Stanford Online

Stanford Online

This lecture explains how Bayesian networks compactly represent joint probability distributions through local conditional probabilities, then contrasts inefficient rejection sampling with Gibbs sampling—an MCMC method that iteratively modifies existing samples to satisfy evidence, enabling efficient approximate inference even with rare events.

Stanford Online

This lecture transitions from model-free and model-based reinforcement learning to probabilistic reasoning, introducing Bayesian networks as a framework for representing uncertain world states. It establishes probability fundamentals—joint distributions, marginalization, and conditioning—using tensor operations (einops) to provide the mathematical foundation for efficient inference in complex domains.

Stanford Online

Veteran AI researcher Terry Winograd argues that rather than focusing on apocalyptic futures or utopian promises, we should recognize AI as an accelerant of existing social problems—particularly in employment, resource allocation, and information integrity—that demand immediate societal attention.

Stanford Online

Rio Wang introduces 𝚿0 (Psi-0), an open foundation model for universal humanoid locomotion-manipulation that moves beyond fixed 18-DoF manipulation systems by leveraging scalable egocentric human data collection and unified whole-body control to address the integration challenges of mobility, dexterity, and reasoning.

Stanford Online

Andrew Ng explains Principal Component Analysis (PCA) as a classic non-probabilistic algorithm for linear dimensionality reduction that identifies principal axes of variation through eigenvector decomposition of the covariance matrix, while cautioning that practitioners often apply it unnecessarily.

Stanford Online

This lecture covers approximate offline methods for solving POMDPs when exact solutions are computationally impossible, focusing on QMDP—a technique that solves the problem as a fully observable MDP and executes policies by weighting pre-computed Q-values with current belief states represented as alpha vectors.

Stanford Online

This seminar argues that GenAI's effectiveness in education hinges on 'knowledge engineering'—the systematic mapping of expert cognitive processes—to ensure high-fidelity, personalized feedback. A randomized trial demonstrates that TAs using AI suggestions based on detailed reasoning rubrics produced significantly better student learning outcomes than human-only feedback.

Stanford Online

Despite AI solving complex closed systems like chess decades ago, autonomous driving remains unsolved due to the 'open world' problem of unbounded physical complexity. This creates fundamental gaps in physical reasoning and safety validation that current foundation models struggle to overcome, requiring new comparative metrics to measure real-world reliability.

Stanford Online

This seminar challenges the traditional linear model of 'informal-to-formal' notation development, arguing that humans dynamically create new notations through collaborative practice while current AI systems are limited to 'instant formalization' into existing structures. The speaker presents a three-stage historical model of notation evolution—from culturally situated invention through community dispersion to institutional sanctification—to guide future HCI system design.

Stanford Online

Stanford professor James Landay argues that truly human-centered AI requires expanding beyond traditional user-centered design to address community and societal-level impacts, while tackling embedded Western cultural biases and the global rise of sovereign AI initiatives.

Stanford Online

Former EY Global Chief Innovation Officer Jeff Wong argues that traditional industries must abandon generational planning cycles for portfolio-based AI innovation strategies, leveraging vast proprietary data assets while requiring leaders to personally adopt the tools they seek to implement.

Stanford Online

This Stanford seminar presents advanced autonomous navigation systems using Gaussian splatting and vision language models to solve traversability and social compliance in outdoor environments, while introducing a paradigm shift toward companion robots that assist aging populations with mobility and health monitoring.