ST 311: Group Project Instruction
Due by 5pm, 5 May, 2023
1 General instruction
• Your project should demonstrate that you have mastered some of artificial intelligence
(AI) topics covered in this course, including, e.g., multi-layer perceptron (MLP), deep
neural networks (DNN), convolutional neural network (CNN), recurrent neural net-
works (RNN), generative adversarial networks (GAN), (deep) reinforcement learning
with a focus on methodology, architecture design, implementation, training, and eval-
uation of the performance of a deep learning or reinforcement learning model. In addi-
tional to various tasks of image and text analysis covered in this course, your project
can apply deep learning methods and its variants/extensions to other AI tasks, e.g.
computer vision, video and audio analysis, natural language processing and machine
translation.
• You can find the dataset from any source. If your project contains deep learning
components, the corresponding implementation must be in PyTorch.
• The project is group-based, typically, a group would consist of two students. You are
expected to form groups among yourself following a separate email instruction and
submit a project proposal (at most one page). Your project proposal must be sent to
the course lecturer,
– Xinghao Qiao (x.qiao@lse.ac.uk),
for approval no later than 10 April, 2023.
• You are expected to split the work on your project among yourself. It is expected from
each group member to make a fair share of technical contributions to the project.
• Your report would typically be in the form of a Jupyter notebook containing Python
code with explanations, along with a Markdown text explaining different parts if
needed. You may also want to write your report in pdf format, which would give
more flexibility in formatting of your project report – this will be in addition to your
code source files (e.g., in a Jupyter notebook).
• You are expected to define and explain concepts used in your project to demonstrate
you understand them.
1
• It is expected from your report to be written in a high professional standard, which
means that is has to be well structured, neat and polished. The report should consist
of a title, abstract, introduction, methodology, numerical evaluation using real data
example, a conclusion section and a reference section, following the standard for aca-
demic papers. In the abstract and more detailed introduction section, please clearly
describe what is the problem studied in your report, why is the problem non-trivial
and interesting, what is your proposed solution and what are the results and findings.
Your report must cite any references that you need. You may use visualizations in
your report. You may also discuss and cite any previously-proposed solutions to your
problem, and compare the empirical performance of your solution with that of other so-
lutions used as baselines in the section of numerical evaluation. The conclusion section
should briefly summarise the main results and findings, and briefly discuss interesting
directions for future research.
• At the end of the report, you must add a section “ Statement about individual contri-
butions”, in which you need to provide the percentage of contribution of each group
member and summarise individual technical contributions.
• The grade for each group member will be a function of the contribution of each group
member using the following equation.
member grade = report grade × member contribution
maximum contribution
.
For example, for a group with 2 members contributing, 60%, 40% and the report grade
is 75 (out of 100), the individual grades are
75× 60
60
= 75, 75× 40
60
= 50.
2 Possible project topics
In Section 2, we provide some possible project topics from last year and other similar courses,
from which you may draw some inspiration. You are not expected to choose a project topic
from these suggestions. In fact we encourage you to come up with some original project
topic idea.
• COVID-19 detection using deep learning techniques.
• Option pricing using deep learning.
• Detecting Cyberbullying using Neural Networks for Text Classification.
• Sentiment classification of Covid-19 related posts on Twitter.
2
• An application of the novel DeepCascade-WR for forecasting volatile time series.
• Classify similar products into groups using product images and titles.
• Colorizing Greyscale Images with Neural Networks.
• Deep Learning for Natural Language Processing: Large Movie Review Sentiment Anal-
ysis using various Deep Neural Networks.
• LSTM with Emotional Analysis for Stock Price Prediction.
• Vehicle Detection and Classification using Yolov4 and SSD.
• Image classification for diagnosing Covid-19 & Pneumonia by Chest X-ray.
• Anomaly detection in images with object localization.
• Conversational ChatBot using deep learning.
• Detecting sentiment using LSTMs and CNNs.
• Abstract text summarization using RNNs.
• Detection of knee injuries using deep learning techniques.
• Diabetic retinopathy detection.
• Generation of financial time series using GANs.
• Analyzing sentiment surrounding Russia’s Ukraine invasion Using BERT.
• Predicting poverty level from satellite imagery using deep neural networks.
• Predicting Meta stock prices using LSTM and DNN.
• Using neural networks to predict the Myers-Briggs Personality Type (MBTI) of users
from their social media posts.
• Identifying Pneumonia by image-based deep learning through X-rays.
• Heart anomaly detection.
• Image classification for diagnosing pneumonia by Chest X-ray images.
• Multi-class classification of news category with BERT.
• Music genre recognition using neural networks.
• Predicting affective content in tweets with deep attentive RNNs.
• Prediction of meterological data using LSTMs and LSTM-CNNs.
3
• Sentiment analysis on Amazon fine food reviews dataset.
• Smart indexing using autoencoders.
• Stock prediction with recurrent neural networks.
• The use of deep neural networks to predict NBA game outcomes.
• Using behavioral patterns in recommender systems.
• Using knowledge distillation to increase accuracy of lightweight CNNs for image clas-
sification.
• Variable selection using deep neural networks.
• Classification of pigmented skin lesions.
• Densely connected convolutional neural network for mammography and invasive ductal
carcinoma histology.
• Exploring image classification techniques to predict poverty levels.
• Music generation with artificial intelligence-creative sequence modelling using LSTM
and recurrent neural networks.
• Measuring political preference with sparse text classification methods.
• Predicting hurricane trajectories using RNN.
• Applying Reinforcement Learning to explore optimal solutions for the container loading
problem (CLP).
• Quantitative trading by reinforcement learning.
• Deep direct recurrent reinforcement learning for algorithmic trading.
• Reinforcement learning for trade execution with Alpha and risk aversion.
• Solving ATT48 by deep reinforcement learning.