Graduate

For robots to effectively operate in our world, they must master the skills of dynamic interaction. Autonomous cars must safely negotiate their trajectories with other vehicles and pedestrians as they drive to their destinations. UAVs must avoid collisions with other aircraft, as well as dynamic obstacles on the ground. Disaster response robots must coordinate to explore and map new disaster sites. In this talk I will describe recent work in my lab using distributed optimization to obtain algorithms for robots to cooperate, and game theoretic methods to obtain algorithms for robots to compete. I will present an algorithm for fleets of autonomous cars to cooperatively track a large number of vehicles and pedestrians in a city, an algorithm for multiple robots to manipulate an object to a goal while avoiding collisions, and a distributed multi-robot SLAM algorithm, all derived using the same underlying distributed optimization framework. I will also discuss algorithms based on the theory of dynamic games, in which each actor has its own objective and constraints. I will describe examples in autonomous drone racing, car racing, and autonomous driving that use game theoretic principles to solve for Nash equilibrium trajectories in real-time, in a receding horizon fashion. Throughout the talk, I will show results from hardware experiments with ground robots, autonomous cars, and quadrotor UAVs collaborating and competing in the scenarios above.

For the past decade, display and sensor hardware developments for mixed reality and smart glasses were merely a shot in the dark, providing enough display immersion and visual comfort for developers to build up apps, especially for the enterprise field. On the sensor side, emphasis was put on 6DOF head tracking and spatial mapping, gesture sensing and later eye tracking. Today, as universal use cases for consumer emerge such as co-presence, digital twin and remote conferencing, new requirements are expressed in the product requirement documents (PRD) to enable such experiences, both on the display and sensing side. It is not only a race to smaller form factor and light weight devices for large field of view (FOV) and lower power, but the requirements are also on additional display and sensing features specifically tuned to implement such new universal use cases. Broad acceptance of wearable displays especially in the consumer field is contingent on enabling these new display and sensing requirements in small form factors and low power.

As of today, more than 20 billion devices, almost three times the number of people on earth, are connected to the internet. More than half of that are Internet-of-Things (IoT) devices and it is expected that flexible electronics will play a big role in developing new types of IoT sensor systems. Applications include environmental monitoring, food packaging, and biomedical applications like vital sign and disease detection on skin or inside of the human body. However, there are several material, device and integration challenges to solve before flexible IoT systems can become a reality. Most flexible substrates require low process temperature (typically <250 °C), which hinders the direct growth of high-quality semiconductors making the choice of materials limited. At the same time, typical device dimensions are on the micron-scale which leads to low performance and high power consumption. Here, I will show how layered and two-dimensional (2D) chalcogenide compounds can offer attractive solutions for the components needed in flexible electronics overcoming the previously mentioned limitations. I will present our recent work on flexible devices including transistors, sensors, solar cells and memory. In future, combining these devices, we can imagine self-powered flexible IoT systems enabled by low energy consumption and integrated energy harvesters.

Join leaders from around the globe for a field-defining conversation about the challenges and opportunities of connecting the next billion. Your ticket includes access to all virtual sessions during the October 22 symposium in addition to all pre- and post-symposium virtual sessions.

If you'd like to join us for our virtual evening gala to celebrate the groundbreaking work of Andrea Goldsmith, 2020 Marconi Fellow, tickets are available separately.

OVERVIEW OF THE MARCONI SOCIETY EVENT

Defining the next decade of equitable connectivity

We believe Internet access is a basic human right. Join leaders in policy, technology, and digital inclusion advocacy to learn about the critical challenges of connecting the next billion and to develop innovative, practical solutions.

The Decade of Digital Inclusion is your opportunity to join a community of innovators working toward a digitally inclusive future. This timely conversation can only be hosted by the Marconi Society, a nonprofit that bridges the communities of advanced technology and digital inclusion.

In addition to the symposium series, we will make history by recognizing and honoring one of the industry's leading stars, Andrea Goldsmith, as our 2020 Marconi Fellow.

Student organizer contact: Kirstin Schauble (kschaub@stanford.edu)

I will present a novel concept that blends spintronics and quantum thermodynamics to generate electricity. This concept is invoked to explain experimental observations of electrical generation across oxide1 and molecular spintronic devices that comprise paramagnetic centers sandwiched between electrodes with full transport spin polarization. The presence of so-called quantum resources2,3, leading to a source of work of quantum origin called ergotropy, appears to be manifest in sub-kBT spectral features, as well in an apparent signature of a phase transition of the spin fluctuations on the paramagnetic centers. I will discuss our present research tracks to better understand this spintronic quantum engine. General info may also be found at www.spinengine.tech.

References:

1. Katcko, K. et al. Spin-driven electrical power generation at room temperature. Communications Physics 2, 116 (2019).

2. Bresque, L. et al. Two-Qubit Engine Fueled by Entanglement and Local Measurements. Phys. Rev. Lett. 126, 120605 (2021).

3. Klatzow, J. et al. Experimental Demonstration of Quantum Effects in the Operation of Microscopic Heat Engines. Phys. Rev. Lett. 122, 110601 (2019).

The classical Erdos–Littlewood–Offord theorem says that for any n nonzero vectors in R^d, a random signed sum concentrates on any point with probability at most O(n^{1/2}). Combining tools from probability theory, additive combinatorics, and model theory, we obtain an anti-concentration probability of n^{-1/2+o(1)} for any o-minimal set S in R^d — such as a hypersurface defined by a polynomial in x1,...,xn,exp(x1),...,exp(xn), or a restricted analytic function — not containing a line segment. We do this by showing such o-minimal sets have no higher-order additive structure, complementing work by Pila on lower-order additive structures developed to count rational and algebraic points of bounded height.

This is joint work with Jacob Fox and Matthew Kwan.

In the Gilbert–Shannon–Reeds shuffle, a deck of N cards is cut into two approximately equal parts which are riffled together uniformly at random. This Markov chain famously undergoes total variation cutoff after (3/2)*log_2(N) shuffles. We prove cutoff for asymmetric riffle shuffles in which the deck is cut into differently sized parts before riffling, confirming a conjecture of Lalley from 2000.

Note: this talk will be held in person in Packard 101, and will start at 4pm. The talk will be streamed on Zoom for those who cannot attend (registration link below).  Please join us for coffee starting at 3:30pm at the Grove outside Packard.

Talk 1: Marketplace platforms use experiments (also known as "A/B tests") as a method for making data-driven decisions. When platforms consider introducing a new feature, they often first run an experiment to test the feature on a subset of users and then use this data to decide whether to launch the feature platform-wide. However, it is well documented that estimates of the treatment effect arising from these experiments may be biased, due to the presence of interference. In this talk, we survey a collection of recent results and insights we have developed on experimentation and decision-making in two-sided marketplaces. In particular, we study the bias that interference creates in both the treatment effect estimates as well as standard error estimates, and show how both types of biases affect the platform's ability to make decisions. We show that for a large class of interventions ("positive interventions"), these biases cause the platform to launch too often. Through simulations calibrated to real-world data, we show that in many settings the treatment effect bias impacts decision-making more than the standard error bias. Based on joint work with Ramesh Johari, Inessa Liskovich, Gabriel Weintraub, and Geng Zhao.

Talk 2: In this work, we design a simple reinforcement learning (RL) agent that implements an optimistic version of Q-learning and establish through regret analysis that this agent can operate with some level of competence in an arbitrarily complex environment. While we leverage concepts from the literature on provably efficient RL, we consider a general agent-environment interface and provide a novel agent design and analysis. This level of generality positions our results to inform the design of future agents for operation in complex real environments. We establish that, as time progresses, our agent performs competitively relative to policies that require longer times to evaluate. The time it takes to approach asymptotic performance is polynomial in the complexity of the agent's state representation and the time required to evaluate the best policy that the agent can represent. Notably, there is no dependence on the complexity of the environment. The ultimate per-period performance loss of the agent is bounded by a constant multiple of a measure of distortion introduced by the agent's state representation. This work is the first to establish that an algorithm approaches this asymptotic condition within a tractable time frame.

Large deviations of nonlinear functions of adjacency matrices of sparse random graphs have gained considerable interest over the last decade. This includes popular examples like subgraph count, or the extreme eigenvalues. For the first half of the talk, we will discuss how the upper tail large deviation problem of subgraph count in a random regular graph can be reduced to a variational problem and how to solve such optimization. Next, we consider Erdos–Renyi graph $\mathcal{G}(n,p)$ in the regime of $p$ where largest eigenvalue is governed by localized statistics, such as high degree vertices. In particular, for $r \geq 1$ fixed, we will discuss the upper and lower tail probabilities of top $r$ eigenvalues jointly.

This talk is based on joint works with Bhaswar B. Bhattacharya, Amir Dembo, and Shirshendu Ganguly.