(The talk will be given in English)

Speaker:     Dr. Roy Schwartz
                   University of Washington and the Allen Institute for AI.

Monday, December 10^th, 2018
15:00 - 16:00

Room 011, Kitot Bldg., Faculty of Engineering

Towards Interpretable Deep Learning for Natural Language Processing

Abstract

Despite their superb empirical performance, deep learning models for natural language processing (NLP) are often considered black boxes, as relatively little is known as to what accounts for their success. This lack of understanding turns model development into a slow and expensive trial-and-error process, which limits many researchers from developing state-of-the-art models. Customers of deep learning also suffer from this lack of understanding, because they are using tools that they cannot interpret. In this talk I will show that many deep learning models are much more understandable than originally thought.

I will present links between several deep learning models and classical NLP models: weighted finite-state automata. As the theory behind the latter is well studied, these findings allow for the development of more interpretable and better-performing NLP models. As a case study, I will focus on convolutional neural networks (ConvNets), one of the most widely used deep models in NLP. I will show that ConvNets are mathematically equivalent to a simple, linear chain weighted finite-state automaton. By uncovering this link, I will present an extension of ConvNets that is both more robust and more interpretable than the original model. I will then present similar observations regarding six recently introduced recurrent neural network (RNN) models, demonstrating the empirical benefits of these findings to the performance of NLP systems.

This is joint work with Hao Peng, Sam Thomson and Noah A. Smith
 

Short Bio

Roy Schwartz is a postdoctoral researcher at the University of Washington and the Allen institute for AI. Roy's research focuses on improving deep learning models for natural language processing by gaining mathematical and linguistic understanding of these models. He received his Ph.D. and M.Sc. in Computer Science and his B.Sc. in Computer Science and Cognitive Science from the Hebrew University. Roy has won a best paper award at RepL4NLP 2018, as well as a Hoffman leadership and responsibility fellowship.

You are invited to attend a lecture

(Selected) Frontiers in plasma science

:By

Shurik Yatom

Princeton University, Princeton Plasma Physics Laboratory

Abstract

Plasma is a partially of fully ionized gas, comprising the 99 % of the visible matter in the universe. The physical plasma is scarce in nature on Earth, however is continuingly gains popularity in the science and technology fields, since being described by Langmuir in 1920s. Currently, multiple directions of plasma related research are prominent in a variety of scientific and technological fields: from the quest for fusion energy, through plasma propulsion for space-travel, to plasma medicine, agriculture and material fabrication and functionalization.  The potential for applications is indeed vast, however so are the scientific challenges that emerge in the context of the above applications. In this talk I will concentrate on two exciting areas: plasma-assisted synthesis of nanomaterials and the plasma interactions with liquids, biological tissues and cells. Both of these research areas are incredibly complex, owing to multiple different interacting species with a wide variety of energies, sizes and chemical potentials, interacting across all the possible phases of matter, from gas to liquid, solid and crystal phases, presenting many exciting physical challenges. Notably, both these areas consider multiphase environment, posing many scientific challenges, such as: large density and temperature, unknown plasma characteristics, the control of the reactivity transfer at the plasma–liquid/solid interface, interfacial charging, droplet transport and self-organization of plasmas in contact with liquid/solid. Similar challenges occur in the interface of plasma with biological tissues and cells. The phenomenon of plasma self-organization into coherent structures and patterns is particularly interesting, because it modulates the characteristics of the plasma, influencing the density and the energy of the charged particles, chemical composition, species transport along and across the plasma-tissue/liquid interface, radiation and electric fields. The aim of this seminar is to introduce these selected frontiers of plasma science, discuss their promise, challenges and the experimental approach to their investigation.

On Thursday, Nov 22, 2018, 15:00

Room 011, Kitot building

You are invited to attend a lecture

Chip-scale metrology: taming atoms, frequency combs and cavities

:By

Liron Stern

National Institute of Standards and Technology (NIST), Time & Frequency division, 325 Broadway Boulder CO, USA

Metrology, the science of measurement, strives to push the limits of human ability to measure quantities such as time, frequency, distance, temperature and mass. This ongoing global effort provides us with a reference “ruler” which is crucial both for understanding the fundamental nature of our universe, as well as an enabling and driving new technologies. Central to such measurements are quantum sensors, such as atomic clocks, magnetometers and superconducting voltage standards which are continuously driving today’s technological revolutions. Nowadays, we are witnessing a universal endeavor to miniaturize such “measuring machines” with two prime motivations: understanding the fundamentals of light-matter interactions at these extreme limits, and enabling new applications in disciplines such as telecommunication, space exploration and medical devices.

In this talk, I will present how we move forward to enable chip-scale frequency metrology. First, I will introduce the recently discovered microresonator based Kerr-soliton frequency combs, and experiments and calculations where we have used these as a spectroscopic tool to directly interact with atomic vapour. I will present experimental results where we construct a new type of micro-machined atomic vapour cells where diffractive optics are merged with atomic physics. I will describe how these devices map atomic states to the spatial distribution of diffracted light, and how we can utilize these effects to allow offset frequency locks with high stability. Finally, I will present a miniaturized cavity with a quality factor of a billion, interfaced with an atomic micromachined atomic cell, providing ~100 Hz of linewidth and stable optical frequency. To conclude, I will show how interfacing atoms, Kerr-combs, and miniaturized cavities paves the way to a fully chip-scale metrology system, and discuss the huge impact such system may cast on science and technology.

On Sunday, November 18, 2018, 13:00

Room 011, Kitot building