Graduate

Machine Learning and its applications have significantly impacted how we live our lives and use technology to enhance it. Quantum computing has been one of the inevitable advances in technology that promises to take us into a new realm of computational power. This talk will be a broad general overview of what quantum computing power added to machine learning techniques could actually give us. The talk will provide a very non-technical introduction to quantum technologies and where it stands at the end of 2019 and similarly a very non technical overview of machine learning and AI. It will then describe a few example projects motivated by real problems in industry and possible approaches to solving them. The goal of the talk is to incite curiosity about the subjects, bust a few myths and hopefully leave the audience with more questions than answers.

About the speaker:

Dr Rishiraj Pravahan is co-founder and Director of INQNET (Intelligent Quantum Networks and Technology) at the AT&T Foundry in Palo Alto and a visiting scientist at Caltech. Rishiraj has created a new model for corporate innovation through collaborative research and development between, industry, government and academia. His work is focused on building quantum networks and technologies and uses of Artificial Intelligence. Prior to joining AT&T, Rishiraj worked for the ATLAS experiment at CERN where he was part of the team that discovered the Higgs Boson. Rishiraj is also a passionate teacher and advocate for science through public talks and seminars in the US, Europe, India and Latin America. His technical interests involve, understanding core networks, privacy and security of data and computation, collection, storage and analysis of sensor data and making advances in the frontiers of statistics, machine learning and Artificial Intelligence.

Ridge regression is a fundamental paradigm in machine learning and statistics, and it has long been known to be closely connected to the implicit regularization properties of gradient descent methods, cf. early stopping. Over the past decade, this connection has sparked research into a variety of directions aimed at developing computationally efficiency estimators, including acceleration, mini-batching, averaging, sketching, sub-sampling, preconditioning, and decentralization. Sparse recovery is another cornerstone of modern statistics and learning frameworks. Yet, here the connection to implicit regularization is not as well developed. Most results in the literature only involve limit statements (holding at convergence, for infinitesimal step sizes), apply to regimes with no (or limited) noise, and do not focus on computational efficiency. In this talk, we address the following question: Can we establish an implicit regularization theory for gradient descent to yield optimal sparse recovery in noisy settings, achieving minimax rates with the same cost of reading the data? We will highlight the key ideas to obtain the first results in this direction, along with a few surprising findings.

Computational analyses are playing an increasingly central role in research. However, many researchers have not received training in best practices and tools for reproducibly managing and sharing their code and data. This is a step-by-step, practical webinar on managing your research code and data for computationally reproducible collaboration. The webinar starts with some brief introductory information about computational reproducibility, but the bulk of the webinar is guided work with code and data. Participants move through best practices for organizing their files, automating their analyses, documentation, and submitting their code and data for publication.

Prerequisites: Participants should bring their own wifi-enabled laptop.

Audience: Researchers who use code in their research and wish to share it.

Workshop goals:
1. Learn best practices for file organization, documentation, automation, and dissemination.
2. Assess possible tools for managing code and data.
3. Build a collaborative workspace for your code and data on Code Ocean.

About Women of Silicon Valley 2020

With a progressive agenda engineered to bring attendees into the next generation of tech, Women of Silicon Valley will continue to support and inspire the women working in America's global capital of innovation. Bringing you an unprecedented line up of speakers, actionable insight, networking and expert-led workshops, this dynamic event will assist you in taking positive action in your own career and in your organization.

This event is built for women, by women and will act as a vital tool in your development and progression; join us and show why women in Silicon Valley do tech better than anywhere else.

RSVP by 5pm TUESDAY, DEC. 3

This talk will discuss maintaining funding for R&D in startups, large companies, and academia/government labs. This lunch seminar is part of a SOS/SPRC series on startups, entrepreneurship, and innovation in photonics focusing on formation/IPO of photonics startups, patents, IP protection and many more topics.

For those unable to come to Stanford, the lecture will be broadcast online as a free livestream.

Abstract:
The number π appears thousands of times in The Art of Computer Programming, in many different contexts. Dr. I. J. Matrix has remarked that its digits, "when properly interpreted," actually convey the entire history of the human race! [See page 41 of Volume 2.] This lecture will examine many interpretations of those digits, both proper and improper.

For more details, please see the event page or check out Dr. Knuth's site.

By 2050, the costs of an average PV and wind plant are expected to fall by 71% and 58%, respectively. Meanwhile, batteries will further depress market prices, which in turn enable the deeper penetration of renewable energies like PV, wind, and electric vehicles (EVs). However, the transition on primary energy resources can be a double-edged sword. Problems such as protective relay is landing and fault detection, protective relay coordination under environmental uncertainty, topology recovery of secondary distribution networks, and EV charging station planning are critical to the security and resilience of the electric systems. This presentation describes several timely solutions to enable more secure and efficient grid operations by analyzing voluminous power system operation data. The aforementioned solutions include the multifunction intelligent relays, an environment-driven adaptive protection scheme, a transformer connectivity inferencing tool, and an EV charging station planning method. Several types of machine learning algorithms are developed in power systems to support renewable energy integration for sustainability.

The panel will discuss future market and technology trends of battery systems and their applications in transportation and the grid. It will feature short presentations by the panelists followed by a Q&A period moderated by Abbas and Ram. The panelists will address future trends in battery technologies at the cell and pack levels and projections of key battery performance metrics, including energy density
power density, cost, degradation, and safety. They will also address questions such as: What are the projected requirements for grid and transportation applications? What technologies will dominate transportation applications? What technologies will dominate grid applications? Will secondary life become real? Will EV to grid become real? What are the challenges to be addressed? What should academia be working on?

PANELISTS:

Yi Cui
Professor,
Department of Materials Science and Engineering
Stanford University

Robert Tietje
VP for E-Mobility
Volkswagen Group of America

Blake Richetta
Chairman and CEO
Sonnen

Exams to conclude autumn 2019 quarter. 

Winter recess follows. Classes resume January 6, 2020.

Many modern data sources arising from social networks, biological data, etc. are best viewed as indexed by combinatorial structures such as graphs, rather than time series. The local weak limit theory for sparse graphs, also known as the objective method, due to Benjamini, Schramm, Aldous, Steele, Lyons and others, provides a framework which enables one to make sense of a stationary process indexed by graphs. The theory of time series is the engine driving an enormous range of applications in areas such as control theory, communications, information theory and signal processing. It is to be expected that a theory of stationary stochastic processes indexed by combinatorial structures, in particular graphs, would eventually have a similarly wide-ranging impact.

Employing the above framework, we introduce a notion of entropy for probability distributions on rooted graphs. This is a generalization of the notion of entropy introduced by Bordenave and Caputo to graphs which carry marks on their vertices and edges. Such marks can represent information on real-world data. For instance, in a social network graph where each node represents an individual and edges represent friendships, a vertex mark represents the type of an individual, while edge marks represent shared data between friends. The above notion of entropy can be considered as a natural counterpart for the Shannon entropy rate in the world of graphical data. We illustrate this by introducing a universal compression scheme for marked graphical data. Furthermore, we introduce an algorithm that can perform such a compression with low complexity.

This talk is based on joint work with Venkat Anantharam.