Welcome to the fast.ai/UQ hackathon thread!
Many of you will naturally have questions, like… How will I find a team? What project should I work on?
The purpose of this thread is to help you understand a little better how we will run the event, and help to kick-start that project and team-building process.
Kicking off next Tuesday afternoon (Jun 28th) at 1pm we’ll have facilitated team building and project-finding exercises, so if you don’t know anybody or don’t have any idea what you want to work on, that’s just fine!
However, if you’ve got an idea (or many ideas!) for a project - let’s hear about them in this thread. If you see somebody propose a project that you’d like to dive into, then start that conversation.
Our judges are really looking to see your application of the ML and AI techniques taught in the course. We are also interested in potential impact of the solution, whether that’s commercial, social or something else entirely. It will also be great to see insightful analysis of things like feature importance which Jeremy has covered at some length.
When thinking about problems to tackle, be mindful of considerations such as
- what is the potential impact of the solution?
- is it feasible to demo or prototype in just a couple of days?
- what data sources are available to train with?
- what technical skills would contribute to developing a good demo or prototype?
- what can you contribute to your group?
We are anticipating 45-50 participants, and aim to form groups of 4-5 people at the most. So, while it’s great to have an idea or several ideas you are keen to tackle, try not to be too wedded to any one concept.
Feel free to @ me (@jwuq) with any specific questions and I’ll do my best to help or clarify.
Democratising eucalypt species identification
I am not sure whether this is an idea suitable for a hackathon yet, but will outline it just in case, even if the post is to be moved to another thread (which is fine). I suspect that a compendium of labelled photos is not yet readily available to train on for a 3-day ML hackathon.
I came across a twitter post from a prominent Australian journalist who is also a keen mountain bushwalker, asking for help to identify eucalypt leaves. This reminded me of Jeremy’s use of rice species identification recently in his walkthroughs.
That prompted me to have a quick look at what is available for eucalypts which are a rather distinctive genus but hundreds of species.
The main reference resource I located is EUCLID, which appears to rely so far purely on human brains for classification.
I also know of iNaturalist for plant identification, but a quick look at a common species such as Eucalyptus blakelyi says that “The current Computer Vision Model does not know about this taxon, so while it might be included in automated suggestions with the “Nearby” label, it will not have the “Visually Similar” label.”.
I cannot help but think “surely it’s been already looked at”. Botany is not my field so I have many and big blind spots.
I thought there was an adaption of the LUCID keys that allowed ID from image. I remember it being demoed to our bushcare group. Not sure what became of it. Possibly a project from Brisbane Herbarium
I was just wondering if all team members need to register separately through the registration link or one team member can fill in their details on behalf of the team. Will there be a different follow up link sent out to fill up details for all the other members?
Is there a “looking for team” thread for those of us without a team?
Feel free to use this thread.
All participants in the hackathon need to register for catering purposes and also to meet the legal requirements around insurance, prizemoney and so on.
Team formation will take place once everybody is registered and at the event.
I’ve just made an edit to the thread description that hopefully clarifies things a little. As Jeremy said, feel free to use this thread. If you’ve got any project concepts, no matter how vague, feel free to throw them out here. You never know, somebody else may reply and want to work on it!
Here’s a text / natural language processing project idea from me. THe US National Institute of Standards & Technology publishes something called the NICE framework - National Initiative on Cybersecurity Education.
It’s a massive table of cyber security related skills, knowledge and abilities, and a corresponding mapping of job roles in different aspects of cyber security - e.g. Secure Provisioning, Oversee and Govern, Operate and Maintain and so on.
The data is here (Excel Spreadsheet format): The Workforce Framework for Cybersecurity (NICE Framework) | NIST
The idea would be to train some sort of classifier on this pre-labelled data, then use it e.g. given a passage of text describing a work role, which NICE category does it best fit?
Or, given some text describing a course or learning module, which NICE skills or knowledge is it most closely aligned with?
This is a hand curated corpus, but it’s quite imbalanced and not a particularly large dataset. Some interesting challenges for sure.
Note that I’m a judge so I can’t get involved in your team sorry!
Another project idea - invasive weed species identification. Research the top invasive weed species according to state or fed government department of natural resources. Train a classifier to identify them. If you have the time/skills, prototype a mobile web app that allows image upload and classification. Add geotagging!
Given @jwuq said our project concepts can be vague … I am interested in building systems that identify people who do not have an advanced care directive and may need one. Only 14% of Australian have an advanced care directive in place. Advanced Care Directives state what healthcare treatments someone wants or does not want, in the event they are seriously ill/injured and can’t communicate decisions about care themselves.
I wondered whether we could use one or more datasets from physionet to build a system that: 1. prioritises patients for proactive engagement to complete an advanced care directive (i.e. patients at highest risk of serious injury/illness in the near term) and 2. identifies the health care treatments that will be most important for those patients to consider in an advanced care directive.
I do need to disclose two matters along with this project concept: 1. My mum is a doctor with a small business that allows patients to complete digitised advanced care directives; 2. Unfortunately, I can’t attend the final day of the Hackathon. If one or both of those matters preclude this project concept from being considered for the Hackathon, I totally understand. Thought I would share, nonetheless.
My name is Tyson Jennings and I am your facilitator for next week, I am excited to meet you all next week and see the amazing Ideas you come up with!
As John @jwuq has stated it’s time to get your creative juices flowing. Over the weekend go about your everyday lives and start to look for problems that could be solved with AI or ML. These could be little annoyances with navigating Bunnings (I will be there this weekend) or something you see on the news or on social media that tugs on your heart strings. The best way to develop a great solution is to be solving a real problem for as many people as possible, that you personally care about. I love some of the examples I have seen already but keep them coming.
For those of you that can’t think of a problem, that is perfectly alright! You have skills that will contribute to your team being super successful! Think of what you are really good at whether it is public speaking, product development, data analysis, research skills, marketing the options are endless!
Have a great weekend and I look forward to seeing you all on Tuesday!
Hello @Tyson and @jwuq , I’m attending the Hackathon tomorrow and Wed. Where are we meeting at 1pm and can you recommend anywhere non-students can park for a reasonable day rate?
Hi @Mattr ,
Your best bet would be Blue Zone casual parking at $5/day - we’re in mid year teaching break at the moment so parking should be pretty easy to come by.
Lots of info here including maps and payment details: Casual parking at St Lucia - Campuses - University of Queensland
Your best option for parking nearby the venue would be Conifer Knoll Carpark (access via Sir William McGregor Drive), or even the casual parking along Sir William MacGregor isn’t far either.
Hi John, it’s still a bit unclear where the location for the Hackathon is?
Is this information supposed to be in the Slack channel we were supposed to be invited to after we registered?
Hi Charlie, I have sent the information to your email and the slack invite should of come through, please let me know if it hasn’t.