Deep learning

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

Final Year Thesis Oral Presentation

Title: "Deep learning"

By

Shuo HAN

Abstract

Deep learning, which refers to a set of algorithms in machine learning
that aim at mimicking the learning mechanism of human brains, has become
increasingly popular nowadays. The word 'deep' refers to the relatively
large number of learning layers, where concepts are extracted from lower
layers and used as input features for upper layers. Similar to the way
that human cognition works, the concepts in upper layers are more abstract
than those in lower layers. In recent years, deep learning methods have
achieved fairly high performance in object recognition, speech
recognition, and natural language processing, etc.

However, deep learning has not been widely appplied on time series data,
especially on financial data such as stock prices. In this thesis, we
first study thoroughly on convolutional neural networks (CNNs), which is a
powerful deep learning model that can detect and learn translate invariant
features from input data. Based on the knowledge we have, we build a CNN
model to identify stock plunges and the model yields good classification
performance. We also make use of CNNs to build a convolutional
autoencoder, an unsupervising learning model that aims to learn better and
more concise representation of the original data, in order to extract
transient but significant perturbations from stock price changes.

Date:                   Tuesday, 5 May 2015

Time:                   1:30 - 2:10pm

Venue:                  Room 5503
                        Lifts 25/26

Committee Members:      Prof. James Kwok (Supervisor)
                        Prof. Gary Chan (Reader)