See a listing of previous talks.

Xin Ru Wang, General Examination

Title: Multimodal machine learning techniques for naturalistic unlabelled long-term neural, audio and video recordings

Advisors: Rajesh Rao and Bingni Brunton

Supervisory Committee: Rajesh Rao (Co-Chair), Bingni Brunton (Co-Chair), Katherine Steele (GSR, ME), and Ali Farhadi

Abstract

In our data-rich world today, the need for better capture and storage of large everyday data are spurring the development of many techniques. However, methods for analysis are still often only trained and tested on small sets of well-curated and labelled samples. To truly garner the potential of the huge amounts of real-world data being recorded every second, we must develop techniques that can not only interpret the signals in lab-like conditions but remain robust in natural circumstances. Data recorded from and for bioelectronic technologies are especially noisy, such as ones for developing useful interfaces between brains and machines (BCI). My research develops and integrates multimodal machine learning techniques to tackle these large, noisy and unlabelled recordings. In my presentation, I will explore the connections between the fields of computer vision, speech recognition, machine learning and neural decoding. I will also summarize my previous work in applying deep learning to a long-term naturalistic multimodal dataset with neural, video and audio recordings to predict future human movements. The proposed work will regress as well as detect future movements, predict natural speech and attempt to map the human brain with semantic associations. As well, I propose to share this rare dataset with the wider scientific community.

When: 4 Oct 2017 - 11:00am Where: CSE 678

Minjoon Seo, General Examination

Title: Learning to Read and Reason by Answering Questions

Advisors: Ali Farhadi and Hannaneh Hajishrizi

Supervisory Committee: Ali Farhadi (Co-Chair), Hannaneh Hajishrizi (Co-Chair, EE), Gina-Anne Levow (GSR, Linguistics), and Oren Etzioni

Abstract: Reasoning can be defined as an ability to produce new knowledge from existing knowledge. While reasoning in a formally defined (logical) environment is well-studied, constructing the ontology and annotating training data for grounding surface forms (natural language) require tremendous amount of effort from human experts, still without guarantee of success. In this talk, I will discuss two recent projects on learning to (1) read and comprehend (ground) natural language and (2) reason over multiple facts in a synthetic setup, both with end-to-end training and minimal supervision of question answers. Then I will demonstrate that while language understanding and reasoning could be tackled one by one, there is a clear challenge in learning both at the same time. I will finally end my talk with my thesis proposal for immediate next steps to address the challenge and contribute towards building a general-purpose reasoning model.

When: 17 Oct 2017 - 3:30pm Where: CSE 403

Sam Elliott, Qualifying Project Presentation

Title: Putting the Checks into Checked C

Advisors: Ras Bodik and Dan Grossman

Abstract: 

Checked C is an extension to C that aims to provide a route for
programmers to upgrade their existing C programs to a safer language
without losing the low-level control they enjoy. The first origin of
unsafe code that Checked C is addressing is spatial memory safety,
such as buffer overruns and out-of-bounds memory accesses.

Checked C addresses these memory safety problems by adding new pointer
and array types to C, and a method for annotating pointers with
expressions denoting the bounds of the objects they reference in
memory. To ensure memory safety, the Checked C compiler uses a mixture
of static and dynamic checks over these additions to the C language.

This talk concerns these Dynamic Checks, and the
algorithms and infrastructure required to support them, including: the
soundness property Checked C is aiming to preserve, propagation rules
for bounds information, and the code generation algorithm for the
checks themselves. This talk includes an evaluation of these dynamic
bounds checks, their overhead, and their interaction with a
state-of-the-art optimizer.

When: 31 Oct 2017 - 3:00pm Where: CSE 203

Xuan Luo, Qualifying Project Presentation

Title: Pepper's Cone: An Inexpensive Do-It-Yourself 3D Display

Advisors: Steve Seitz and Jason Lawrence (Google)

Abstract: In this talk, I will introduce a simple 3D display that can be built from a tablet computer and a plastic sheet folded into a cone. This display allows viewing a three-dimensional object from any direction over a 360-degree path of travel without the use of special glasses. Inspired by the classic Pepper's Ghost illusion, our approach uses a curved transparent surface to reflect the image displayed on a 2D display. By properly pre-distorting the displayed image our system can produce a perspective-correct image to the viewer that appears to be suspended inside the reflector. We use the gyroscope integrated into modern tablets to adjust the rendered image based on the relative orientation of the viewer. Our particular reflector geometry was determined by analyzing the optical performance and stereo-compatibility of a space of rotationally-symmetric conic surfaces. We present several prototypes along with side-by-side comparisons with reference images.

When: 31 Oct 2017 - 4:00pm Where: CSE 624

John Thickstun, Qualifying Project Presentation

Title: Supervising Music Transcription

Advisors: Sham Kakade and Zaid Harchaoui (Stat)

Abstract: Music transcription can be viewed as a multi-label classification problem, in which we identify notes present in an audio recording at time t based on a two-sided contextual window of audio surrounding t. We introduce a large-scale dataset, MusicNet, consisting of music recordings and labels suitable to supervising transcription and other learning tasks. We will discuss the construction of the dataset, including optimal-alignment protocols and the value of side-information. We will then turn our attention to network architectures and data augmentations that lead to state-of-the-art performance for music transcription. Along the way, we will consider several scientific questions: what are the low level features of musical audio? What are the invariances of these recordings?

When: 2 Nov 2017 - 1:30pm Where: CSE 303

Pavel Panchekha, General Examination

Title: Automated reasoning about web page layout

Advisors: Michael Ernst and Zachary Tatlock

Supervisory Committee: Michael Ernst (Co-Chair), Zachary Tatlock (Co-Chair), Andy Ko (GSR, iSchool), Shoaib Kamil (Adobe Research), and James Fogarty

Abstract: Web pages are a common way for people to interact with applications critically important to their health, education, and livelihood. Web pages must correspondingly be accessible to all and easily usable. However, the technologies that web pages are developed in, chiefly HTML and CSS, are complex, with many unusual features and special cases. This talk discusses Cassius, a new attempt to make it easier to test and verify web page accessibility and usability. Cassius is a mathematical description of the web page layout algorithm used by web browsers, and an encoding of this specification to Satisfiability Modulo Theories, a format that allows efficient automated reasoning about web pages. Cassius has been used to build VizAssert, a tool to verify accessibility and usability assertions of a web page over all possible user configurations. The talk will discuss Cassius, VizAssert, and chart out ongoing work to make Cassius modular, allowing efficient reasoning about large web pages and web page components.

When: 3 Nov 2017 - 9:30am Where: CSE 303

Antoine Kaufmann, General Examination

Title: TBA

Advisor: Tom Anderson

Supervisor Committee: Tom Anderson (Chair), Scott Hauck (GSR, EE), Simon Peter (University of Texas- Austin), Xi Wang, Arvind Krishnamurthy, and Michael Taylor

Abstract: TBA

When: 6 Nov 2017 - 2:30pm Where: CSE 403

Qiao Zhang, General Examination

Title: TBA

Advisors: Tom Anderson and Arvind Krishnamurthy

Supervisory Committee: Tom Anderson (Co-Chair), Arvind Krishnamurthy (Co-Chair), Sreeram Kannan (GSR, EE), and Xi Wang

Abstract: TBA

When: 21 Nov 2017 - 11:00am Where: CSE 203

Haichen Shen, General Examination

Title: TBA

Advisor: Arvind Krishnamurthy

Supervisory Committee: Arvind Krishnamurthy (Chair), Raadhakrishnan Poovendran (GSR, EE), Matthai Philipose (Microsoft), Tom Anderson, and Ali Farhadi

Abstract: TBA

When: 28 Nov 2017 - 2:30pm Where: CSE 203

Aaron Bauer, General Examination

Title: TBA

Advisor: Zoran Popovic

Supervisory Committee: Zoran Popovic (Chair), Andy Ko (GSR, iSchool), Steve Tanimoto, and Dan Weld

Abstract: TBA

When: 29 Nov 2017 - 9:00am Where: CSE 203

Alex Takakuwa, General Examination

Title: TBA

Advisor: Yoshi Kohno

Supervisory Committee: Yoshi Kohno (Chair), Edward Mack (GSR, Asian Languages & Literature), Alexei Czeskis (Google), and Franzi Roesner

Abstract: TBA

When: 1 Dec 2017 - 2:00pm Where: CSE 203