Learning Orders via Algorithmic Approach and Deep Learning Approach

The Hong Kong University of Science and Technology
Department of Computer Science and Engineering

Final Year Thesis Oral Defense

Title: "Learning Orders via Algorithmic Approach and Deep Learning Approach"

by

LIN Zizheng

Abstract:

Despite the advancement of Recommender Systems and Data Mining, there are 
still some real-life top-k ranking problems which are extremely 
challenging for machines. Hence, the intervention of human domain experts 
is essential. The crowdsourced top-k query is a paradigm that addresses 
this issue, where the preferences of numerous domain experts between two 
items are aggregated to serve as order information.

One challenge facing crowdsourced top-k query applications is that 
constantly asking human experts for their input may not be realistic. To 
mitigate this problem, there are several commercial platforms or business 
models allowing frequent interactions with experts during ranking.  
Nevertheless, the time required to constantly consult human experts might 
affect customer satisfaction. Therefore, the prior partial order 
information (i.e., the order information that is known without asking 
domain experts) can be exploited to minimize the interaction. This leads 
to the top-k Sorting Under Partial order Information (SUPI) problem. But 
the state-of-the-art solution is still not optimal and require too much 
preprocessing time.

Hence, we propose an algorithm, called LInear Quick Selection Sort 
(WOLIQSS), that is asymptotic optimal in terms of the total number of 
queries issued. We also greatly improve the efficiency of the 
preprocessing step. Both theoretical and empirical analysis of the 
proposed algorithm are provided.


Date            : 4 May 2019 (Saturday)

Time            : 10:40 - 11:20

Venue           : Room 2127A (near lift 19), HKUST

Advisor         : Dr. WONG Raymond Chi-Wing

2nd Reader      : Dr. NG Wilfred Siu-Hung