Academic year: 2021/2022
My interests are in data analytics, computer vision (images or video data) or machine learning applications in general. I've worked on projects involving instrumented vehicles ("self-driving" cars); smart cities (how crowds move); security and medical image data.
MCM (CA685 - Data Analytics)
Please consider the following questions before approaching me to discuss a project.
- What is your topic, problem or domain?
- What datasets might you use? (how big?, is it labelled?, is it ethical?)
- How will you know if you have solved or contributed to the problem? (or how will you measure the performance of your methods? how will you test your implementation?)
Frequently Asked Questions
- Will you supervise my project? Maybe. Projects that I'm less likely to supervise (and there are people with more experience) - text analytics or natural language processing; games; cloud systems.
- What data should I use? It should be large enough to test and train systems. It should be labelled (eg. with the objects or actions you're interested in). You should have permission to use it.
- I want to do deep learning. Can you give me a GPU? No (sorry). There are a few systems but I don't have general access to GPUs. Students should look at options from Google Colab; amazon web services and labs at DCU. Note that these systems are not enough to train large video datasets (insufficient RAM or process timeout constraints). It takes time to learn, implement, train and test CNN/Deep learning systems so consider carefully if this is what your project depends upon.
Computer Vision Challenge: Use a labelled dataset from one of the computer vision challenges or these datasets to explore a problem in image analysis (e.g., Detection, Classification, Recognition , Tracking, Segmentation , Foreground/Background, Saliency detection, video Surveillance, Multi view, Action, Human Pose/Expression). Be aware that there's not much time in the project to use a very large dataset!
Another source of computer vision challenges is the workshops associated with the Computer Vision and Pattern Recognition workshops - skin imaging, UAV detection, fashion, autonomous driving, etc.
These projects shouldn't just apply a model to the dataset but should try to make a change to either the data pre-processing (augmentation?), to the training methodologies (hyper-parameter optimisation is insufficient) or the network architecture. You should be trying to improve performance or to ask a question about how a technology works in this context.
Phone: ext. 6052