Events

Speaker: Nabiha Asghar, PhD Candidate

We propose an online, end-to-end, neural generative conversational model for open-domain dialogue. It is trained using a unique combination of offline two-phase supervised learning and online human-in-the-loop active learning.

Speaker: Mohammed Alliheedi, PhD Candidate

Over the past decade, the focus on argumentation mining has been growing significantly in different areas of Artificial Intelligence (AI) research. The incentive to build Natural Language Processing (NLP) systems to automatically identify and analyze argumentative components in various genres of texts has increased because knowledge of argumentative structure facilitates various tasks such as text summarization and opinion mining for commercial purposes.

Speaker: Michael Cormier, PhD Candidate

In this talk I will present recent advances in our vision-based web page segmentation method. This method is an edge-based, Bayesian image segmentation algorithm, capable of estimating the semantic structure of a web page from an image of the page.

Speaker: Carolyn Lamb, PhD Candidate

Computer software has frequently been used to perform creative tasks, such as generating artwork or discovering new mathematical proofs. But if a computer is intended to be creative, how do we measure its performance or decide if it is doing a good job?

Speaker: Carolyn Lamb, PhD Candidate

Artists, hobbyists, cognitive scientists, and computer scientists are all trying their hands at writing computer programs that generate poetry. Why? What do these poems look like, and how are they made? What might meta-poetry writers on both sides of the art-science spectrum learn from each other?

Speaker: Ricardo Salmon, PhD Candidate

Stochastic satisfiability (SSAT) problems are an extension of SAT problems that is PSPACE-complete and in the same complexity class as quantified Boolean formula (QBF) and partially observable Markov decision processes (POMDPs).

Speaker: James Wright, Microsoft Research

In order to do a good job of interacting with people, a system must have an adequate model of how people will react to its actions. This is particularly true in strategic settings: settings that contain multiple agents, each with their own goals and priorities, in which each agent's ability to accomplish their goals depends partly on the actions of the other agents. Standard models of strategic behavior assume that the participants are perfectly rational. However, a wealth of experimental evidence shows that not only do human agents fail to behave according to these models, but that they frequently deviate from these models' predictions in a predictable, systematic way.

Speaker: Haifeng Xu, University of Southern California

Strategic interactions among self-interested agents (a.k.a., games) are ubiquitous, ranging from economic activity in daily life and the Internet to defender-adversary interactions in national security.  A key variable influencing agents' strategic decision making is the information they have available about their environment as well as the preferences and actions of others. In this talk, I will describe my work on computational questions pertaining to the role of information in games.

Speaker: Daniel Recoskie, PhD candidate

We propose a method for learning wavelet filters directly from data. We accomplish this by framing the discrete wavelet transform as a modified convolutional neural network. We introduce an autoencoder wavelet transform network that is trained using gradient descent.

Speaker: Daniel Recoskie, PhD candidate

We propose a new method for learning filters for the 2D discrete wavelet transform. We extend our previous work on the 1D wavelet transform in order to process images. We show that the 2D wavelet transform can be represented as a modified convolutional neural network (CNN). Doing so allows us to learn wavelet filters from data by gradient descent. Our learned wavelets are similar to traditional wavelets which are typically derived using Fourier methods.