Class Times: Monday and Wednesday: 3:00pm - 4:30pm Units: 3-0-9 Location: 46-5193 Instructors: Tomaso Poggio (TP), Shimon Ullman (SU), Daniel Harari, Daniel Zysman.
TA: Darren A. Seibert Office Hours: Friday 2-3 pm in 46-5156, Poggio's lab lounge (by appointment) Email Contact : 9.54@mit.edu Further Info: 9.54 on the Stellar system: http://stellar.mit.edu/S/course/9/fa14/9.54/ Synopsis
This class takes a computational approach to learning in the brain by neurons and synapses. We will examine supervised and unsupervised learning as well as possible biological substrates, including Hebb synapses and the related topics of Oja flow and principal components analysis. We will discuss hypothetical computational primitives in the nervous system, and the implications for unsupervised learning algorithms underlying the development of tuning properties of cortical neurons. This course also focuses on a broad class of biologically plausible learning strategies.
Prerequisites
Course 9.40 - Intro to Neural Computation.Grading
- Homework assignments (PSETS) : 40%
- Midterm exam : 20%
- Final exam : 30%
- Class attendance and participation : 10%
Class attendance
Classroom attendance is mandatory. You should notify the instructor if you have varsity sport events, medical school interviews, or sickness (doctor's notice required) that will prevent you to attend class.
Homework assignments
There will be a total of four (4) homework assignments. Release and due dates with a specific time deadline are indicated in the following table:
Homework assignment Release date Due date Problem Set #1 9/10 9/24 11:59 pm Problem Set #2 9/24 10/8 11:59 pm Problem Set #3 10/8 10/20 11:59 pm Problem Set #4 11/26 12/5 11:59 pm Assignments require the use of MATLAB. MATLAB is provided for free to registered students (you need a valid MIT certificate). Instructions for downloading and installation are available on this website: http://ist.mit.edu/matlab/all/student
Collaboration is encouraged on problem sets, but you must write up your own solutions and develop your own MATLAB code. Therefore, you may work with others to develop the algorithms for MATLAB, but you should produce the code on your own. Please list the names of all your collaborators on the top of each problem set.
Policy on Late Assignments:
Excused extensions on assigned work will be given only for significant illness or family crisis. Either a Dean's note or a note from MIT medical should properly document this. If an excused extension or postponement is requested, you must notify us prior to the class period for which the work is due.
In all other late submission cases, we will deduct 20% for the first late day and 10% for all days thereafter for a given homework.
Instructions for submitting homework assignments:
Please submit your problem sets through the Stellar course website as a .zip file containing:Institute Policy on Integrity:
- A single file (.pdf or .docx) containing the answers to all questions, in order. This includes figures. It does not include MATLAB code (see below). The easiest way to achieve this is to create a word-processing or LaTeX document (the latter is strongly suggested), type your answers, and paste in your figures. For problems involving equations, if you are not comfortable typing equations in your word processor, you can hand-write something and scan it (using a scanner, strongly suggested). If you instead take pictures with your cell phone, ensure that they are crisp, clear and easy to read when you paste them to the document. This time we will not allow resubmission of unclear, hard to read or blurred images. Note that we still require the answers to be in order, so you will need to paste your scanned answers into the proper place in the document. If you use a word processor other than MS Word, you must save your document into ".pdf" format and submit that.
- All the MATLAB code you used to get the answers and plots. This can be as many ".m" files as you need, as long as they are named in a way that makes it easy for us to know which file is the code for which problem. If you use a single ".m" file to run different questions of the problem set, and you comment out lines to obtain different answers, make comments that will allow us to reproduce your figures with minimal effort.
- All plots must have labelled axes, units, and figure captions.
The Institute obliges us to remind you of its policy on integrity. It can be found at the website http://web.mit.edu/academicintegrity/. Please read it if you have not already done so.Syllabus
Some of the later classes may be subject to reordering or rescheduling.
Class Date Title Instructor(s)