ACIT4030 Machine Learning for 3D Computer Vision Emneplan

This course will present the state of the art in algorithms for machine learning for the 3D environment. We will cover topics related to deep learning for 3D data such as point clouds, multi-view images, and shape graphs. The course covers applications like classification, segmentation, shape retrieval and scene representation.

The following required coursework must be approved before the student can take the exam:

A project plan document containing a description of the chosen data set, a preliminary research question and suggested tools and method to apply.

No formal requirements over and above the admission requirements.

Upon successful completion of the course, the candidate:

Knowledge

• has knowledge of problems within graphics and imaging that are applicable to machine learning, such as classification, segmentation, correspondence detection, and shape retrieval.
• has a good understanding of problems related to 3D shape and image synthesis.

Skills

• is able to apply state-of-the art machine learning algorithms to real-world problems related to imaging and 3D graphics.

Competence

• is aware of the state of the art in algorithms for machine learning on 3D data.
• has experience with real world problems within the course domain, with a focus on solutions using deep neural architectures.

Teaching approach is a combination of traditional weekly lectures and practical work on a semester group project. Lectures will present influential research for relevant topics. The semester group project exposes the student to a chosen real-world problem relevant to the course topic.

Practical training

The student will be exposed to programming with repositories such as ShapeNet and will have created solutions for real-world problems related to deep learning for 3D data.

The following required coursework must be approved before the student can take the exam:

Two mandatory group assignments consisting of technical tasks, summarized in reports (about 10 pages each).

The exam consists of three parts:

1. Oral presentation of 15 minutes (20% of the final grade), individual or in a group of two
2. Written evaluation of another student presentation, 500-1000 words (10% of the final grade), individual or in a group of two
3. Final project report between 6000 and 11,000 words (70% of the final grade), individual or in a group of two.

All three parts of the exam must be passed in order to pass the course.

The oral examination cannot be appealed.

New/postponed exam

In case of failed exam or legal absence, the student may apply for a new or postponed exam. New or postponed exams are offered within a reasonable time span following the regular exam. The student is responsible for registering for a new/postponed exam within the time limits set by OsloMet. The Regulations for new or postponed examinations are available in Regulations relating to studies and examinations at OsloMet.

The course covers the foundations and recent advances in Machine Learning from the point of view of Statistical Learning Theory. The goal of this course is to provide students with the practical skills to support the theoretical knowledge to (1) develop machine learning solutions to challenging problems and (2) to be able to develop the acquired expertise further.

The theoretical aspects of statistical learning will be illustrated with concrete problems and tasks in Python.

No formal requirements over and above the admission requirements.

The student should have the following outcomes upon completing the course:

Knowledge

Upon successful completion of the course, the student:

• will have a good understanding the different concepts and methods of supervised and unsupervised statistical learning and how to apply them on large data.
• has advanced knowledge of probabilistic formulation of the various learning problems.
• has focused knowledge of theoretical aspects of the different methods in machine learning and statistical learning, as well as a deep knowledge of concepts and assumptions behind each method.

Skills

Upon successful completion of the course, the student:

• can apply different high-dimensional regression techniques on data
• can apply different classification techniques on data
• can apply clustering techniques on data
• can apply dimension reduction techniques on data
• can make informed decisions on which method suits best for a particular problem and/or data set
• can derive learning algorithms for new models and analyze new data with them.

General competence

Upon successful completion of the course, the student:

• can apply different predictive models on data and assess their performance
• can use supervised and unsupervised learning in different real life problem

This course has three main parts.

The first three weeks cover revisions on statistics and scientific method, briefly presenting also the basics in (Python) programming, selection of a problem, etc. (i.e. revisions of several of the focus points in DATA3800).

After that there will be 7 weeks with presenting and solving exercises, covering the book "Introduction to Statistical Learning".

Finally the course has 3 to 4 weeks bridging the content presented during the weeks before with other scientific fields and topics, namely in Applied AI and in Mathematical Modelling. In particular, connections to the courses " Applied and Computational Mathematics (ACIT4310)" and "Evolutionary Artificial Intelligence and Robotics (ACIT4610)" will be addressed.

The weekly classes will be divided in three parts: (1) a theoretical exposition of the new content introduced each week, (2) one set of exercises/problems implementing the content presented during the theoretical exposition, and (3) supervision of each student in his/her specific project (see "Assignment").