Haofu Liao

Candidate for Ph.D. in Computer Science
University of Rochester, Rochester, NY



About Me

I'm a Ph.D. candidate in Computer Science at University of Rochester, Rochester, NY, advised by Prof. Jiebo Luo. Prior to coming to Rochester, I received my M.S. degree in Electrical and Computer Engineering from the Northeastern University, Boston, MA in 2015 and my B.E. degree from the Beijing University of Posts and Telecommunications, Beijing, China in 2012.

I'm interested in leveraging the latest advances in deep learning and machine learning to solve problems in data mining and computer vision.

News

  • [11/28/2016] One paper is accepted by AAAI 2017 Workshop (W3PHIAI).
  • [10/10/2016] One paper is accepted by BigData 2016.
  • [08/22/2016] One paper is accepted by ICPR 2016.
  • [04/24/2016] I am going to have an internship at VisualDx during the summer.
  • [04/23/2016] I am the reviewer of the ACM Multimedia (MM) 2016 .
  • [10/12/2015] I am the reviewer of the journal Pattern Recognition .

Publication

Conference Proceedings

Haofu Liao, Jiebo Luo, "A Deep Multitask Learning Approach to Skin Lesion Classification" AAAI 2017 Joint Workshop on Health Intelligence W3PHIAI 2017 (W3PHI & HIAI), San Francisco, CA, February 2017.
Haofu Liao, Yucheng Li, Tianran Hu, Jiebo Luo, "Inferring Restaurant Styles by Mining Crowd Sourced Photos from User-Review Websites" Big Data (Big Data), 2016 IEEE International Conference on, Washington, D.C., 2016, Acceptance Rate 19.39%.
Haofu Liao, Yuncheng Li, Jiebo Luo, "Skin Disease Classification versus Skin Lesion Characterization: Achieving Robust Diagnosis using Multi-label Deep Neural Networks",Pattern Recognition (ICPR), 2016 23rd International Conference on, Cancun, Mexico, 2016
Haofu Liao, Ping Chen, "Explore a Way of Improving the Computational Efficiency of the Block in Scilab", in Proceedings of 2011 IEEE International Workshop on OSSC, pp. 13-16, Oct. 12-14, 2011.
Zhenzi Liu, Ping Chen, Haofu Liao, Huichao Zhou, "The Modeling and Simulation of Rayleigh Flat Fading Channels based on Scilab", in Proceedings of 2011 IEEE International Workshop on OSSC, pp. 27-31, Oct. 12-14, 2011.

Book Chapters

Ping Chen, Xinghua Dong, Huichao Zhou, "Communication System Simulation Experiments Tutorial Based on Scilab\SCICOS", Tingshua University Press, Nov. 16, 2012. Contributed Section 3.1, 3.3 and 3.4 on SCICOS development introduction.

Research

A Deep Learning Based Decision Support System For Dermatology Diagnosis
Advisors: Prof. Jiebo Luo, Visual Intelligence & Social Multimedia Analytics (VIStA) Laboratory
Sep. 2015 - Present

 
Created a dermatology image dataset from multiple on-line skin disease atlases. To our best knowledge, our dataset, which contains 75,665 dermatology images from 2,113 skin diseases, is the largest and most comprehensive dermatology image dataset
 
Built several deep convolutional neural networks to classify skin diseases, skin lesions and their anatomical locations
 
Designed a deep multi-task learning based skin lesion classification method using the state-of-the-art CNN model
 
Developed a prototype of the skin lesion classification system on IOS
 
Two papers came out of this project

Inferring Restaurant Styles From User-Review Restaurants
Advisors: Prof. Jiebo Luo, Visual Intelligence & Social Multimedia Analytics (VIStA) Laboratory
Jan. 2016 - Jun. 2016

 
Created a restaurant photo dataset based on user-submitted photos from TripAdvisor
 
Built a multi-label convolutional neural network to label user-submitted photos with restaurant style attributes
 
Designed the pseudo tagging and restaurant profiling algorithms to infer restaurant styles from labeled restaurant photos
 
One paper came out of this project

Do They All Look the Same? Deciphering Chinese, Japanese and Koreans Using Big Data Computer Vision
Advisors: Prof. Jiebo Luo, Visual Intelligence & Social Multimedia Analytics (VIStA) Laboratory
May. 2016 - Sep. 2016

 
Built an Asian face classifier using state-of-the-art CNN model
 
Conducted several experiments upon the Asian face classifier and analyzed the experimental results
 
One paper came out of this project

High-Performance Kernel Smoothing Toolbox for Biomedical Image Processing
Advisors: Prof. Deniz Erdogmus, Cognitive System Laboratory (CSL)
Fall 2013 - Summer 2015

 
Studied kernel density estimation (KDE), bandwidth selection, kernel density derivative estimation (KDDE) algorithms and related applications on biomedical images
 
Designed a KDE based mathematic library in C++, which included a set of univariate kernel cores, bandwidth selectors, KDE and KDDE modules. It enabled users to build their own methods by selecting components from this library. The C++ Armadillo library was used to support linear algebra computations.
 
Accelerated the toolbox using parallel processing with both CUDA and Pthread
 
Added a full Matlab support for the library

Vessel Segmentation for Retinal Images and Cerebral Microvasculature Images
Advisors: Prof. Deniz Erdogmus, Cognitive System Laboratory (CSL)
Fall 2013 - Summer 2015

 
Studied curvilinear structure tracing, vessel enhancement filtering, and density estimation techniques, such as principal curve tracing, Frangi filter, and kernel smoothing
 
Implemented a decision-tree algorithm in Matlab, which measures the ridgeness based on the combined decision from a principal curve formula, the eigenvalues of Hessian matrix, and kernel density values
 
Improved the performance of Frangi filter by introducing a kernel curvature estimator with an unconstrained bandwidth which allowed the kernel estimator to smooth in any direction whether coordinate or not

Fast Level Set Method Based Image Segmentation for Luggage Inspection
Advisor: Prof. David Kaeli, Computer Architecture Research Group (NUCAR)
Summer 2014

 
Investigated a series of level set methods and their speed function design
 
Improved existing fast level set method by an adaptive speed function based on Chan-Vese model
 
Implemented the improved fast level set method on GPU and optimized new speed function with parallel reduction
 
Optimized new algorithm by the latest Dynamic Parallelism and Hyper-Q features in CUDA

802.11 Physical Layer Beacon Design and Implementation on USRP
Advisor: Prof. Kaushik Chowdhury, MathWorks SMART Lab
Summer 2013

 
Helped design the 802.11 PHY Layer protocol for USRP
 
Designed an 802.11 WLAN transmitter & receiver in MATLAB
 
Connected the 802.11 WLAN transmitter & receiver to USRP and tested their performance

Work

VisualDx, Rochester, NY
Research Intern, Advisor: Dr. Art Papier M.D.
Jun. 2016 - Sep. 2016

 
This is the follow-up research internship for the "A Deep Learning Based Decision Support System For Dermatology Diagnosis" project, see research experience for details

EMC Corp., Hopkinton, MA
System Engineer, Co-op (Full-time), Advisor: Dr. Sachin Adnan
Spring 2014 - Summer 2014

 
Designed a pipeline parallelism based T2 data processing tool on distributed servers, built a T2 data database for statistic analysis of storage system's workload, and generated a series of reports about read/write characteristics of the storage systems
 
Designed a Disk QOS simulation tool, simulated several Disk QOS algorithms and generated a report about their latency and throughput performance
 
Learned the configuration steps about VMAX and maintained several servers and storage systems in the lab

Shanghai Academy of Spaceflight Technology, Shanghai, China
Researcher, Intern (Full-time), Advisor: Prof. Ping Chen
Spring 2012

 
Studied a series of independent component analysis algorithms, such as Infomax, FastICA, and JADE
 
Designed a simulation tool for satellite communication in Matlab
 
Designed a satellite signal reception algorithm based on blind signal separation and independent component analysis
 
Simulated and optimized the performance of this algorithm for real satellite signals

Anjin Medical Technology (Beijing) Co., Ltd, Beijing, China
Embedded Engineer, Intern (Part-time), Advisor: Dr. Wei Zhao
Fall 2010 - Spring 2011

 
Developed the circuit for the LCD module of the surgery workstation
 
Designed a GUI in C for the same LCD module
 
Learned and supported the development of CPU module of the surgery workstation

Projects

Subset-Sum Problem Solvers: Implementation and Evaluation
Advisor: Prof. Waleed Meleis, Course Project for Combinatorial Optimization
Fall 2013

 
Developed a benchmark that consists of instances generated from different test problems for the evaluation of Subset-Sum Problem
 
Looked up all complexity results related to the Subset-sum Problem and its sub-problems
 
Studied and implemented the exhaustive, greedy, ILP, deepest decent and tabu search algorithms for the Subset-sum Problem
 
Evaluated the performance of Subset-sum Problem solvers and gave the best solver for different types of Subset-sum problems

Face Recognition Using Eigenfaces
Advisor: Prof. Deniz Erdogmus, Coding Project for Cognitive System Laboratory Interview
Summer 2013

 
Wrote a C++ API for Matlab, which allowed the C++ ALGIB package to communicate with Matlab
 
Studied and implemented a fast PCA algorithm which proved to have a better performance on face recognition
 
Implemented the eigenface method based on the fast PCA algorithm and C++ ALGIB package, and classified human face images from the AT&T face database

Human Detection with HOG Descriptors and SVM
Advisor: Prof. Jennifer Dy, Course Project for Pattern Recognition
Spring 2013

 
Implemented several static HOG descriptors in Matlab
 
Designed and trained a SVM classifier based on the feature vectors extracted from the INRIA Person dataset and MIT pedestrian dataset using HOG descriptors.
 
Developed a GUI supported human detection toolbox in Matlab

Spatial Domain Image Enhancement Toolbox Using Matlab
Advisor: Prof. Vinay Ingle, Course Project for Two Dimensional Image & Signal Processing
Fall 2012

 
Studied and implemented a set of spatial domain image enhancement techniques in Matlab
 
Built an image enhancement toolbox to demonstrate the relationship between transformation functions and the processing results
 
Analyzed effectiveness of different image enhancement techniques

Cloud Based Intelligent Irrigation Systems
Advisor: Prof. Li Guo, Group Leader, Innovation Lab of BUPT
Summer 2011

 
Established the develop environment (OS and interfaces) for embedded irrigation device
 
Created a database for storing historical irrigation data such as temperature and humidity on Windows Azure cloud platform
 
Developed a Windows Communication Foundation (WFC) based backend communication tool, which supported web communication between embedded irrigation device and smartphone or internet users