- School of Computer Science
and Engineering, Seoul National University
- Instructor: Prof. Byoung-Tak Zhang
- TA: Min-Oh Heo, (Room : 302-314-1, Tel :
02-880-1847)
- Classroom: 302-105 (Lab
for Practice: 302-311-1)
- Time: Tue
14:00-15:15 and Thu 14:00-15:15
- Course Description
This course aims to
understand the principles and mechanisms of cognitive agents that learn
continuously and sequentially over lifetime in dynamic environments. In this
lifelong learning setting, the target distribution of the learning is
typically non-stationary and changing. This is in sharp contrast to
the usual assumptions made in most supervised and unsupervised learning
algorithms. That is, lifelong sequential learning violates the assumption of
the examples being i.i.d (independent and
identically distributed) and the assumption of availability of all the
training data at the outset of learning. However, lifelong learning research
is still in its infancy and there is no single universal algorithm that
addresses all these issues. Thus, in this course we discuss three specific
aspects of lifelong learning problems, i.e. sequence learning, sequential
decision making, and moving target tracking. We review the existing machine
learning algorithms that best deal with these problems so far, namely hidden
Markov models, reinforcement learning, and particle filters. Based on this
knowledge, we will discuss how to develop a humanlike intelligent agent that
that can learn actively, sequentially, and lifelong in real-time from
real-life sensing data in real-worlds.
- Texts:
[1]
Candy, J. V., Bayesian Signal Processing, Wiley
& Sons, 2009. (Chs. 1-3, 7, & 9)
[2] Sutton,
R. S. & Barto, A. G., Reinforcement Learning, MIT Press, 1998.
[3] Zhang, B.-T., Communication as moving target tracking: Dynamic
Bayesian inference with an action-perception-learning cycle, In: Wachsmuth, I. et al. (Eds.), Alignment in
Communication: Towards a New Theory of Communication, Chapter 7, Simon
& Shuster, 2013 (to appear)
[4] Barber, D., Cemgil, A. T., &
Chiappa, S., Inference and Estimation in Probabilistic Time Series
Models, In: Bayesian Time Series Models, Chapter 1, Cambridge University
Press, 2011.
- References:
[5]
Thrun, S., Burgard,
W., & Fox, D., Probabilistic Robotics, MIT Press, 2005.
[6] Bishop, C. M., Pattern Recognition and Machine Learning, Springer-Verlag, 2006.
- Projects:
- Learning sequential patterns from a music corpus
- Learning mobile behaviors of smartphone users
- Evaluation:
- One open-book
exam (40%)
- Two miniprojects
+ one presentation + one report (50%)
- Participation in
discussion (10%)
- Announcement
- Text [1] pdf files were uploaded. You may download them.
- 9/17 class will be held
in the room 302-105.
- 9/24 class will be held
in the Lab 302-311-1 for the
1st project based on HMM.
- 9/26 class is cancelled
due to the official college event.
- Supplementary lecture
will be given on Oct 4, 13:30 ~
15:30 (Room 302-105)
- 10/15 class is cancelled
due to SNU foundation day.
- Exam: (Open book)
- You can bring any notes,
papers, and books you want on the exam except electronic devices such
as notebooks.
- For your preparation of
the exam, the following materials would be especially useful:
- HMM: text [4]
& Ch 13 in Reference [6]
- RL: Chs. 3, 4, 6, & 7 in text [2]
- Particle Filters: Chs. 2 & 3 in text [1], text [4]
- Project Poster Presentation will be done on separate 2 days (11/26, 11/28) due
to too much presenters.
- Please check your pt-date
in HERE.
- Presentation Venue: 302-308 (Not
302-105)
- All of presenters should
bring your printed poster.
- Pizza will be served!!
- New late policy
is updated. (see the project page)
|
