• Showcases CS548 Fall 2017
• Showcases CS548 Fall 2016
• Showcases CS548 Spring 2016
• Showcases CS548 Spring 2015
• Showcases CS548 Spring 2014

Sept. 18: Decision Trees

• Students: Holly Nguyen, Hongyu Pan, Lei Shi, Muhammad Tahir.

• Application Topic/Title: "Using partial decision trees to predict Parkinson's symptoms: A new approach for diagnosis and therapy in patients suffering from Parkinson's disease"

• Short description: This article proposes a new method using partial decision trees and association rules to predict Parkinson's Disease (PD) symptoms using features relating to medical history, physical examinations, and medications. The partial decision trees and corresponding rules were applied to a dataset of 230 PD patients from the University Hospital of Ioannina. The proposed method for PD symptom prediction was designed to use in the PERFORM system which uses individual wearable monitors to capture motor information, which can then be used to predict new PD symptoms for each patient.

• Slides: CS548F18_Showcase_Decision_Trees

• Link to paper: Showcased Paper

Sept. 20: Model and Regression Trees

• Students: Geri Dimas and Han Liu

• Application Topic/Title: "Local models-based regression trees or very short-term wind speed prediction"

• Short description:
  This paper presents different variations of the Regression Tree algorithm and evaluates each in relation to the real-world problem of predicting short-term wind speeds from data collected on wind farms. The paper evaluates Regression Tree methods with other classical approaches finding favor in the Regression algorithms. In particular, the paper finds good performance in terms of the prediction errors on the testing data set and low computational learning and forecasting times.

• Slides: CS548F18_Showcase_Model_and_Regression_Trees

• Link to paper: Showcased Paper

Oct. 9: Artificial Neural Networks and Deep Learning

• Students: Serdarcan Dilbaz, Ricardo Flores, Rajendra Kante, Ashley Elizabeth Lutz

• Application Topic/Title: "Dermatologist-level classification of skin cancer with deep neural networks"

• Short description:
  This article demonstrates the diagnostic capabilities of deep convolutional neural networks (CNN) in classifying skin lesions as benign or malignant based on clinical images. The clinical images are fed to the CNN without any preprocessing. The paper utilizes the Inception v3 CNN architecture and outperforms the average of the dermatologists at skin cancer classification.

• Slides: CS548F18_Showcase_ANNs_and_DeepLearning

• Link to paper: Showcased Paper Note: search for this paper from the WPI Library webpage to access the full article.

Oct. 25: Clustering

• Students: Yingxue Zhang, Mo Cheng, Ashish Gurung, Jake Scheide.

• Application Topic/Title: "T-Drive: Enhancing Driving Directions with Taxi Drivers' Intelligence`"

• Short description:
  The paper discusses a route navigation system that uses a time-dependent landmark graph, to generate the fastest possible route to a destination. The data was collected from 33,000 taxi drivers over a period of 3 months. The model uses VE-Clustering to estimate travel time as a distribution. Two separate routing algorithms were used to generate the fastest route. 60-70% of the generated routes were found to be faster than the routes generated by competing applications.

• Slides: CS548F18_Showcase_Clustering

• Link to paper: Showcased Paper

Nov. 8: Anomaly Detection

• Students: Abhishek Shah, Marie Solman, Mohamed Mahdi Alouane, Satishraju Rajendran and Eno-Obong Inyang

• Application Topic/Title: "Anomaly Detection in Thermal Images using Deep Neural Networks"

• Short description:
  Infrared thermography is an important tool in industry to allow technicians to detect anomalous temperature conditions and prevent expensive maintenance. While infrared cameras are less expensive than ever, they do require a system and setup and manual detection can be costly and impractical. The proposed work applied a Deep Neural Network (DNN) to real thermal images of objects in comparison to their normal visual images. The work is done both with actual images and simulated images to predict anomalous temperatures for various instruments and in different environments.

• Slides: CS548F18_Showcase_Anomaly_Detection

• Link to paper: Showcased Paper

Nov. 13: Text Mining

• Students: Charlie Parsons, Sola Shirai, Shyam Subramanian, Erin Teeple and Jack Zhang.

• Application Topic/Title: "Deep Learning for Answer Sentence Selection"

• Short description:
  This Text Mining Showcase presents the work of Lei Yu, Karl Moritz Hermann, Phil Blunsom, and Stephen Pulman on Deep Learning for Answer Sentence Selection. Previous work in the area of question-answering and answer sentence selection has relied on feature engineering from text data and semantic encodings. The authors avoid this resource-intensive process by proposing a neural-network based model for matching questions with answers using semantic encoding to perform selection of answers to questions from among a set of candidate answers. Answer selection performance is evaluated in this study by comparing the mean average precision (% retrieved documents that are relevant) and mean reciprocal rank (the inverse rank of the first correct document in an ordered list of correct documents).

• Slides: CS548F18_Showcase_Text_Mining

• Link to paper: Showcased Paper

Nov. 29: Data Visualization

• Students: Russell Davis, Kianoosh Ghazi, Renee Sweeney, Huaming Sun.

• Application Topic/Title: "SmartAdP: Visual Analytics of Large-scaleaxi Trajectories for Selecting Billboard Locations"

• Short description:
  This paper presents SmartAdP, an interactive visual analytics framework using large-scale GPS trajectory data to deals with two challenges: finding good solutions in a huge solution space and comparing the solutions in a visual and intuitive manner. Particularly, it combines a novel visualization-driven mining model with several well-designed visualization and interaction techniques. This system can effectively and efficiently formulate good candidate solutions for not only advertising planning but also other location selection problems.

• Slides: CS548F18_Showcase_Data_Visualization

• Link to paper: Showcased Paper

Dec. 4: Web Mining

• Students: Mallak Alkhathlan, Jason Dsouza, Paritosh Goel, Preeti Havannavar, Sayed Inamdar

• Application Topic/Title: "Image-Matching Based Identification of Store Signage Using Web-Crawled Information"

• Short description:
  This paper addresses automatic matching of street images with relevant web resources to enable the identification of store signage in street images. The SIWI (Signage Identifier in Web-crawled information) system uses automated web-mining that includes data filtering and pruning strategies for dataset generation. Hybrid Image Matching approach is used for signage identification which combines the feature vector with feature point matching technique to achieve accuracy of 91% in real-life application.

• Slides: CS548F18_Showcase_Web_Mining

• Link to paper: Showcased Paper