ML internship, Masason, habitat conservation, & more math again May/June/July 2021

Slowly this newsletter transitions from bi-monthly to quarterly

Hi Everyone,

Clearly I have forgotten to update this. Well, I haven’t forgotten per se but I’ve pushed it back to the bottom of the to-do list at the end of each month for well, about four months now.

But, hey, at least I’m not spamming, right? I’ve clearly been taking a rule out of Supreme’s book in terms of posting. Lower the supply and up goes the demand, right?

Well, better late than never. Here’s why I’ve been too busy to do all of this. Maybe I’ll try to keep things back to a bi-monthly schedule. Who knows?

Machine Learning Engineer at Quantiphi

I’ve been fortunate enough to be able to work as a Machine Learning Engineer intern at the wonderful company Quantiphi. The learning curve has been steep but it has felt rather linear with the constant support and help of my mentor, Darien Schettler (I feel like every time I mention his name it is accompanied with around 3 or 4 “thank yous”, but seriously, I could not do it without him).

At Quantiphi I am working on a project to detect Covid-19 from chest radiographs, as well as locate regions of interest in the lungs. The current norms of using PCR and sometimes CT scans to detect Covid-19 are surefire but slow. X-ray imaging takes minutes, and if a reliable algorithm was created to detect Covid-19 from these images, patients could be diagnosed and given proper help quicker.

I have been tasked with creating the data pipeline and image classification part of the task. Other than learning Tensorflow (as a former Pytorch user), I have read quite a few papers now to learn about the algorithms I’m implementing (EfficientNet, FPNs, ResNet, MobileNet, Squeeze-and-Excite Blocks, etc…).

The dataset is quite tough to classify, and these are some of the predictions of my highest scoring model so far (roughly 0.67 accuracy on validation, which in the context of the challenge is quite good). You can’t really complain about the bottom right image, haha.

I have a little over two weeks left at Quantiphi and but I’m trying to work four weeks worth. I’ll update you in two or three or six months on how that went.

Associate Member of Masason Foundation

I’ve been privileged enough to be selected as one of the 31 youth globally to become a fellow of the Masason Foundation, an initiative started by Softbank founder Masayoshi Son about half a decade ago to empower youth with funding and a community of like-minded individuals.

I’m rather excited to meet people leagues smarter than me.

Learning Tensorflow

This is a bit tied to my internship, but hey, why not make it an extra thing. I’m spending a lot of time familiarizing myself with Tensorflow, which I actually so far like a lot more than Pytorch. It’s not as cluttered as I’ve been told, I think, or perhaps that’s just because I’m only starting out.

Instantiating and modifying the pre-made EfficientNetB0. As it is often with ML, creating the model is pretty much the easiest part - what’s that, like six lines?

Anyways, I’m getting some good practice with the tf.data module, creating and using pre-made models (like EfficientNet and ResNet) from the tf.keras module. I’m looking forward to pretty much only dealing with tensorflow from now on.

Creating an Online Course for Deep Learning

I’m currently putting together the first online course to go with the textbook Deep Learning by Ian Goodfellow, Yoshua Bengio, and Aaron Courville. It’s a bit of a slow burner since I am pretty busy this summer, but I try to at least film a few lectures each week. Along with lectures, I’m also creating quizzes and articles. I’m currently in touch with one of the three authors, and things might happen if I am able to finish it.

getting very animated talking about positive semidefinite matrices

Also, a quick sneak peek at the course hub, because why not? Writing about this is inspiring me to get back at it again (after a rather stressful time trying to explain marginal probability in a way that is not completely and utterly confusing)

An article that I wrote explaing Principal Components Analysis for the course in the first chapter, Linear Algebra (although I’m probably going to move it)

I’ll keep you updated on this one, maybe. I’m looking forward to seeing it perhaps get somewhere, and hopefully I’ll have enough time to get around to finishing it.

Learning Pure Maths (Real Analysis)

Despite all this machine learning stuff, the more and more I do all of this, the more I realize that my true love might be mathematics. To test this hypothesis, I’m beginning to self study my first “pure mathematics” subject: Real Analysis.

In other words, exploring the peculiarities of the real number system and re-defining calculus in a more mathematically rigorous manner.

So far, I’m loving it, and I think I might become a mathematician*.

A screengrab of the textbook I’m using (S. Abbott, Understanding Analysis). This is one of my favorite-est concepts I’ve gotten to so far, Cantor’s Diagonilization Theorem. It’s incredibly simple yet an absurdly creative way to prove things relating to uncountable sets.

I’m really trying to spend every spare moment doing this stuff.

* Maybe. I love this stuff, I really do.

Land Conservation Project (wow, that’s different!)

Totally different from all this stuff is another side project I’m working on, which is trying to conserve a local plot of land in my city (Coquitlam) that is planned to be developed in the next few years. It’s an uphill battle, but I’ve made some strides, and either way, if I’m not able to protect the land, I can make sure the negative effects are mitigated and that the city compensates in other ways.

A family of Northern Flickers I photographed at the lot a few weeks back. The unparalleled amount of snags in the lot gives great opportunity for nesting woodpeckers (as I detail in the report!)

This little plot of land has incredible breeding bird diversity, as well as serving as an incredibly important corridor for wildlife moving through the city, especially migratory birds. I met with a local naturalist to do a bat survey in the lot, and found regular transit of both Big Brown Bats (E. fuscus) and Hoary Bats (L. cinereus) over the lot, yet another reason to protect the land.

I wrote a three page paper summarizing the ecological significance of the land, followed with a short proposal. I have quite the plan for this, but it’s also sort of a slow burn as things will mostly heat up during municipal elections a year from now. I’m currently slowly putting together a documentary of some of the footage I was able to capture of nesting birds and animals.

---

Thanks for reading, folks. See you in a few months, I need to get back to work (actually, I need to sleep. It is 1:21 am as of writing).

don’t forget to subscribe or whatever

cheers, adam