Passing my A2 levels for Mandarin and moving on from the INFLOW project.

Wrapping up INFLOW

Another presentation this month on the INFLOW-AI model, this time at the Red Cross “Long Night of Research” – a somewhat strange initiative where presentations run for 24 hours through the night. My presentation was at 7.00AM BST, which was pretty unfavourable to both American and European audiences. Nonetheless, attendance and participation was quite high (about 20 attending, with one Red Cross employee specifically emailing me afterwards to thank me for the work and presentation).

One of the core improvements to the model designed to improve predictions on out-of-sample extremes by seasonally differencing targets.

Seasonally differencing targets forces the out-of-sample extreme distribution to more closely resemble the typical case distribution, the benefit is even higher when predicting recursively (by summing predicted deltas).

I also finally wrapped up writing the draft of the Geoscientific Model Development paper for INFLOW-AI. With that behind me (after almost an entire year of this being my primary project), I’m ready to move onto other things and work with student volunteers again.

One last major update to the model before I let it go more permanently: creating a trigger mechanism to alert humanitarian agencies when the forecasted flood is predicted to exceed a given threshold change from the dry season of that year.

Student Projects

Finally ready to try and scale up some work again by working with students. I intend to run at least two student projects this summer, and possibly more in the Fall term. It all depends on which collaborations I can get. I attended a lecture at the Martin School on what turns spillovers into pandemics in search of potential academics interested in our AMR surveillance tool LexiMine.. I’d love to run a project that build on what we did last year, to actually test the tool in a real-world project.

I’ve also been meeting with CIGI, who have promised to connect me to people at the Canadian Department of National Defence with regards to building a model to predict global conflicts, similar to ACLED’s Conflict Alert System (CAST). Will know more about this later.

I’m working with a friend back in Chicago to set up a GitHub repo structure to better facilitate data cleaning for student projects, since my biggest barrier to working with students last year was a lack of consistency in their work. Hoping to fix this by introducing modularity and pre-written tests they can run to integrate their components of the code with the overall pipeline. I plan to send out research position ads next month, and will be running projects from Trajan House again.

Other Work

Taking some time over the next few months to volunteer as an expert in the UN working group FAIR Cities (Foster AI for Inclusive and Responsible cities). I’m not sure how impactful this working group is, but I’m interested to learn more about how the UN works and contribute to areas in which I know some things. Thus far my experience has been fully underwhelming; it’s clear the group is not going to produce any document at the level of specificity needed to be actually useful for AI practitioners at the municipal level.

I’m doing my best to recommend the group focus on specific issues with specific applications of AI in cities by breaking down potential AI uses into different typologies, but it’s not seeming to stick. Everyone wants to just sombrely note that we need to consider human rights and ethics, which is great, but not actual useful advice for a document allegedly going to be used by AI practitioners to inform decisions about fair implementation of AI in cities. (Does anyone actually consider theories of change before they create a UN working group? Who is this for?)

My “AI Typology for Cities” I keep pushing at the group so we can get beyond the level of just writing the words “human rights” in the chat dozens of times. An aspirational goal, but more specifics are needed to actually achieve any benefit.

For example, in the category of “predicting future events or trends”, subcategory “disaster risk modelling”: we could be talking about how urban data usually does not include representations of where unhoused people live, meaning disaster alerts might not reach them or might underestimate potential impacts of various hazards, harming resource allocation. This could be fixed by working with local law enforcement to maintain a relatively up-to-date layer on large encampments and areas frequented by unhoused people. (See? Specific example, AI fairness issue that practitioners might not consider, and achievable solution.) Yet, I’d be willing to bet any money that the report is going to come out with mostly fluff like “AI must respect human rights :)”.

If I get my way I’d like to try and make the report more specific.

I’m also working on getting funding for my DPhil. Not strictly useful to anyone besides me right now, but (in my opinion) determinative to whether or not I will accept my offer. Time is running out, as the deadline is the end of June. I have a fairly strong lead with the Barrie Rotary Foundation – will update on this later.

Life on the Side

Passed my A2 in Mandarin! I hope my extra credit contributions from my viva video played a role. Also saw a student production of Fiddler on the Roof at Jesus College which was very impressive.

Queer Formal at Mansfield.

I’ve also started life drawing at Mansfield (free lessons offered by a former law student, now in art school).

Looks a bit strange at the moment but I’m sure I’ll improve.

Have you ever seen Attack on Titan?.

I attended the Oxford Summer Eights for the first time, where the most exciting event for me was learning College heraldry.

For example, St. Catherine’s has a wheel as their symbol as Emperor Maxentius tortured St. Catherine on a torture wheel (however the wheel allegedly broke and she ended up beheaded instead).

At least the event provided a lovely opportunity for some water colouring.

Oh, and I guess I’ve aged again – making this an entire year since I wrote my first post on this site. Has enough changed?

my_age += 1.

Chill celebration at Kazbar with the flatmates.