Machine Learning Saved My Life, and it’s Personal

Yim Register
Nov 1 · 12 min read
Scrabble pieces on a pink background spelling out “Love Yourself”

Major Content Warning: the following post contains some details about extreme sexual violence, child abuse, disability struggles, trauma recovery, and an odd amount of optimism surrounding machine learning and statistics.

Today was a big breakthrough in therapy and I’m inspired to write it out in a public blog post. I’ve been sitting on these ideas for a while, but they’re not pretty and not easily digestible. Some of this public sharing is solely for my own healing; because this is not my shame to carry on my own. Another part of this is honing in on why I’m studying what I’m studying in my PhD program. I study if/how personal connection to machine learning algorithms can help us learn and advocate for ourselves. After today, I know why I do this. Lastly, I share this because my connection to ML and statistics has helped me to explore vast corners of myself; I introspect with algorithmic metaphors and observe the world like a complex network where we can affect change for the better. Just because my story is traumatic and sad doesn’t mean that you shouldn’t contest the idea that Machine Learning is an amazing introspective tool (because in many cases it is not). It just means that maybe you’ll go on a short 5 minute journey with me about how statistics can help us heal.

Today in therapy I realized how machine learning saved my life. I’ve hinted at it before, in almost everything I talk about. But we got to the root of it today. Even as I’m writing this, I’m saying to myself “dear god you do not have to publish this, and you shouldn’t!” So let’s see if I click publish at the end. I’m also compulsively eating popcorn right now, apparently that’s a coping mechanism (lol).

When I was 12 years old I was in a horrifically abusive relationship. It lasted for two years, and in that time, my childhood was stripped away from me in every way you could imagine. Some of the most horrific moments including being locked in a closet all night without any clothes after being raped and nearly being forced to piss in a bottle. Yes, it’s fucking criminal, No, I don’t feel shame anymore. I’m shaking as I write this but I promise it’s all related. If he didn’t like the clothes I was wearing, he would tear them in half and spit on me. I once made him a plaster gift of my hand and he crushed each finger off and laughed. I wasn’t allowed to eat Italian Dressing (that one got me really laughing in therapy. Yes, this stuff is heavy. But dear lord what a stupid situation). Every day for two years I met him at his bus stop to be taken home and forced into sex, criticized and critiqued in front of a mirror, and sent home crying. He called me a dog and a slave for two years, and I hid it from the world. You might ask how I finally got out? Well, a lot of therapy for one. Even recognizing it was abusive was not immediately obvious to me. The final straw was that I wanted to play rugby at my highschool. It seemed fun, and powerful. He told me I wasn’t allowed to, as it was too manly (jokes on you! I’m trans! woop!). Somehow, that was the last straw. Because I really wanted to play. I played for 8 years, and if I do say so myself, kicked some serious ass.

So how does something like this happen? How does something like this happen to a child? What on God’s good Earth made me believe that it was okay to be treated this way?

My foundations were rocky at best. My models of how humans should interact were incredibly faulty. I am autistic, and I can describe my state of mind as basically starting out like a blank slate. My theory of mind used to suffer a lot more than it does now (now I have pretty decent insight from all of the work I’ve done). As a child, I knew that the world hurt. Things were way too loud, I wasn’t allowed to spin or stim because it was distracting to the class, and teachers yelled at me a lot for getting overwhelmed. My home situation was incredibly rocky, and I genuinely feel bad about going into it so publicly despite there being major pitfalls in how I was raised. I was undiagnosed as a child, with most teachers and family members sensing that something was different but not really knowing what. On top of that, I identified as a boy and no one knew how to listen to that. Puberty was trauma for us all, but it was especially difficult for someone who loudly spoke about being a boy and being laughed at. If I wasn’t a boy, then I must be a girl. And girls are sexual creatures used for pleasure by boys. Somehow, that was my view of the world. I had a lot of meltdowns as a child. If you’ve never heard of an autistic meltdown, it’s not a temper tantrum. Temper tantrums are usually a big fuss to get attention or a need met (which is also pretty reasonable to be honest). Meltdowns are when the body pretty much goes into shock. The senses are overwhelmed, it’s too exhausting to live in a neurotypical world, and the body needs to dispel some of the boiling energy inside. Sometimes autistic kids take off running, like I would in large, crowded spaces. They call it “elopement”. The response to my elopement? People would strap me down, restrain me, slap me, and scream at me. I’ve even gotten the police called on me several times, putting me in handcuffs or strapping me into an ambulance and punishing me for my meltdowns. I turned to drugs for a while, drinking for a while, and an eating disorder for a while. What else is a kid supposed to do? In college, I was sexually assaulted numerous times because it was just how things were supposed to go. Unconscious and alone, trapped in another room, my patterns were deeply rooted models of how I thought I should be treated.

Most of that was before I ever even hit 15 years old. Today in therapy, I asked the question “how did it get so bad? How did I let it get so bad?”

It still hurts, and I’m grateful every day for good trauma therapy. But even more than that, I am grateful to machine learning and statistics.

Rational me doesn’t blame me in the slightest. I have so much compassion for the little Yim who went through all of that so that I could be who I am now. I think they were god damn brilliant. A shining star that the world didn’t know what to do with. It still hurts, and I’m grateful every day for good trauma therapy. But even more than that, I am grateful to machine learning and statistics.

It sounds ridiculous. I know. But I’m going to try to communicate what started happening for me when I started gaining data science knowledge. First came a math class with Patrick Cooney, I must have been about 16. His love of math was contagious. He was funny, and brilliant, and obnoxiously in love with math. Patrick Cooney was a man who dressed up as a “Rectangle” for Halloween by being an “angle” (arrows on his hands) and a car “wreck” through his middle. A “wrecked” “angle”. A “rectangle”. What a ridiculously lovable person. Anyways. He adapted to students and how they were learning. He put in so much effort into his classes; desperately wanting to share some of that joy. And I dove in. Numbers had an answer. Numbers only had joy for me.

I did a project where we had to collect data from our own class and perform the chi-squared test by hand. I had some weird theory at the time about gender and handwriting size. I literally collected hundreds of people’s handwriting, printing out the phrase “I love pina coladas and getting caught in the rain”, and measured the letters with a ruler. I have no idea where that theory came from but it was one of my first explorations into gender and statistics (it was dumb, there was no significance, and I was 16). He introduced me to fractals, a concept I fell in love with for years to come. (Did you know that the B in Benoit B. Mandelbrot stands for Benoit B. Mandelbrot?)

blue, yellow and white image of the ever-recursing geometric mandelbrot set that goes on and on until infinity (or until you run out of pixels).

Next, I produced an award-winning “Extended Essay” (for the International Baccalaureate Program) about the history of the Golden Ratio in Egypt and Greece. Numbers run deep, back to hieroglyphs plastering the Pyramids. Numbers make beautiful shapes, and hold history within the bounds of their geometric representations. Patrick Cooney gave me a textbook from his college math course, History of Math. I pored through it. Numbers were my sweet, sweet relief from going home or facing all the damage I’d done to myself through drinking, self-harm, and drug use. Numbers gave me an unparalleled high. I felt cared for, inspired, and seen by this amazing teacher and the joy he carried.

Fast-forward to university where I began to explore computing and math. It was like something in my brain unlocked. I resonated deeply with Douglas Hofstader’s Godel, Escher, Bach: an Eternal Golden Braid. It’s a kooky exploration into linguistics, computation, and recursion. Suddenly I was gaining words to frame my own way of thinking. I wasn’t someone that horrible things happened to. I was someone who could reflect on my own thinking; my own cognition; my own journey through the world. I wasn’t just a product of my priors (a Bayesian outlook that I’ve taken seriously in the last few years). I wasn’t just the product of pre-programmed computations and signals of power coming from outside of myself. No, I was the data scientist who could choose what to measure, and what to learn. I was the data scientist that could sample from different populations, record my own feelings and thoughts, and make predictive models that could help me thrive.

Machine learning gave me the language to save myself.

a little cardboard robot looking at a heart made of skittles. don’t ask why, I just liked it.

I’m particularly fond of the idea that we start with some prior distribution of what we are likely to do or what is likely to happen. My priors were super distorted, as the data I received in childhood was that I wasn’t going to be respected. That prior distribution was true. It reflected true data in the world. It doesn’t mean that it was right. The Metropolis Hastings algorithm is a particular flavor of MCMC (Markov Chain Monte Carlo) that I was introduced to when doing research in the Computation and Language Lab. MCMC starts with some prior weights, and then explores the (infinite) hypothesis space to produce likely hypotheses that match data it was given. My life is an infinite hypothesis space made up of primitive functions. And I can explore that space, and overfit or underfit, just like machine learning algorithms. I can get caught in local maxima (places in life that I think are the best thing and I couldn’t possibly optimize more but in reality if I just go back down I’ll find a higher global maximum state).

A note about the Computation and Language Lab and the Kidd Lab. Steve Piantadosi, Celeste Kidd, Jessica Cantlon, and others that I knew less well were some of my new data that showed me that sexual assault will not be tolerated. They sued the University of Rochester for a cover-up and retaliation case after coming forward about Florian Jaeger’s behavior. They gave up their jobs and their lives to do what was right. Celeste and Jessica go on to be featured as some of TIME Magazine’s person of the year for standing up against institutional violence. Lindsay Wrobel and I led protests against the University of Rochester administration to fire the professor, resulting in the president stepping down but not removing the professor. We won the Susan B. Anthony award which was dope and I only bring this up because look how far I have come!!!

Because my data was so skewed, I needed a way to “jump” out of that data sample; into something better where I could learn how to be treated like a person. Some machine learning theorists call this “the teleport” or the “random jump” in things like gradient descent. We get caught in local maxima, where we think it’s the best scenario we could have. But really there’s an infinite world out there that could feel better; help us thrive. So we have to take that algorithmic leap of faith. We jump to a new place, a new way of doing things, a new way of treating ourselves. Just to try it out. We may be pleasantly surprised to see that it starts an entire feedback mechanism where we climb higher and higher in who we are and how we feel.

Machine learning metaphors gave me the mechanisms to jump, to believe in new data, to believe in a better world. I process social interactions in this way; and label them like a reinforcement learning algorithm would. I try to create clusters when I observe people or my research field just like a semi-unsupervised algorithm might. I connect deeply, and personally, to machine learning as therapy. I’m starting to write a book called Machine Learning Therapy, where I teach algorithms through a trauma recovery lens. If I say that here, maybe it will make it real. I want to share this tool with the world because despite the violence and bias that machine learning technology has enacted and perpetuated, there is also this huge light for me and my experience with it. With knowledge of the algorithms, I see the world in a way that is more within my control. I question how I would measure things differently, or try to help others with technology at scale. How would it hurt me? How would it help me?

I’m starting to write a book called Machine Learning Therapy, where I teach algorithms through a trauma recovery lens. If I say that here, maybe it will make it real.

When I say, “it’s personal”, you probably think I mean all the divulging of sexual violence. But that’s not what I’m talking about. I’m talking about the personal connection I share with algorithmic metaphors and the empowerment I gathered through thinking in a computational way about my own life and my own struggles. What moves can I make to heal my body and spirit? What function am I optimizing for? Is it joy? Is it health? Is it friendship? Is it success? Is it a linear or non-linear combination of all of those things? (probably). If I’m optimizing for one function, what features do I have to throw out? Given the curse of dimensionality, I might get severely stuck if I always tried to optimize for my whole being. So instead, I choose different functions for different tasks and different functions for different parts of my life.

When I say, “it’s personal”, you probably think I mean all the divulging of sexual violence. But that’s not what I’m talking about. I’m talking about the personal connection I share with algorithmic metaphors and the empowerment I gathered through thinking in a computational way about my own life and my own struggles.

Sure, I’ve learned a hell of a lot of machine learning along the way; now I can write the big algorithms from scratch, dig in with scikit and R and tensorflow and CUDA and all that. Sure, I can probably tell you a lot about how your financial model will fail, or your medical model could be improved. But machine learning has, and always will be, a tool that I use to reflect on myself and my growth. It is a therapeutic tool for me, and I want to share a little bit of that. If this resonated with you, on whatever dimension, I’d love to hear your thoughts.

Time to return to the work that I do; researching how to teach machine learning literacy through personal connection to algorithms and their consequences. My current study is situating learners as surveyors of their own data, learning tools to advocate for themselves against possibly harmful machine learning models. I dream of a world where they have the tools to be empowered; the knowledge to be heard; and the deep understanding to guide their own lives in an machine learned world.

Yim Register

Written by

Radical developer. Radical optimist. Machine learning literacy.

Welcome to a place where words matter. On Medium, smart voices and original ideas take center stage - with no ads in sight. Watch
Follow all the topics you care about, and we’ll deliver the best stories for you to your homepage and inbox. Explore
Get unlimited access to the best stories on Medium — and support writers while you’re at it. Just $5/month. Upgrade