2016-2017 2017-2018 2018-2019 2019-2020 2020-2021
Browse
by subject...
    Schedule
view...
 
  COVID-19 Scheduling Updates!
Due to recent announcements about Autumn Quarter (see the President's update), please expect ongoing changes to the class schedule.

1 - 2 of 2 results for: MS&E 321: Stochastic Systems

MS&E 321: Stochastic Systems

Topics in stochastic processes, emphasizing applications. Markov chains in discrete and continuous time; Markov processes in general state space; Lyapunov functions; regenerative process theory; renewal theory; martingales, Brownian motion, and diffusion processes. Application to queueing theory, storage theory, reliability, and finance. Prerequisites: 221 or STATS 217; MATH 113, 115. (Glynn)
Terms: Spr | Units: 3

OIT 611: The Drift Method: from Stochastic Networks to Machine Learning

Overview: This course is an introduction to the drift method in sequential decision-making and stochastic systems, a family of simple, yet surprisingly powerful, meta-algorithms that in each step the greedily and incrementally minimizes some potential function. Manifested in various forms, the drift method powers some of the most popular algorithmic paradigms in stochastic networks (MaxWeight, BackPressure), oneline learning, optimization and machine learning (SGD, Langevin dynamics, TD-learning). Using the Drift Method as a unifying theme, we will survey major developments in these areas and answer questions such as: What may explain the method¿s effectiveness? How can we rigorously evaluate its performance? We will develop rigorous probabilistic and optimization methodologies for answering these questions, such as Lyapunov functions and stability theory, state-space collapse, weak convergence and Stein¿s method. In terms of application topics, the course is roughly evenly divided bet more »
Overview: This course is an introduction to the drift method in sequential decision-making and stochastic systems, a family of simple, yet surprisingly powerful, meta-algorithms that in each step the greedily and incrementally minimizes some potential function. Manifested in various forms, the drift method powers some of the most popular algorithmic paradigms in stochastic networks (MaxWeight, BackPressure), oneline learning, optimization and machine learning (SGD, Langevin dynamics, TD-learning). Using the Drift Method as a unifying theme, we will survey major developments in these areas and answer questions such as: What may explain the method¿s effectiveness? How can we rigorously evaluate its performance? We will develop rigorous probabilistic and optimization methodologies for answering these questions, such as Lyapunov functions and stability theory, state-space collapse, weak convergence and Stein¿s method. In terms of application topics, the course is roughly evenly divided between stochastic queueing networks versus optimization + machine learning. Objective: For students to acquire fundamental methodologies that can be applied to tackling problems in dynamic decision-making, stochastic modeling and machine learning. Target Audience: Graduate students / advanced undergraduates with a solid grasp of probability and stochastic processes (Stat 310A / MS&E 321, or equivalent). Strong background and interests in queueing networks is highly recommend.
Terms: Aut | Units: 3
Instructors: Xu, K. (PI)
Filter Results:
term offered
updating results...
teaching presence
updating results...
number of units
updating results...
time offered
updating results...
days
updating results...
UG Requirements (GERs)
updating results...
component
updating results...
career
updating results...
© Stanford University | Terms of Use | Copyright Complaints