Artificial Intelligence Coursework Guideline
This order does not have tags, yet.
Artificial Intelligence Coursework Guideline
This coursework concerns the automated classification of images through machine learning techniques.
You will work on blood cell image data, where training samples and their ground truth are provided.
You will develop suitable classification techniques to classify unseen examples, a poster to introduce
the problem. You will need to submit a Jupyter notebook containing your code for training and
evaluation, your prediction outcome of the test dataset, and a A1 poster by 9th of May. Any of your
experiment figures and/or tables included in your poster need to be reproducible in the Jupyter
notebook that you submit.
You are provided with the datasets contained in AI Data.zip.
• train: a folder containing 10,000 training images.
• train.txt: a txt file containing 10,000 ground truth labels of the training images.
• test: a folder containing 5,000 testing images.
These are datasets about blood cell image, and there are 8 classes including healthy and different
types of diseases in this dataset need to be classified. The data were extracted from a microscopic
peripheral blood cell image sets and each image was revised into 28 x 28 x 3 images. More
information about the original dataset can be found in .
You should submit the following three files:
1. A Jupyter notebook containing code to (i) train and evaluate your developed algorithms with
training data, (ii) save your developed model, (iii) load your developed model, (iv) perform
evaluation on a test dataset that generate the test.txt described below. The submitted Jupyter
notebook needs to include the figures for evaluating performance which you included in your
2. A text file called test.txt containing the predicted labels of the test dataset, following the format
3. A poster that describes the problem, literature review, your approach and evaluation result.
4 Coursework poster
The poster should be in A1 size, in digital format (No hard print is needed). It is important to
demonstrate your understanding of the subject area, your approach to the coursework and present
the outcome. There are many useful guidelines on how to design an academic poster, for examples:
https://www.makesigns.com/tutorials/. You are also provided with a sample poster as a research
presentation poster example. You are free to choose either the landscape or portrait orientation.
You should consider including sufficient evidence of your work through the poster, covering the
following areas in the marking criteria.
5 Marking criteria
You will be evaluated based on the following criteria with your poster and submitted.
• Introduction to the project: providing appropriate background information and detailing
the objectives of the project. Literature review should be carried out on relevant topics and
how they may relate to your work. You are expected to research and discuss other sources of
information but must show their origins by referencing all sources used. The reference list
should be included in the poster. (10/100)
• Analysis of the problems: research and technical issues, challenges, and description of the
approach you have developed with justification. Effective graphic illustration will be
appreciated. Multiple classification techniques can be designed and implemented. It could be
also multiple versions of the same technique, with different hyperparameters or different ways
of training approaches, so you can compare the performance across different models in the
• Implementation and evaluation of the methods: You need to discuss the system results,
present the data you used, how many of them for training and validation in your model, and
what are suitable performance metrics etc. (hint: you may want to include not only accuracy
here as your performance metrics). Figures and/or tables can be useful to present such
information, especially when multiple approaches are developed, and comparison of various
approaches/choices are conducted. Figures included in the poster need to be reproducible as
demonstrated in your Jupyter notebook (otherwise they will not be counted in the scores).
• Performance of the model: The marking is based on the performance of the final prediction
result of the test dataset. Your final submitted test.txt will be compared with the ground truth
label (withheld from you) by the Module team. If students have achieved the baseline result
(accuracy on the test set ≥ 90%), 30 marks out of 30 will be awarded. If your model achieves
less than 90%, your mark will be determined by your accuracy score, i.e.:
You must submit the Deliverable 1-3 in Section 3 as evidence of your work for these marks. (30/100)
 Andrea Acevedo, Anna Merino, Santiago Alf´erez, Ángel Molina, Laura Boldú, and Jos´e Rodellar.
A dataset of microscopic peripheral blood cell images for development of automatic recognition
systems. Data in Brief, 30, 2020.