Techniques and applications of deep learning

The aim of this seminar is to share research experience of deep learning techniques and applications, including work from the three research groups.

Location: Room C/2.07, Queen's Buildings, School of Computer Science and Informatics.
Date: 29 November 2017 at 14:00

The Chair for this seminar will be Yukun Lai.

In recent years, we have seen significant progress in the development of deep learning techniques. Consequently, different groups in the School are using deep learning techniques to solve research problems, which can vary significantly.

Talk 1 - Malware prediction using Recurrent Neural Networks (RNNs) with Matilda Rhode

Recurrent Neural Networks (RNNs) are the only machine learning algorithm capable of processing highly varied, numeric, time-series data. We have used RNNs to try and detect malware based on the initial sequence of behaviours exhibited by a file.

To achieve higher accuracy earlier in the file execution. we select a group of networks, which have learned to place emphasis on different features, and use these to create an ensemble classifier.

Talk 2 - Deep learning tools with Humphrey Sheil

We will talk about deep learning tools.

Talk 3 - A prototype systems architecture for Coalition Situational Understanding (CSU) with Dan Harborne

We will demonstrate how services from across a coalition can be brought together to facilitate low-to-high level reasoning and how this compares to an end-to-end deep learning approach.

We will also showcase the affordances of each method with regard to the core factors of Coalition Situational Understanding (CSU).

In this work specifically, we focus on examples of machine decision interpretability and the impacts of information flow constraints (and the links between these two factors).

Talk 4 - Deep learning applications in visual computing with Tom Hartley

Deep learning has found applications in many areas of visual computing. In this talk, we will cover some of the basic tools we use before looking at how deep learning is used in visual computing.

This will include the use of two stream networks and 3D convolutions in action recognition, techniques for visualising networks to ensure trust, and training networks with small amounts of data.