MLP 2017-18  |  News Archive  |  Lectures  |  Labs  |  Group Project  |  Coursework  |  Feedback  |  Computing  |  Piazza  |  Github

Machine Learning Practical (MLP) 2017-18: Group Project


MLP in semester 2 will be based on projects done in groups of 2-3 students. The projects, which should be done using one of the deep learning toolkits (TensorFlow is recommended) can be chosen from a variety of topics and data sets.

Project Groups

The projects are intended to be done in groups of 2-3 students. By working in a small group you can discuss ideas and work things out together. You can form your own groups. You can use the Piazza 'Search for Teammates' to help you form a group, if you like.

You may discuss any aspects of the assignment with your group and divide up the tasks however you wish; but we encourage you to collaborate on each part rather than doing a strict division of tasks, as this will enable better learning for all of you.

Register your group on this Google Sheet

The Projects

Data sets that you might wish to explore include CIFAR-10/100 object recognition, the Million Song Database (or a subset of it) for tasks like music genre recognition, Painter-by-numbers to predict if images of two paintings are by the same artist, large movie review dataset for sentiment analysis, ... If you have a suitable data set and task then you are free to Bring Your Own Data (BYOD).

Approaches you might want to explore include multitask learning, curriculum learning, one-shot learning, generative models, Bayesian deep learning, meta-learning....

In some projects the entry point to the project might be an interesting data set / task, and you may focus on engineering fairly standard approaches to work well. In other projects you may look at a challenging approach, in which case it could make sense to work on dataset you already understand and have good baselines for. The choice is yours, both types of project are valid, and you can get excellent marks on both types.

So the first part of your project will be to make a plan - what data you will be using, what approaches you will investigate, what are the research questions. You will then carry out your plan!


Each group will set up a google doc which is to report progress (including results), give plans, and raise any questions they have, updated each week. Each group will have a tutor, who will review the group's google document. There will be tutor sessions each week, in which groups can meet with their tutor. Tutorial sessions will involve 5-6 groups: an advantage of sessions involving with several groups is that many of the questions/discussions will be similar across groups, and there is of course opportunity for discussion between everyone in the session. Since the tutor will have read the group reports, there will be less need for groups to report on their progress in detail in the meeting, and more opportunity for discussion.


There will be a daily helpdesk, starting next Monday. The helpdesk will run Monday-Friday from 2-3pm in AT-5.08 South Lab.

There will not be weekly lab sessions this semester; they are replaced by tutor sessions and the help desk.


We are planning on making a new GPU system (with a reasonable number of GPUs) available for use, hopefully from next week. I hope this will provide reasonable compute for the sort of projects that I know people want to do. More details on that next week.


Please make sure you have read and understood the academic misconduct policy.

Copyright (c) University of Edinburgh 2015-2018
The MLP course material is licensed under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International Public License
The software at is licensed under the Modified BSD License.
This page maintained by Steve Renals.
Last updated: 2018/02/14 14:04:41UTC

Home : Teaching : Courses : Mlp 

Informatics Forum, 10 Crichton Street, Edinburgh, EH8 9AB, Scotland, UK
Tel: +44 131 651 5661, Fax: +44 131 651 1426, E-mail:
Please contact our webadmin with any comments or corrections. Logging and Cookies
Unless explicitly stated otherwise, all material is copyright © The University of Edinburgh