This post takes a deeper look at variational auto encoders (VAE). Like regular auto encoders, a VAE maps a large dimension onto a small set of latent variables, from which we can reconstruct something very close to the original input. However, unlike regular auto encoders, a VAE uses a probability distribution in the latent space.
Read onConvolutional neural networks are a very intuitive (and fast) type of network for image processing problems. It uses a mathematical operation called a convolution, to analyze/generalize different sections of the input.
Read onWhen I took my machine learning classes at Michigan, we learned how to code our architectures using Julia. It’s a very performant language (compared to something like Python) and was built with math and data science in mind.
Read onAt its core, machine learning is a statistical science, and like statistics, a great deal of care must be taken when formatting and analyzing data. They way we chose to quantify abstract topics and/or measure error will have a significant impact on the quality of our result.
Read onIn practice, machine learning algorithsm can be tricky to setup and tune. This post discusses the different metrics you can use to evaluate your algorithm, and how to fine-tine your setup based on how those metrics perform.
Read onAcross the vast field of robotics and controls, there is a common question of “how do we get from point A to point B?”. Without any constraints, the answer is pretty easy - go in a straight line - but this is not very useful. So how can we formulate a more useful question?
Read onIn this article, I talk about how to frame a trajectory problem as a mathematical optimization problem. Keep in mind that this is only one way to plan a trajectory. And when I say math, I really do mean math (the difficult kind). You have been warned.
Read onThere is so much literature out there on machine learning that it can get pretty overwhelming. This post introduces machine learning one concept at a time, without getting into any math.
Read onMachine vision refers broadly to any imaging analysis task performed by a computer. Meanwhile, simultaneous localization and mapping (SLAM) is a specific domain focused on obtaining an understanding of the environment and your position within it. These two often go hand in hand (but not always!).
Read onThe underlying principle of an optimization is pretty straightfoward - find the best way to do something. Best can mean fastest, safest, most efficient, shortest or thousands of other things, but the common denominator is that we are looking for the most (or least) of something. Mathematically then, this is pretty simple: we’re solving for a maximum.
Read on