Mathematical Foundations of Machine Learning
Prof. Matthieu Bloch
Wednesday, September 18, 2024 (v1.0)
Last time
- Last class: Monday September 16, 2024
- We talked about least square regression
- To be effectively prepared for today's class, you should
have:
- Come to class last week
- Gone over slides
and read associated lecture
notes
- Logistics
- Office hours
- Jack Hill: 11:30am-12:30pm on Wednesdays in TSRB and hybrid
- Anuvab: 12pm-1pm on Thursdays in TSRB and hybrid
- Gradescope
- Make sure you assign each solution to each question
- Make sure you do some quality control on your submission
Solving the least-squares problem
Any solution to the
problem must satisfy This system is called normal
equations
Facts: for any matrix
We can say a lot more about the normal equations
- There is always a solution
- If , there
is a unique solution:
- if
there are infinitely many non-trivial solution
- if , there
exists a solution for
which
In machine learning, there are often infinitely many
solutions
Minimum norm 2 solutions
Reasonable approach: choose a solution among
infinitely many with the minimum energy principle
- We will see the solution is always unique using the SVD
For now, assume that , so that the
problem becomes
Regularization
- Recall the problem
such that
- There are infinitely many solution if is non trivial
- The space of solution is unbounded!
- Even if , the
system can be poorly conditioned
- Regularization with consists in solving
- This problem always has a unique solution
- Note that is in the
row space of
Ridge regression
We can adapt the regularization approach to the situation of a
finite dimension Hilbert space
- We are penalizing the norm of the entire function
Using a basis for the space , and constructing
as earlier, we
obtain with the Gram
matrix for the basis.
If is invertible, we find the
solution as and we can reconstruct the function as .
If is well conditioned,
the resulting function is not too sensitive to the choice of the
basis
Least-Squares in infinite dimension Hilbert spaces
In , the problem has a solution
- with
- is dimension independent!
- We will be able to extend this to infinite dimensional Hilbert
spaces!
Let be a Hilbert space
and let be the function
we are trying to estimate
- This happens in Reproducing Kernel Hilber Space (RKHS)
Next time
- Next class: Monday September 23, 2024
- To be effectively prepared for next class, you
should:
- Go over today's slides
and read associated lecture
notes
- Work on Homework 3 (to be posted soon)
- Optional
- Export slides for next lecture as PDF (be on the lookout for an
announcement when they're ready)


1/1
Mathematical Foundations of Machine Learning
Prof. Matthieu Bloch
Wednesday, September 18, 2024 (v1.0)