Data-Driven Decision Processes Boot Camp 2022
2022-08-22..2022-08-26
Berkeley, United States
[Mathematics of Online Decision Making 2020, Quantifying Uncertainty: Stochastic, Adversarial, and Beyond, Optimization and Decision-Making Under Uncertainty (Simons Institute, September 2016), Real-Time Decision Making 2018, Foundations of Machine Learning Boot Camp 2017, Adversarial Approaches in Machine Learning 2022]
who it's for
machine learning researchers, reinforcement learning researchers, bandit researchers, theoretical computer scientists, algorithm designers, economists and mechanism designers, graduate students in CS
topics
stochastic multi-armed bandits; adversarial multi-armed bandits; contextual linear bandits; ucb and Thompson sampling; regret bounds and concentration; Markov decision processes; td learning and q-learning; online reinforcement learning regret; competitive analysis of online algorithms; prophet inequalities and secretary problems; mechanism design with limited commitment; multicalibration for fair decision making
| title | author | length |
|---|---|---|
| Stochastic Bandits: Foundations and Current Perspectives | Shipra Agrawal | 01h 09m |
Covers concrete algorithms and guarantees: explains UCB (upper confidence bound) and Thompson Sampling, shows how UCB uses concentration bounds to get… | ||
| Stochastic Bandits: Foundations and Current Perspectives | Shipra Agrawal | 01h 01m |
Covers algorithms and precise guarantees for stochastic bandits with feature vectors: a linear contextual model where contexts are seen before choice, are… | ||
| Adversarial Bandits: Theory and Algorithms | Haipeng Luo | 01h 21m |
Covers adversarial multi-armed bandits with concrete algorithms and proofs: full-information FTRL/Hedge analysis, bandit reduction via importance-weighted loss… | ||
| Bridging Stochastic and Adversarial Bandits | Thodoris Lykouris | 01h 15m |
Single algorithm requirement: derive one policy that attains log(T)/Δ regret when rewards are IID and Θ(√(KT)) worst-case when adversarial. Concrete… | ||
| Fundamentals of Markov Decision Processes | Weina Wang | 57m |
Covers finite-state, finite-action discounted MDPs with gamma in (0,1); derives the Bellman equation and shows the Bellman operator is a contraction so the… | ||
| Foundations of RL | Rayadurgam Srikant | 01h 05m |
Covers concrete algorithms and finite-time claims: presents TD learning, Q‑Learning and Natural Policy Gradient and shows how stochastic‑approximation tools… | ||
| Foundations of RL | Rayadurgam Srikant | 01h 11m |
Covers three concrete algorithms: TD learning, Q‑learning, and Natural Policy Gradient, with claims of finite‑time bounds for linear function approximation… | ||
| Online Reinforcement Learning and Regret | Christina Lee Yu, Sean Sinclair | 01h 17m |
Tutorial compares three interaction models (fully known, generative, and an online episodic model) and gives concrete algorithms and regret bounds: model-based… | ||
| Competitive Analysis of Online Algorithms (Part 1) | Anupam Gupta | 56m |
Survey defines competitive ratio, uses rent-or-buy to show deterministic best policy buys at B days giving ratio ~2 and randomized optimal ratio is e/(e-1)… | ||
| Competitive Analysis of Online Algorithms (Part 2) | Anupam Gupta | 01h 04m |
Defines competitive analysis and proves concrete bounds: ski‑rental gets 2 - 1/B deterministically and ~e/(e-1) randomized as B→∞; cat‑and‑mouse is… | ||
| Competitive Analysis Meets Stochastic Input: Secretary Problems And Prophet Inequalities | Shuchi Chawla | 01h 14m |
Algorithms assume a partly adversarial, partly stochastic input and give concrete, constant-factor guarantees: a single threshold yields a 2-approx for… | ||
| Machine Learning For Algorithm Design | Ellen Vitercik | 01h 14m |
Survey of concrete claims and tools: performance of parameterized combinatorial algorithms is piecewise constant in parameters on each input, so learning… | ||
| (Sequential) Information Design | Laura Doval | 01h 02m |
Surveys the information‑design toolkit and gives concrete characterizations: distributions over actions and states must satisfy a martingale (prior) constraint… | ||
| Mechanism Design with Limited Commitment | Laura Doval | 59m |
A revelation‑principle analogue shows up for short‑term commitment settings: mechanisms can be written as direct Blackwell mechanisms that map types to… | ||
| Multicalibration: Towards Fair Decision Making | Michael Kim | 01h 21m |
Multicalibration requires predictions be calibrated not just overall but simultaneously over a computable class of subpopulations and supplies an algorithmic… | ||
| Multi-Distribution Learning, for Robustness, Fairness, and Collaboration | Nika Haghtalab | 01h 28m |
Argues that worst-case multi-distribution learning can avoid an O(K) blowup and often needs only log K dependence by recasting the problem as a min‑max game… | ||
| Blackwell’s Approachability Theorem | Rakesh Vohra | 57m |
Blackwell’s approachability theorem gives a concrete claim: if each stage’s vector outcome has expected inner product ≤0 with the outward normal at the… | ||
| Online Learning via Offline Greedy Algorithms: Applications in Market Design and Optimization | Negin Golrezaei | 43m |
Presents a concrete offline-to-online reduction that Blackwell approachability converts greedy, locally robust offline algorithms into online learners… | ||
| The Blackwell-Rao Theorem | Daniel Alabi | 34m |
The Rao‑Blackwell theorem claims that replacing an estimator by its conditional expectation given a sufficient statistic strictly lowers mean‑square error (and… | ||
| Understanding, Simplicity, And Decision Trees | Bin Yu | 35m |
Shows that CART-style trees suffer a provable gap on smooth additive models by proving a lower bound using information theory that decision trees need many… | ||
| Reflections of David Blackwell on the Work of David Blackwell | Bill Massey | 43m |
Survey of David Blackwell's proofs and legacies emphasizes concrete claims and tools: a simple stopping-time proof of Waldlike identities using the strong law… | ||
| David Blackwell: Teacher And Colleague And A Sample Of His Work | Peter Bickel | 41m |
Generalizes Von Neumann to vector payoffs and proves: any closed convex target set S is either approachable or excludable. Strategy is explicit and dynamic —… | ||
| nothing matches | ||