This tutorial provides an introduction to the Markov process in a simple example where the state transitions are probabilistic. The aims of this tutorial is to help you understand Markov processes and history dependence, as well as to explore the behavior of a two-state telegraph process and understand how its equilibrium distribution is dependent on its parameters.
This tutorial builds on how deterministic and stochastic processes can both be a part of a dynamical system by simulating random walks, investigating the mean and variance of a Ornstein-Uhlenbeck (OU) process, and quantifying the OU process's behavior at equilibrium.
The goal of this tutorial is to use the modeling tools and intuitions developed in the previous few tutorials and use them to fit data. The concept is to flip the previous tutorial -- instead of generating synthetic data points from a known underlying process, what if we are given data points measured in time and have to learn the underlying process?
This tutorial is in two sections:
This lecture provides a summary of concepts associated with linear dynamical systems, covered in Linear Systems I (Intro Lecture) and Tutorials 1 - 4, and also introduces motor neuroscience/neuroengineering, brain-machine interfaces, and applications of dynamical systems.
This tutorial introduces the Sequential Probability Ratio Test between two hypotheses 𝐻𝐿 and 𝐻𝑅 by running simulations of a Drift Diffusion Model (DDM). As independent and identically distributed (i.i.d) samples from the true data-generating distribution coming in, we accumulate our evidence linearly until a certain criterion is met before deciding which hypothesis to accept. Two types of stopping criterion/stopping rule will be implemented: after seeing a fixed amount of data, and after the likelihood ratio passes a pre-defined threshold. Due to the noisy nature of observations, there will be a drifting term governed by expected mean output and a diffusion term governed by observation noise.
This tutorial covers how to simulate a Hidden Markov Model (HMM) and observe how changing the transition probability and observation noise impacts what the samples look like. Then we'll look at how uncertainty increases as we make future predictions without evidence (from observations) and how to gain information from the observations.
This tutorial covers how to infer a latent model when our states are continuous. Particular attention is paid to the Kalman filter and its mathematical foundation.
This tutorial presents how to estimate state-value functions in a classical conditioning paradigm using Temporal Difference (TD) learning and examine TD-errors at the presentation of the conditioned and unconditioned stimulus (CS and US) under different CS-US contingencies. These exercises will provide you with an understanding of both how reward prediction errors (RPEs) behave in classical conditioning and what we should expect to see if dopamine represents a "canonical" model-free RPE.
In this tutorial, you will learn how to act in the more realistic setting of sequential decisions, formalized by Markov Decision Processes (MDPs). In a sequential decision problem, the actions executed in one state not only may lead to immediate rewards (as in a bandit problem), but may also affect the states experienced next (unlike a bandit problem). Each individual action may therefore affect affect all future rewards. Thus, making decisions in this setting requires considering each action in terms of their expected cumulative future reward.
In this tutorial, you will implement one of the simplest model-based reinforcement learning algorithms, Dyna-Q. You will understand what a world model is, how it can improve the agent's policy, and the situations in which model-based algorithms are more advantageous than their model-free counterparts.