well as complexities due to disability conditions (spasms and involuntary movements). For both activities, the state of the world may not always be fully observable due to occlusions (e.g. feeding inside the mouth) or uncertainty in perception (e.g. wiping transparent water/gel on a limb). Using insights from human studies, I will showcase algorithms and technologies that leverage multiple sensing modalities to perceive varied object properties and determine successful control policies for feeding and bed-bathing. Using feedback from all the stakeholders, I will show how we built autonomous robot-assisted feeding and bed-bathing systems that use these algorithms and technologies, and how we deployed them to work in the real world with real users. I will end the talks with some open problems in the field of robotic caregiving.
Jordan Boyd-Graber, University of Maryland
The questions that computers still cannot answer: how to find them, why they’re fun, and what they mean for AI
You’ve probably heard of ChatGPT and other large language models. They can do some neat stuff! But they still have some weaknesses. After briefly reviewing how transformer language models work, we will go into the weaknesses of models in 2023: recognizing unconventional language, numerical and logical reasoning, multihop questions, cross-cultural understanding, and incorporating new information. Next, we’ll describe how to elicit questions that challenge today’s computer models using adversarial human-in-the-loop authoring and how computers and humans compare at their ability to answer those questions.
Laxman Dhulipala, University of Maryland
Building Scalable and Practical Batch-Dynamic Graph Algorithms
In this talk, I will give an overview of recent work on parallel dynamic graph algorithms and the closely related graph-streaming setting where a continuous stream of graph updates is mixed with graph queries. The study of these settings are motivated by the need to answer time-sensitive queries over rapidly changing data, e.g., detecting fraud in financial networks, identifying emerging trends, or providing real-time recommendations. To cope with the rapid rate of change in these settings, parallel algorithms that can process batches of updates in parallel are necessary. I will present our work on (1) parallel batch-dynamic data structures for representing dynamic graphs, (2) ongoing work on parallelizing several important dynamic graph problems, and (3) end by describing open questions and directions for future work in this area.
Moritz Hardt, Max Planck Institute for Intelligent Systems (MPI-IS)
The power of predictions
In this tutorial, I will introduce the calculus of performative prediction for its use in reasoning about the effects of algorithmic predictions on human populations. Performative prediction reveals a distinction between two mechanisms fundamental to social prediction. One is to discover patterns in a population. The other, less recognized, is to steer the population through predictions. Building on performative prediction, we develop a notion of power tailored to digital platforms operating predictive systems. From here we examine the power of platforms in digital markets through theory, observational causal inference, and randomized experiments. We end on a discussion of collective algorithmic strategies to effectively resist the power of platforms.
Owolabi Legunsen, Cornell University
Specializing Runtime Verification for Software Testing
The spate of costly and harmful bugs in deployed software underscores the need for techniques that can help find more bugs during software development. Runtime verification (RV) is such a technique. RV checks program executions against formally specified properties and produces violations if execution traces — sequences of events — do not satisfy the properties. In these talks, I will first introduce RV and discuss my prior work on using RV to amplify the bug-detection capability of existing test suites and on leveraging software evolution to scale RV. Then, I will highlight several challenges that still need to be addressed if RV is to become widely used among developers. Finally, I will discuss work in progress that my research group is doing to specialize RV for software testing, towards addressing these challenges.
Murphy Yuezhen Niu, University of Maryland
Introduction to quantum control for superconducting qubits and beyond
After over 30 years of research and engineering in controllable quantum devices, we are at the cusp of discovering the first real-world applications for quantum computing systems. A critical element that connects high-level applications to low-level quantum system designs is quantum control. In my talk, I will review on a high level the design and calibration of quantum gates, analog quantum operations, quantum state preparation, and measurement for superconducting qubits.
Yiting Xia, Max Planck Institute for Informatics (MPI-INF)
Towards High-Performance and Power-Efficient Network Infrastructure for Cloud Computing
The data center network is a key infrastructure for cloud computing. Emerging cloud applications such as machine learning, big data, and Internet of Things are driving the demand for highly available and highly performing data center networks, which brings cost and carbon footprint into play. Optical data center networks show promise to serve as the next-generation cloud infrastructure with their performance and power benefits. In this talk, the speaker will introduce optical data center networks, summarize the challenges for practical deployment, and propose a solution towards full adoption of the technology.
Yixin Zou, Max Planck Institute for Security and Privacy (MPI-SP)
Human-Centered Security and Privacy
There is an increasing appreciation for human factors in security and privacy research. The knowledge of people’s concerns, needs, and expectations provide valuable insights for improving security and privacy systems. Meanwhile, people often do not use existing tools and strategies to the full extent – and it is not their fault. In this lecture series, I will draw from my research to demonstrate the value of incorporating human factors in designing security and privacy mechanisms, and the need of considering digital equity in people’s ability to protect themselves. In the first part, I will feature my line of work on data breaches as a case study, showing how examining consumer reactions could inform the design of more effective breach notifications. In the second part, I will feature my work with various marginalized populations–such as survivors of intimate partner violence, older adults, and Muslim-American women–and trauma-informed computing as a unifying framework for creating safer technology experiences for all. Throughout the lectures, I will highlight how this human-centered approach can lead to positive impacts on industry practices, public policy, and educational efforts around security and privacy. Content warning: some parts of the lectures will include descriptions of physical/emotional violence, harassment, and trauma.