TU Wien:Deep Learning for Visual Computing VU (Hermosilla Casajus)

Daten

Zuordnungen

Masterstudium Data Science	Modul MLS/EX - Machine Learning and Statistics - Extension (Gebundenes Wahlfach)
Masterstudium Business Informatics	Modul DA/EXT - Data Analytics Extension (Gebundenes Wahlfach)
Masterstudium Visual Computing	Modul AI for Visual Computing (Gebundenes Wahlfach)

Inhalt

noch offen, bitte nicht von TISS/u:find oder Homepage kopieren, sondern aus Studierendensicht beschreiben.

Ablauf

noch offen

Benötigte/Empfehlenswerte Vorkenntnisse

noch offen

Vortrag

noch offen

Übungen

SS24:

Some of the worst and unclear exercise assignments I have ever seen. There were random code snippets to fill in, but there was no clear explanation of the existing package structure, what classes we have and how they should be used. Function documentation was bad, parameters were unclear and not explained anywhere. There was just no logical thread to follow. Information was rather randomly spread out across the assignment document, the (3rd party) dataset documentation or the source code, but often not in the place where you would expect it. Most of the time it feels like the assignment wants you to implement the architecture/classes in a certain way, but it gives you no pointers on how to do that. There are some function stubs with TODOs and a rough explanation on what the function should do and some (verbal) limitations on data types and stuff, but that's mostly not very helpful in pointing you in the right direction. Having at least some data types/fields pre-defined would be tremendously less frustrating for finding the clearly intended solution. The assignment document itself also doesn't provide any guidance or overview on what it's about. There is nothing to give you a hint about what you're actually going to be doing other than "becoming familiar with pytorch usage". "Read the code and make sure to understand what the individual classes and methods are doing" is not very helpful when we have to write half the classes and methods ourselves, and the documentation does not tell you what the class should be doing, but rather just gives some vague hints on how to implement it. There is no high-level overview, and the supposed "good starting point" ("train.py" file) is a file where every second line is incomplete and has to be filled in. So in other words, it's not a good starting point AT ALL. You basically have to read through the entire assignment first to try and piece together what it's even about.

There was A LOT of boilerplate code to implement (file loading/writing, datatype conversion to the required format, field initialization, __init__ methods, etc.) which has basically zero learning effect for the topic at hand. And little to no provided verification/test code that you can run to make sure that what you implemented is behaving correctly. Comparing this with the recommender systems lecture in the same semester, which had a similar style of fill-in-code assignments, but where you had the expected output for verifying the correctness of your code at basically every corner of the jupyter notebook, the difference is night and day. And this isn't really about difficulty either, it just makes the assignment frustrating to solve and vague, because you have no idea whether you're on the right track, or if your basic data structure/format is completely wrong, since there is clearly an intended solution, but nothing to make sure you don't veer off the correct path.

In terms of time investment, the course suggests 34h for all 3 assignments. This is hopelessly, ridiculously out of touch with reality. You will probably use most of this time on the first assignment alone. I ended up dropping the course during the first exercise, because it just wasn't worth the time that I would have to put into it.

Prüfung, Benotung

noch offen

Dauer der Zeugnisausstellung

noch offen

Zeitaufwand

noch offen

Unterlagen

noch offen

Tipps

noch offen

Highlights / Lob

noch offen

Verbesserungsvorschläge / Kritik

noch offen

Materialien

Diese Seite hat noch keine Anhänge, du kannst aber neue hinzufügen.