SystemX

Ethics is often messy and chaotic—it's hard to see the method in the madness. In contrast to the way ethics is usually taught, this talk will offer a new framework that lays out the process or ethics workflow, in order to systematically think through the issues in technology ethics, especially AI ethics. We will also look at a few AI case studies in applying this framework.

The success of the abstract model of computation, in terms of bits, logical operations, algorithms, and programming language constructs makes it easy to forget that computation is a physical process. Our cherished notions of computation and information are grounded in classical mechanics, but the physics of our universe is quantum. A natural question to ask is how computation would change if we adopted a quantum mechanical, instead of a classical mechanical, model of computation. In the early 80s, Richard Feynman, Yuri Manin, and others recognized that certain quantum effect could not be simulated efficiently on conventional computers. This observation led researchers to speculate that perhaps such quantum effect could be used to speed up computation more generally. Slowly, a new picture of computation arose, one that gave rise to a variety of faster algorithms, novel cryptographic mechanisms, and alternative methods of communication. In the first part of the talk, I will introduce key concepts underlying quantum computing and correct misconceptions. In the second part of the talk, I will discuss research being done by NASA's QuAIL group on quantum algorithms, quantum supremacy, elucidating quantum resources for computation, quantum programming and compilation, quantum-inspired classical algorithms, and assessing future applications of quantum computing, all within the broader context of the rapidly evolving field.

Ion trap quantum computers at first glance have no materials challenges. Ions are all the same by nature and they are trapped and manipulated by electromagnetic fields. The materials challenges arise in how these fields are delivered to the ions. In this talk, I will describe how ion trap quantum computers work and then review the materials challenges that were recently described in Nature Reviews Materials (https://doi.org/10.1038/s41578-021-00292-1).

Autonomous and anthropomorphic robots are poised to play a critical role in manufacturing, healthcare and even our homes in the near future. However, for this vision to become a reality, robots need to efficiently collaborate and physically interact with their human partners. Rather than traditional remote controls and programming languages, adaptive and transparent techniques and interfaces for human-robot collaboration are needed. In particular, robots may need to interpret implicit behavioral cues or explicit instructions and, in turn, generate appropriate responses. In this talk, I will present ongoing work which leverages machine learning (ML), natural language processing and virtual reality to create different modalities for humans and machines to engage in effortless and natural interactions. To this end, I will describe Bayesian Interaction Primitives - an approach for motor skill learning and spatio-temporal modelling in physical human-robot collaboration tasks. Further, I will discuss our recent work on language-conditioned imitation learning and self-supervised learning in interactive tasks. The talk will also cover techniques that enable robots to communicate information back to the human partner via mixed reality projections. To demonstrate these techniques, I will present applications in social robotics and collaborative assembly.

Many descriptions of trench warfare in World War I describe it as "months of boredom punctuated by moments of extreme terror" (Guy's Hospital Gazette (1914), The New York Times Current History of the European War (1915), among others). Those responsible for public safety today can relate to this description. Critical decisions in public safety often have to be made with incomplete and uncertain information under varying degrees of stress. Life and death hang in the balance as a consequence of many of these decisions. The gravity of the task attracts a considerable set of requirements for documentation. Approximately 30% of a first-responder's time is spent on documentation. A 911 call-taker works on multiple screens running different applications simultaneously entering the same data many times. Post-incident investigations require multiple sources of data to be integrated manually, and roughly 80% of the aggregated case files likely have errors. Assisting in the monotonous, time-consuming, and error-prone task of filling out forms and documenting events can save lives. I'll talk about how automatic speech recognition, language understanding, object and activity recognition in video, and effective UX design can help with this problem.

SLAC is a multi-faceted, R&D institution with one-of-a-kind onsite tools that lead the world in probing the fundamentals of nature. In this talk, we'll discuss the technologies that have enabled these tools and how SLAC has applied these same technologies to multiple uses. Examples of how the accelerator technologies that enabled these capabilities have been applied to medicine, national security and industry will be presented. In the area of advanced instrumentation for research, examples of detector systems for South-Pol, mountain peak in Chile, to Space-based experiments are presented.

The flow of professional talent, both permanent and temporary, is a prevalent aspect of globalization. High-skilled talent moves across national borders in search of better academic, professional, and social opportunities. But the concept of talent flow is not narrowly confined to the physical relocation of talented individuals. In the era of the knowledge economy, mobile talent contributes to the creation and diffusion of knowledge, and one cannot disregard the social capital value of talent that incorporates the connections between cultures and the potential for transnational collaboration.

The migration of high-skilled professionals is not a zero-sum game in which the host country receives a net inflow of human capital from the home country. A phenomenon commonly referred to as "brain drain" for the home country and "brain gain" for the host country can in fact offer opportunities for brain circulation or brain linkage—that is, home-host interactions that create a win-win, positive-sum situation for both sides. When high-skilled migrant talent stays engaged with the home country, both home and host countries gain from the productive capacity embodied in the ties and networks linking many individuals and organization.

Transnational social capital and ties spanning geographic and cultural distance remain vital to today's global market economy, even more so in a time of political tensions at home and abroad. Speakers Gi-Wook Shin and Dexter Simpson will discuss the positive gains of global talent flows and how migrant talent can create mutually beneficial ties—or "brain linkages"—between the United States and their home countries, even during the times of heightened political tension and rising anti-immigrant sentiment in the United States.

With slowing technology scaling, specialized accelerators are increasingly attractive, but naive specialization limits accelerators to narrow domains. This is problematic practically because algorithms are constantly evolving, and intellectually as innovations are often siloed into their respective domains.

We believe this problem can be solved jointly by, 1. making accelerators more general, and 2. automating their design. I'll first overview a "general purpose accelerator ISA" that can abstract the typical behaviors of domain-specific accelerators. Our evaluation shows accelerators with these can achieve order-of-magnitude improvements over GPUs, without sacrificing programmability. However, many of their features are expensive and not useful for every workload. Therefore, our second direction is automated codesign: We developed a framework, DSAGEN, that enables users to search for the best programmable architecture given a set of input C/C++ kernels, using principles of modular hardware and compilation. Our overall vision is that hardware and ISA design can be nearly completely automated.

Materials engineered with atomic precision promise unprecedented control over their structure and properties, with profound implications for future device technologies. Atomically thin two-dimensional (2D) materials provide a versatile platform for atom-scale engineering: they exhibit a variety of superlative electronic characteristics, and their discrete layered structures and van der Waals (vdW) interlayer bonding enable them to be grown, patterned, and stacked to generate heterostructured solids with atomically-precise vertical composition and band structure tailored by moiré superlattices. However, current approaches to stacking 2D layers into high-quality vdW heterostructures are slow, stochastic, and artisanal. In this talk, I will discuss the development of automated manufacturing of vdW heterostructures with unprecedented speed, patternability, and angle control. Additionally, I will outline ongoing efforts in my lab to manufacture van der Waals heterostructures with high-quality 2D materials grown at the wafer scale to enable quantum device applications.

After decades of intensive theoretical and experimental efforts, the field of quantum information processing is at a critical moment: special-purpose quantum information processors are cutting into a regime of quantum complexity where classical computers can no longer predict their outputs: we can "program complexity", unable to predict the outcome. Meanwhile, new technologies to connect quantum processors by photons give rise to quantum networks with functions impossible on today's "classical-physics" internet.

However, to harness the power of quantum complexity in "noisy intermediate-scale" quantum computers and networks requires advanced methods in quantum control and noise mitigation -- perhaps to the ultimate goal of fault tolerant computing. This talk discusses one approach in that direction: large-scale programmable photonic integrated circuits (PICs) designed to control photons and atomic or atom-like quantum memories.

The second part of the talk considers another "complexity frontier": that encountered in machine learning and signal processing when trying to process exponentially growing quantities of data. These problems present new opportunities at the intersection with quantum information technologies -- specifically, we will consider new directions for processing classical and quantum information in deep learning neural networks architectures. [References for abstract linked here]