Deep Learning
Theory and applications for Computer Vision
046003, Spring 2020
Meyer 815, EE building, The Technion.
Description
Deep convolutional neural networks (CNNs) have revolutionized computer vision, achieving unprecedented performance in high-level vision tasks such as classification, detection, segmentation an low-level vision tasks.
This crash course aims to give a first dive into the world of deep learning, with hands-on experience. During an intensive one-week course, students will learn to implement, train and debug their own neural networks.
We will focus on teaching how to set up the problem of image recognition, the learning algorithms (e.g. back-propagation), practical engineering tricks for training and fine-tuning the networks. The final assignment will involve training a large scale convolutional neural network and applying it on a practical data-set.
This course is based on Stanford's CS231n: Convolutional Neural Networks for Visual Recognition course.
Course Staff
- Organizers: Tamar Rott Shaham, Idan Kligvaser, Yair Moshe
- Invited speakers: Yuval Bahat, Regev Cohen, Yona Coscas
Registration
- EE undergrad students: through UG: https://ug3.technion.ac.il/rishum/course/046003/201902
- EE grad students: Graduate Studies Secretariat
- Free listeners (or students who wish to take the course with out academic credit): Please register through the Free Listeners Registration Form (the number of free listeners is limited)
Prerequisites
Knowledge of the following is required:
- Machine learning 046195 or 236756 (students from W19 semester are welcome).
- Proficiency in Python.
- Linear algebra.
- Basic Probability and Statistics.
Free listeners please fill the following registration form.
Class Time and Location
February 23-27, Sunday-Thursday 9:30am-12:20pm.
Meyer 815, EE building, The Technion.
Grading Policy
The course is designed for students planning to attend all lectures and would like to dive into deep learning.
The course is worth one credit point (not included in any chain of studies, but can be regarded as a "faculty" points (בחירה פקולטית)).
Course Project (%70):
The project will involve implementing and training Neural Networks for a given computer vision task. Submission is in pairs, and the due is within 3 weeks after the course end.
Course Exam (%30):
The exam does not require any preparation for those who attended all lectures.
Course Project
submission deadline: 19/04/2020 (postponed). Please submit the project to this email.
You can look for a project teammate in the following spreadsheets.
Useful tutorials:
- How to fetch data from kaggle by using google colaboratory (link).
- PyTorch 60 minutes blitz tutorial (link).
- Transfer learning tutorial (link).
Contact Information
- For registered students: Yair (myair@technion.ac.il)
- For free listeners: Tamar (stamarot@campus.technion.ac.il)
- Regarding the final project: Idan (kligvasser@campus.technion.ac.il)
Registration to the course is available through the links above (there is no need to write us for registration)