Transcript for:
Linear Algebra Refresher Course

Hello and welcome to the "Linear  Algebra" day of the Pre-reqs refresher!   I'm Ella Batty. I am a Lecturer and Curriculum  developer for Computational Neuroscience at   Harvard. Within Neuromatch, I am the Coordinating  Academic Officer for the Comp Neuro course.   As you can tell, I'm interested in comp neuro  education, and I'm also interested in using   Machine Learning approaches to model neurons, and  in methods for understanding complex networks.   Outside of science, I like reading, dogs, and  (most importantly) binging murder mysteries.   So, I want to start with why you should care  about linear algebra. Why we're asking you to   spend the next couple hours of your life learning  it? Linear algebra is a really foundational math.   It's central to most other areas of math and a  lot of applications. So, it's used all the time   in physics, computer science, economics,  and others. Importantly for us, it's also   used a lot in neuroscience, and especially in  computational and theoretical neuroscience. And   so, we'll actually use concepts from linear  algebra in over half of the comp neuro days.   It's a good fit for neuroscience because you can  think of linear algebra as the language of data.   It's how we organize, transform, analyze  data. As a really simple example of that,   let's assume that you are recording from three  different neurons. So, neuron 1, neuron 2,   and neuron 3. While you are recording from these  neurons you present an image to the animal.   So, let's say you present an image of a dog,   and you find that "neuron 1" fires at "10 Hertz",  where Hertz is the unit spikes per second.   "Neuron 2" fires at "50 Hertz". And "neuron 3"  at "2 Hertz". Then you can show another image,   let's say of a cow, and you can again record  the activities of these three neurons.   So, we only have 6 numbers, but it's already  starting to get a little bit messy. I have to   tell you a lot of different pairings of: "neuron  2" to Dog is this, "neuron 3" to Cow is this,   and so on. And so we can organize this data  using a linear algebra concept called "vectors".   And these brackets are denoting  that this is now a vector.   So, vector is basically an ordered list of  numbers. So, here the ordering is our neural   ordering, so you know that the first component  in the vector is always "neuron 1", the second   component is "neuron 2", and the third is "neuron  3". So now if I give you a new vector and tell you   it defines rates to a tree stimulus, you instantly  know which number corresponds to which neuron.   And we can also do operations on these vectors.  So, we'll talk about this operation later in the   course, but we can do "vector subtraction". And  so we can do this to start to understand how the   neurons are responding differently to dogs and  cows. So, when we do vector subtraction, we're   subtracting the individual components. So this is  "10 - 12" is "-2", "50 - 8" is "42", and so on.   So, this difference vector tells us how  much more the neuron responded to the   dog image than the cow image. So, today you  will start with this tutorial on vectors.   You'll learn about the definition of a vector,  more about vector properties and operations,   and how to define space through vectors. In the  second tutorial, you'll learn about matrices,   so you'll learn about how matrices can transform  space, about their properties and operations,   and about eigenvalues and eigenvectors.  And then if you have time (but absolutely   no worries if you don't), you can tackle the  bonus tutorial on "Discrete Dynamical Systems".   So in this tutorial, you'll model a very  simple neural circuit, and then you'll   understand the dynamics of this neural circuit  using eigenvalues. And don't worry if all of   these words are nonsense to you. Hopefully,  by the end of the day, they will not be.