Please note that for this term, the lab meetings will be held from 1230 hours every Wednesday. The venue for the lab meeting remains the same, TASC1 9408.
In the lab meeting on Wednesday, May 8, Ann Clifton is going to discuss the paper Emergence of Gricean Maxims from Multi-Agent Decision Theory. A short description is as follows:
” This paper trains multiple agents to perform a cooperative task using linguistic information and pragmatic reasoning through a decentralized partially observable markov decision process model. The authors argue that the models learned exhibit the properties of the “cooperative principle”, a pragmatic linguistic theory of how natural language conversation is conducted. ”
The lab meeting will start at 1100 hours at TASC1 9408.
In the lab meeting on Wednesday, April 24, 2013, Majid Razmara will give a talk on the paper Marginalized Denoising Autoencoders for Domain Adaptation. The meeting will be at TASC1 9408 and start at 1100 hours.
In the lab meeting on 10 April, 2013, Bruce Krayenhoff will give a talk. The details are below :
We will discuss the first-half of the paper Similarity Estimation Techniques from Rounding Algorithms. ‘Locality sensitive hashing schemes’ enable us to generate hashing functions h such that the probability that h(x)=h(y) is equal to some measure of the similarity between x and y. I will cover what locality sensitive hashing schemes are and what properties they have. In addition, we will consider one such hashing scheme for vectors.
Finally we will consider the definition of the EMD (Earth Mover Distance), and how it can be interpreted. The paper discusses how to use hashing functions to approximate the EMD, however I will be leaving this to a future lab meeting.
The lab meeting will be at TASC1 9408 and start at 1100 hours.
David Lindberg will give a talk in the lab meeting on Wednesday, April 3 at TASC1 9408. A short description is below :
I will begin with a quick review of my question generation(QG) research, describing the context, scope and goals. After discussing previous efforts in evaluating the output of question generation systems, I will conclude with a discussion of the results and implications of an evaluation of my QG system’s output
The talk will begin at 1100 hours.