191 - 200 of 236 results for: MS

This Ph.D. course will study advanced topics in causal inference, with a focus on nuances of experimental design and policy evaluation, particularly in settings with interference. We will emphasize discussion of a range of experimental designs, as well as applications in networks and marketplaces. The course will be taught in a seminar format, with an emphasis on in-depth discussion of recent research papers at the frontiers of this area. The class is restricted to Ph.D. students; exceptions require instructor approval.

Advanced stochastic modeling and control of systems involving queueing and scheduling operations. Stability analysis of queueing systems. Key results on single queues and queueing networks. Controlled queueing systems. Dynamic routing and scheduling in processing networks. Applications to modeling, analysis and performance engineering of computing systems, communication networks, flexible manufacturing, and service systems. Prerequisite: 221 or equivalent.

An in-depth treatment of algorithmic and game-theoretic issues in social choice. Topics include common voting rules and impossibility results; ordinal vs cardinal voting; market approaches to large scale decision making; voting in complex elections, including multi-winner elections and participatory budgeting; protocols for large scale negotiation and deliberation; fairness in societal decision making;algorithmic approaches to governance of modern distributed systems such as blockchains and community-mediated social networks; opinion dynamics and polarization. Prerequisites: algorithms at the level of 212 or CS 161, probability at the level of 221, and basic game theory, or consent of instructor.

Random graph theory, Erdos-Renyi, and other network models, the algebra of graph homomorphisms, limits for dense and sparse graphs, and applications in algorithm design, graph representation learning, and others.

Within a couple of decades, or less, it is plausible that humans will create an AI that is much smarter than humans in practically all domains of human activity. We refer to such an AI as a superintelligence. The alignment problem is how to make sure that such a superintelligence acts according to its creator's intent. This course is intended for a technical audience interested in thinking about this problem. Prerequisites: one graduate-level machine learning course and one course that studies agents (e.g., AI, RL, decision analysis, economics).

The advent of cryptocurrencies, NFTs, and more generally, online financial instruments that are not controlled by large governments or banks, has resulted in many new and innovative mechanisms for decentralized finance. This class studies these new mechanisms from the viewpoint of algorithms, optimization, and market design. While blockchains have been a primary motivation behind many of these new developments, the resulting algorithmic and market design issues are more general, and we abstract away the specifics of blockchains and the underlying cryptographic primitives. Topics include distributed and multi-asset exchanges, liquidity pools, automated market makers, credit networks, and rollups (off-chain transactions). We draw connections between these new market mechanisms and more well-studied directions such as prediction markets and Arrow-Debreu equilibria. Includes a research project and a discussion of open directions. Prerequisites: algorithms at the level of CS 161 and optimization at the level of MS&E 111.

The first half of this course provides a rigorous introduction to the foundations of stochastic systems and control theory in discrete-time. The second half explores the associated applications in machine learning theory, with a particular emphasis on reinforcement learning and generative diffusion models. Throughout the course, financial applications will be a central theme, including topics such as algorithmic trading (optimal execution, portfolio optimization, and smart order routing), reinforcement learning for market making, and the generation of financial scenarios and time series using diffusion-based generative models.

This course is taught in 3 modules - (1) Markov Processes and Planning Algorithms, including Approximate Dynamic Programming (3 weeks), (2) Financial Trading problems cast as Stochastic Control, from the fields of Portfolio Management, Derivatives Pricing/Hedging, Order-Book Trading (2 weeks), and (3) Reinforcement Learning Algorithms, including Monte-Carlo, Temporal-Difference, Batch RL, Policy Gradient (4 weeks). The final week will cover practical aspects of RL in the industry, including an industry guest speaker. The course emphasizes the theory of RL, modeling the practical nuances of these finance problems, and strengthening the understanding through plenty of programming exercises. No pre-requisite coursework expected, but a foundation in undergraduate Probability, basic familiarity with Finance, and Python programming skills are required.

This doctoral seminar critically examines the research frontiers of Decentralized Finance, focusing on advanced topics in tokenomics, the intricacies of market design on various blockchain architectures, and the evolving market microstructure of digital assets. Students engage deeply with current literature from economics and computer science, analyze incentive structures and technical constraints, and utilize novel blockchain datasets for quantitative inquiry. The course equips students with theoretical and empirical tools for conducting cutting-edge research in this interdisciplinary field. Familiarity with game theory is recommended.