**SPRING 2018 SCHEDULE**

**This seminar is held on Mondays at 4:30-5:20 PM in Building 200, Room 305**(unless otherwise noted).

**April 2: NO SEMINAR**

**April 9: NO SEMINAR **(due to special seminar below)

The Stanford Data Science Initiative, the Institute for Computational & Mathematical Engineering, and the Kavli Institute for Particle Astrophysics and Cosmology invite you to a special seminar on April 9 at 4:45pm in Oak Lounge at Tressider Union ---

Speaker: Sir Roger Penrose, Emeritus Rouse Ball Professor of Mathematics in the University of Oxford, Emeritus Fellow of Wadham College, Oxford

Title: Why Algorithmic Systems Possess No Understanding

Abstract: Many examples of highly effective algorithmic systems, such as AI devices, have been constructed in recent years. We have computer-controlled machines like self-driving cars and algorithmic systems that play chess and GO at levels that can out-perform even the best of human players. But do such devices actually “understand” what they are doing, in any reasonable sense of that word? I argue that they do not, and as an illustrative example I present a recently composed chess position that a human chess player, after briefly examining it, would correctly conclude that it is an obviously drawn position. Nevertheless, when it is presented to the top-level chess-playing program Fritz, set at grandmaster level, Fritz incorrectly claims that it is a win for the black pieces and eventually Fritz blunders dreadfully (though “correctly” according to its algorithm) to be soon check-mated by white. This demonstrates Fritz’s remarkable lack of any actual understanding of the game of chess, despite its vast computational abilities.

More sophisticated examples come from mathematics, most particularly with human understanding of the infinite, and it can be shown that this quality cannot plausibly be encapsulated by any algorithm arising from the processes of natural selection. I argue that the quality of understanding is a feature of consciousness, and that consciousness can come about only through physical processes not yet properly understood, most likely at the boundary between quantum and classical processes, as argued for in the Orch-OR proposal.

Bio: Penrose is known for his work in mathematical physics, in particular for his contributions to general relativity and cosmology. He has received several prizes and awards, including the 1988 Wolf Prize for physics, which he shared with Stephen Hawking for the Penrose–Hawking singularity theorems.

More detailed biographical information at his Wikipedia page and the Encyclopedia Brittanica

**April 16:**

**Hao Yin, ICME**

Title: The Local Closure Coefficient: A New Perspective On Network Clustering

Abstract: The clustering of edges in real-world networks is a fundamental property underlying many ideas and techniques in network science. Clustering is typically quantified by the clustering coefficient, which measures the fraction of pairs of neighbors of a given center node that are connected. However, many common explanations of edge clustering attribute the triangle closing edge to the head node (and not a center node) of a length-2 path—for example, “a friend of my friend is also my friend.” While such explanations are common in network analysis, there is no metric for measuring edge clustering that can be attributed to the head node.

Here we develop local closure coefficients as a metric quantifying head-node-based edge clustering. We define local closure coefficient as the fraction of length-2 paths emanating from the head node that induce a triangle. This subtle difference in definition leads to remarkably different properties from traditional clustering coefficients. We analyze correlations with node degree, connect the closure coefficient to community detection, and show that closure coefficients measured from a static graph closely correspond to temporal edge information.

This is a joint work with Austin R. Benson and Jure Leskovec.

Bio: Hao Yin is a third-year Ph.D. student in the Institute for Computational and Mathematical Engineering at Stanford University, under the supervision of Professor Tze Leung Lai. His research interests lie in network analysis, temporal analysis, and broadly speaking, machine learning and data mining. Before coming to Stanford, he obtains his B.S. degree in Mathematics from Fudan University.

**April 23: NO SEMINAR**

**April 30: Pavan B. Govindaraju, Mechanical Engineering**

Title: On Simulating Combustion of Multicomponent Fuels

Abstract: Alternative fuels are being designed to replace conventional fossil fuels and this currently relies on in-flight testing which is tedious, time-consuming and most of all, incredibly expensive. Computational methods, in conjunction with optimization strategies, can make a big impact in accelerating the design process, and help identify novel fuel mixtures. Prior to that, one must understand the difficulties in modeling the combustion process of these complicated fuels, make the problem computationally tractable and identify approximation regimes. This talk will discuss the simulation of multicomponent fuels in a spray-combustion setting and this requires i) a simplified framework for predicting multicomponent droplet evaporation ii) formulation of reduced descriptions of fuels iii) identifying regimes of multicomponent spray combustion and iv) the effect of turbulence on ignition of multicomponent sprays.

Bio: Pavan B. Govindaraju is a 5th year Ph.D. student in the Mechanical Engineering department. He is an ICME alumni and received an MS in 2016. Prior to this, he completed his

undergraduate degree from IIT Bombay, majoring in Mechanical Engineering with honors and minor in Computer Science. His research broadly focuses on using computational approaches to study energy and environmental solutions, particularly on alternative fuels, desalination and combustion.

**May 7: Chao Chen, ICME**

Title: A new rank-structured linear solver, its parallelization and application to ice sheet modeling.

Abstract: Solving linear systems is an important building block in many science and engineering applications, such as reservoir simulation, Gaussian regression, structural mechanics and ice sheet modeling. Most of existing solvers fall into two categories: direct methods (e.g., LU and Cholesky) and iterative methods (e.g., CG, MINRES and Multigrid). I will present a new rank-structured (a.k.a. data-sparse) method, its parallelization and application to ice sheet modeling. Compared to existing methods, the new solver can be faster than direct solvers and more robust than iterative methods. To solve large-scale problems e.g., linear systems arising from ice sheet modeling, the parallel algorithm was developed on distributed-memory machines.

Bio: Chao Chen is a 6th year Ph.D. student in ICME. Prior to this, he completed his undergraduate degree from Nankai University, majoring in Applied Mathematics. His research mainly focuses on using tools from both numerical analysis and parallel computing to develop algorithms and their implementations on modern supercomputers. The targeted applications he has studied include climate simulation, fluid/solid/structural mechanics, dislocation dynamics and so on.

**May 14: Kailai Xu, ICME**

Title: Isogeometric Analysis for the Fractional Laplacian

Abstract: Recent years have witnessed a boom in the research interest in the modeling using nonlocal operators. The fractional Laplacian, which is the generator of a symmetric $\alpha$-stable process, has been used for modeling turbulence, anomalous diffusion, option price, quantum physics, image denoising, etc. In this talk, I will present a novel isogeometric analysis approach for the fractional Laplacian on a 2D bounded domain. The method can deal with complex geometries and shows O(h^2) convergence numerically. I will also discuss the special regularity issues of the fractional Poisson equation and the proper assumptions on the data.

Bio: Kailai Xu is a second year Ph.D. student in ICME. He is currently working with Prof. Eric Darve on the fractional PDEs. He graduated from Peking University with a Bachelor's degree in computational mathematics. His research interest is in numerical methods, such as numerical PDEs, numerical analysis, optimization, etc.

**May 21: NO SEMINAR**

**May 28: NO SEMINAR (HOLIDAY)**

**June 4: Hadley Wickham, Chief Scientist at RStudio and Adjunct Professor of Statistics at University of Auckland, Stanford University and Rice University**

Location: 260-113

****Please note change in location****Time: 4:30-5:20pm

Title: An informal fireside chat with Hadley Wickham.

Hadley will answer questions about:

* writing books on R

* teaching R

* teaching STATS 337 (Readings in Applied Data Science)

* recent software projects

... and anything else you might be interested in.

Bio: Hadley Wickham is the Chief Scientist at RStudio and an Adjunct Professor of Statistics at University of Auckland, Stanford University and Rice University. He obtained his Ph.D. in Statistics from Iowa State University. He is interested in building tools that make data science easier, faster, and more fun. His work includes tools for data science (ggplot2, dplyr, tidyr); tools for data import (readr, httr, rvest); and software engineering tools (devtools, roxygen2, testthat). In 2006 he received the John M. Chambers Statistical Software Award for his work on ggplot and reshape R packages. He was named a Fellow by the American Statistical Association in 2015 for "pivotal contributions to statistical practice through innovative and pioneering research in statistical graphics and computing". He has also written books on R packages and R for data science.