Learning, adaptation, and uncertainty: machine learning for cognitive modeling 2500-EN-CS-SM-02
This course is structured around the concept of uncertainty. In
different statistical learning tasks uncertainty is modeled in different ways. In supervised learning past (historical) data is certain, and uncertainty concerns the future. In reinforcement learning uncertainty concerns outcomes of actions made in the present. With sequential decisions past actions influence future states, which leads to – uncertain – indirect consequences. Finally, when uncertainty enters observation space, the present state of the system is uncertain as well.
The lecture will start with a primer on basic machine learning concepts (classification, regression, cost function, overfitting, etc.) in the context of decision modeling. Linear and logistic regression will be introduced as prototypical regression and classification algorithms. A presentation of multilayer neural networks as a natural generalization of logistic regression will follow. Then, a significant portion of the course will be dedicated to reinforcement learning techniques (temporal difference learning, policy gradient methods) with applications to cognitive modeling and robotics. The last section of the course will be devoted to swarm intelligence algorithms (evolutionary algorithms, ant colony optimization) and techniques for dealing with uncertainty in the observation space (Kalman filter, particle filter).
In the computer lab classes, students will implement and apply the algorithms discussed during the lecture. These classes will have the form of a playground in which learners discover important properties of the algorithms through experimentation.
Learning activities:
Lecture with a short quiz at the beginning of each class.
Group discussions in class.
Class in a computer lab focused on self-experimentation.
Learning outcomes
Student knows and understands:
- basic concepts of machine learning, popular techniques and software libraries (K_W07)
- different families of statistical techniques for dealing with uncertainty (K_W07)
- role of machine learning in artificial intelligence, cognitive processes modeling, experimental data analysis (K_W01, K_W03, K_W04)
Student is able to:
- discuss modeling assumption behind particular techniques (K_U03, K_U04)
- design and implement a procedure of machine learning experiment using existing data (K_U06, K_U07)
- build their understanding of machine learning on their own through reading scientific literature and own experimentation (K_U11, K_U12)
Assessment criteria
a) Assessment methods: exam, test, ongoing work in class, and a small project.
b) Components of the final grade and their weights:
The final grade from the lecture is based on a written exam covering the lectures and selected literature.
The final grade from the computer lab class is based on work done in class and (optionally) a small project.
Lecture and class in computer lab will be graded separately, however:
1. Obtaining a positive grade from computer lab class is a prerequirement for taking the exam.
2. A good grade from the class may increase the final exam grade.
Lecture final grade components:
30% Middle of semester test (after lecture 6).
50% Written exam.
20% Computer lab bonus.
Computer lab class final grade components:
60% Work in class.
40% Project.
c) Grading scale:
- over 50%: 3
- over 60%: 3+
- over 70%: 4,
- over 80%: 4+
- over 90%: 5
d) Requirements for retaking the assessment: N/A
e) Exams in the exam session:
i) Requirements for taking the exam: adequate attendance, obtaining a positive grade from computer lab class.
ii) Exam can be retaken in the case of a negative exam grade. It is
not possible to retake an exam if the grade is positive.
iii) No early exam session (“zerówka”) will be offered.
Attendance rules:
Attendance to the lecture and the class is obligatory, 2 unexcused absences are allowed.
Additional information
Additional information (registration calendar, class conductors, localization and schedules of classes), might be available in the USOSweb system: