WPI Worcester Polytechnic Institute

Computer Science Department

AIRG Topics - Fall 2015

Version: Wed, 2 Dec 2015 11:02:29
Under Construction

AIRG meets on Thursdays at 11:00 a.m. in Beckett Conf. Rm.

Dates and topics for this semester are as follows:

Sept 3
CS AI Faculty
"AI Research @WPI Overview"
  • VIDEO Dmitry Berenson
  • Joe Beck
  • Dave Brown
  • Neil Heffernan
  • Carolina Ruiz

Sept 10
CS AI Faculty
"AI Research @WPI Overview"
  • Xiangnan Kong
  • Dmitry Korkin
  • SLIDES Yanhua Li
  • Eugene Eberbach

Sept 17
Dave Brown
"Some Issues in Computational Design Creativity"
    Computational Creativity is still at the cutting edge of AI research: we don't yet know how to do it in general, but a start has been made. While biased towards Engineering Design, much of this applies to evaluating the creativity of a computational system in any domain.

Sept 24
Xinyue Liu
"Organizational Chart Inference"
Advisor: Xiangnan Kong
    Nowadays, to facilitate the communication and cooperation among employees, a new family of online social networks has been adopted in many companies, which are called the "enterprise social networks" (ESNs). ESNs can provide employees with various professional services to help them deal with daily work issues. Meanwhile, employees in companies are usually organized into different hierarchies according to the relative ranks of their positions. The company internal management structure can be outlined with the organizational chart visually, which is normally confidential to the public out of privacy and security concerns. In this paper, we want to study the IOC (Inference of Organizational Chart) problem to identify the company internal organizational chart based on the heterogeneous online ESN launched in it. IOC is very challenging to address as, to guarantee smooth operations, the internal organizational charts of companies need to meet certain structural requirements (about its depth and width). To solve the IOC problem, a novel unsupervised method "Create" (ChArT REcovEr) is proposed in this research. It consists of 3 steps: (1) social stratification of ESN users into different social classes, (2) supervision link inference from managers to subordinates, and (3) consecutive social classes matching to prune the redundant supervision links. Extensive experiments conducted on real-world online ESN dataset demonstrate that Create can perform very well in addressing the IOC problem.

Oct 1
Antonio Umali
"Early stage work on developing a semi-autonomous robot assistant for PPE removal"
Advisor: Dmitry Berenson
    Compliant and cost-effective personal and industrial robots have allowed for more opportunities for Human-Robot Collaboration (HRC). In this work, we present a framework for using the Baxter robot in aiding health-care workers in the removal of biologically-contaminated clothing and equipment. Wearing this protective equipment is difficult for humans because it traps body heat and limits dexterity. Thus our approach seeks to minimize risk while also taking the least amount of time possible to execute. We use Baxter as a semi-autonomous assistant instead of a fully autonomous robot and take advantage of its compliant actuation. We present a systematic way of decomposing high-risk tasks into a series of human-robot actions which reduce the human's risk of exposure to infection. We also present a few key modifications to the Baxter robot which are likely to significantly improve its effectiveness for such tasks.

Oct 8
Neil Heffernan
Educational Data Mining

Oct 15
No Meeting: last day of A term

Oct 22
Break: no meeting

Oct 29
Ahmedul Kabir
"Predictive Modeling of the Outcome of Patients after Stroke"
    The outcome of a stroke patient is assessed after 90 days of stroke onset, and is measured by the Modified Rankin Scale (mRS) score. The objective of our research is to predict this outcome at the time of the patients' discharge from the hospital. For this purpose, we have built a dataset in collaboration with the UMass Medical School, Worcester, consisting of the stroke patients' demographic information, past medical history, treatment records and laboratory test results. Numerous techniques for predictive analysis have been applied on the data, including well-known classification algorithms, prediction via clustering, instance-level complexity analysis and discriminant analysis. This presentation discusses the application and results of these techniques in details, and sheds lights on future directions in this research.

Nov 5
John Boaz Lee
"Deep Learning Architecture with Dynamically Programmed Layers for Brain Connectome Prediction"
Advisor: Xiangnan Kong
    This paper explores the idea of using deep neural network architecture with dynamically programmed layers for brain connectome prediction problem. Understanding the brain connectome structure is a very interesting and a challenging problem. It is critical in the research for epilepsy and other neuropathological diseases. We introduce a new deep learning architecture that exploits the spatial and temporal nature of the neuronal activation data. The architecture consists of a combination of Convolutional layer and a Recurrent layer for predicting the connectome of neurons based on their time-series of activation data. The key contribution of this paper is a dynamically programmed layer that is critical in determining the alignment between the neuronal activations of pair-wise combinations of neurons.

Nov 12
No meeting

Nov 19
Hao Wan
What should the agent do next?
Advisor: Joe Beck
    Many tasks in AI can be phrased in terms of "what should the agent do next?" One approach for approaching such tasks is to treat them as multi-armed bandit problems. The multi-armed bandit framework looks at the observed past payoffs for various actions, and gives strategies for selecting which action to try next that balances obtaining additional information vs. selecting the action currently believed to be best. We examine an online educational platform called ASSISTments in the context of a bandit problem. For our task, we look at which method of presenting information to a student leads to the most learning. We used three different bandit algorithms (epsilon-greedy, softmax, softmax with simulated annealing) and, as a control, a simple algorithm based on statistical hypothesis testing (t-test). We evaluated the algorithms on two past experiments in ASSISTments. Our findings were that all of the approaches did better than a purely random strategy with respect to allocating students. Surprisingly, the simple statistical approach was hard to beat, and in many cases it performed the best. We suspect the strong performance of our control technique is due to characteristics of our problem: namely, a small number of decisions (2 or 3 experimental conditions) relative to other control problems, and a relatively homogeneous set of payoffs.

Nov 26
Thanksgiving Break

Dec 3
Dmitry Korkin
Machine learning techniques in studying complex diseases

Dec 10
title tbd

Dec 17
Ermal Toto
Predicting the Pulse of the City: A Real Time System for Multi-Granular Predictions of Crowd Flow
Advisor: Yanhua Li
    In recent years, the fast pace of urbanization has caused the rise of complex multi-modal transportation systems that operate under dynamic and interrelated traffic conditions. Accurate forecasts of traffic dynamics throughout the transportation system would enable smart services to mitigate urban challenges. In this presentation we explore 21 days of data from the Shenzhen subway network and discuss methods for dynamic feature selection based on network properties that are learned from data. Preliminary results show that temporal and spatial properties can be learned and utilized to reduce model complexity and increase accuracy.

-- Dec 18 -- Last day of Semester

[Return to the WPI Homepage] [Return to the CS Homepage]

AIRG Coordinator