Modeling Sequences with Memoization
Dongwoo Kim is an assistant professor at the Department of Computer Science and Engineering, POSTECH. Prior to POSTECH, he worked as an assistant professor (lecturer) at the Australian National University. He received his Ph.D. from KAIST in 2015, under the supervison of professor Alice Oh. Applications in his interests include machine learning and its implication on human understanding.
Sequence prediction has been one of the fundamental problems in machine learning. From the hidden Markov models to recent recurrent neural networks, most of the prediction approaches have been focused on internal state transition, which inherently suffers from a long dependency structure in a sequence. In this talk, I will seek some alternative ways to model and predict discrete sequence based on memoization perspective. I will first present the importance of repeating sub-sequences, motifs, in a long sequence and derive an algorithm to predict the sequence based on the motifs. Statistical analysis will follow to show the theoretical guarantee on the performance of the proposed algorithm. In the second part of this talk, I will discuss a way to increase the representation performance of the proposed approach by incorporating a neural network structure. I will conclude the talk with a discussion on the implication of the proposed approach including a preliminary study on social media dataset.
ID : 897 821 7407
PW : 1nTQDY