Machine learning models aren’t just the future of computer science. They’re not just infinite amounts of nested if statements and they’re certainly not just a bunch of linear algebra. I wouldn’t even go so far as to say these ML models are just fancy applied statistics. Machine learning models are all of those things combined to make some extremely sexy hot graphics and diagrams.
Who doesn’t remember that first time they threw that hot fresh moving figure up on a power point slide or runway model and just watched everyones jaws drop in aw. Even undergraduate students typically remember their first exposure to this modern technology when awakening something primal in their gut.
The hotness of ML models isn’t only appreciated by the nerds. Even program managers salivate as soon as they begin to talk to stake holders about how hot their teams machine learning algorithms are. There’s nothing that can attract more funding faster than some hot buzz wordy block diagrams and machine learning models to update upper management with. Judging by nothing more than hotness, here are THE hottest machine learning models.
12. Expectation Maximization
Expectation? I think I remember that lecture in first week of stats. Maximization? Yeah, that’s like an excel function right? Expectation Maximization (EM)? Damn that is one sexy clustering method. Typically making this list every year since the 80’s, Expectation Maximization shows you can be an old algorithm and still be hot enough to walk that runway. EM is everyone’s first crush with its simple two step process. What cluster graphic couldn’t use a little EM to get the viewers drooling.
11. Deep Boltzman Machine
Oh my God is that an undirected probabilistic graphical model walking towards me?!? That’s not just any Markov random field, that’s a deep Boltzmann machine. In only three layers it’s true some of the most beautiful things come in small packages. Some haters might just say “Oh, it just looks like any normal Deep Belief Network.” But do DBNs have bidirectional connections? I don’t think so.
10. Locally Estimated Scatterplot Smoothing (LOESS)
Woah, better throw away your canned spline functions, LOESS is the hottest new way to fit a curve without over fitting! LOESS is the answer we’ve all been looking for after late nights of fitting messy data. LOESS is just that relaxed go with the flow model that won’t let those outliers tell her nothing.
9. Hierarchical Clustering
Expectation Maximization isn’t the only clustering algorithm model to make the list welcome the one, the sexy, the Hierarchical Clustering. This model has it all, easy to understand tree diagrams, easy to demo code, simple metrics, and even graphics a manager can understand. Hierarchical Clustering might not be the youngest clustering algorithm or even the fastest at O(n^3) speed but what a hot demo.
8. Mixture Discrimination Analysis
It’s really rare to find a linear discriminant analysis model on these lists but how could we not include something as hot and adaptive a classification model as Mixture Discrimination Analysis (MDA). This gorgeous new model is actually the cousin of this lists long time favorites EM. MDA, ever the hot model, instead uses more robust S-estimators to find those tricky unknown parameters instead of EM’s M-step easily fooled by outliers. Wow! I think it’s starting to really get steamy in here. That’s like if Ryan Gosling were to date Scarlett Johansson and they were ML models and had a baby. That baby would be MDA!
7. Self Organizing Map
Is that some boring Map coming down that runway? Nope, it’s only one hot Artificial Neural Network. Self Organizing Map (SOM) is the wild one of the bunch by taking control of its own life and even learning completely unsupervised. You can’t tell this model what to do, you can’t guide it, and you certainly can’t even supervise this model! That much unbridled machine learning certainly is hot enough to make our list.
6. Approximate Bayesian Computational Method
Oh damn, is that computational method not even evaluating the likelihood function and still estimating a posterior distribution? SO. FREAKIN. HOT. I don’t know what floats your boat, but I always find myself Simping for a classic Bayesian approach. Bayes is THE buzz word that will get a model on this list. There’s nothing more sexy than using that fundamental law of statistics to make your algorithm sound better on paper. The fact that you can do it without all that math from solving for a likelihood function makes it even more sexy.
5. Stein Variational Gradient of Descent
Ya know what’s even sexier than a Machine Learning model based off a Markov chain Monte Carlo (MCMC) posterior distribution? A fast iterative method to find that MCMC posterior distribution by minimizing its Kullback-Leibler divergence, that’s what. I can hear your jaws dropping to the floor like a cartoon right now almost as fast as the Stein Variational Gradient of Descent (SVGD) can transport a set of particles to match a target distribution. It’s so hot I even think This makes variation methods efficiently solvable! Damn, let the other models catch up SVGD!
4. Explainable Neural Networks
Always the most revealing ML model, Explainable Neural Networks are always opening up that black mysterious box. Sometimes it’s the mystery that makes these artificial intelligences so hot but with this model it’s that honest transparency that’s captured all of our hearts!
3. Long Short Term Memory Network (LSTM)
Who hasn’t fallen in love with this model after following a generate text using an LSTM neural network tutorial to end with a “well the text it generated was
incomprehensible nonsense something? I know I have had that exact experience and wow what a model. Hubba Hubba, baby’s got back propagation! This super hot RNN doesn’t just process single instances of data but sequences like audio, Tarot Card readings or video! Good lord it’s no wonder this is one of the hottest methods in classification and prediction applications for time series data!
2. Multi-Headed Attention
Ya know what’s newer and hotter than a deep learning neural network? Multiple Recurrent Neural Networks because one is just not fast enough! You heard that right, the Multi-Headed Attention model can query and look up answers simultaneously from three memory banks! Say goodbye to narrow pipelines! Goddamn, I don’t think you could think of a sexier way to make a natural language processor.
1. Generative Pre-trained Transformer (GPT)
It should be no surprise that the most hot Machine Learning model is none other than the Generative Pre-Trained Transformer. As an open sourced artificial intelligence software released to the public from OpenAI, GPT is that model we’ve all been dreaming about. It’s that sort of sexy software package that can load data, train and generate predictive data in three lines of code which couldn’t possibly be problematic. What’s in the model you ask? Who cares! JUST LOOK AT THAT AMAZING HOT DEMO OUTPUT!
With the advances in GPT-3 to easily generate text there is a growing rivalry in between the GPT fans and the hard liner “implement your model from scratch” LSTM users. Sounds a lot like the TensorFlow Scikit-learn rivalry that erupted in Chicago! I’m sure this won’t be the last we hear of these hot ML algorithms beef for quite some time!
If you enjoyed this article please like, share, and subscribe with your email, our twitter handle (@JABDE6), or our facebook group here for weekly content.
I’m sure some of you readers are also complaining about your favorite ML/Stats model not showing up on our list despite it’s massive hotness. If you care that much about arbitrary lists sexualizing STEM topics you sound like a contributor and should check out our contribution page.