Stanford Online

Jul 31, 2026 · 1:20:06

Stanford CS229 Machine Learning | Spring 2026 | Lecture 10: GMM (EM), PCA

This lecture provides a deep dive into two foundational unsupervised machine learning techniques: Gaussian Mixture Models (GMM) solved via the Expectation-Maximization (EM) algorithm, and Principal Component Analysis (PCA). The discussion emphasizes the mathematical underpinnings of these methods. For GMM, the EM algorithm is shown to solve for latent variables by constructing a tractable lower bound using Jensen's inequality. For PCA, the technique identifies directions of maximum variance by performing Eigen Decomposition on the data's covariance matrix, highlighting the critical need for proper data centering and scaling.

Key takeaways

1. GMM via EM Algorithm 23:20

   The Expectation-Maximization (EM) algorithm is used to estimate parameters ($ heta$) for GMM. It operates by constructing a lower bound $L( heta|T)$ of the log-likelihood, which allows optimization through alternating steps: E-step (calculating soft assignments/probabilities $Q_i(Z)$) and M-step (re-estimating parameters $oldsymbol{ heta}$).

2. PCA Core Principle 1:03:20

   PCA is a dimensionality reduction technique that finds orthogonal unit vectors (principal components, $U_k$) corresponding to the directions of maximal variance in the data. The process relies on finding the Eigen Decomposition ($oldsymbol{A} = oldsymbol{U} oldsymbol{ ext{diag}}(oldsymbol{ ext{eigenvalues}}) oldsymbol{U}^T$) of the covariance matrix.

3. Mathematical Rigor (Jensen's Inequality) 27:10

   The EM algorithm leverages Jensen's inequality to transform an intractable expectation into a manageable lower bound, enabling iterative optimization. The selection of the conditional probability $Q(Z)$ is key to achieving this tight lower bound.

• Stanford Online
• Gaussian Mixture Models (GMM)
• Expectation-Maximization (EM) Algorithm
• Principal Component Analysis (PCA)
• Dimensionality Reduction
• Latent Variables
• Eigen Decomposition
• Stanford Online
• CS229 Machine Learning Course Website

Watch on YouTube Full article

Jul 31, 2026 · 1:16:30

Stanford CS229 Machine Learning | Spring 2026 | Lecture 9: K-Means and GMM (non-EM)

This lecture provides a deep dive into unsupervised machine learning algorithms, focusing on K-Means clustering and its probabilistic extension, Gaussian Mixture Models (GMM). The core mathematical framework for solving GMM is the Expectation-Maximization (EM) algorithm. Key concepts include understanding how to model structure without explicit labels, utilizing latent variables, and employing convex analysis via Jensen's inequality to derive the iterative estimation procedure.

Key takeaways

1. Unsupervised vs. Supervised Learning 2:00

   Unlike supervised learning (where labels define separation), unsupervised learning aims to model inherent structure or clusters within unlabeled data, making it a fundamentally more challenging problem that requires stronger assumptions and accepting weaker guarantees.

2. K-Means Clustering 5:50

   K-Means is an iterative algorithm where points are assigned to the nearest cluster center ($oldsymbol{ ext{mu}}_i$). The process involves two steps: (1) assigning each point to its closest $oldsymbol{ ext{mu}}$, and (2) recalculating the new cluster centers based on the arithmetic mean of all assigned points. Finding the optimal clustering is NP-hard, meaning initialization matters.

3. Gaussian Mixture Models (GMM) 10:50

   GMMs are a probabilistic relaxation of K-Means, modeling data as mixtures of Gaussian distributions. Instead of hard assignments, points are assigned probabilities to belong to each source/cluster. The model is defined by means ($oldsymbol{ ext{mu}}$), covariances ($oldsymbol{ ext{Sigma}}$), and mixing proportions ($oldsymbol{f}$).

4. Expectation-Maximization (EM) Algorithm 17:30

   The EM algorithm is used to estimate the parameters of latent variable models like GMM. It alternates between two steps: the E-step (calculating the probability $W_{ij}$ that each point belongs to each source given current parameter estimates) and the M-step (re-estimating all model parameters, including means and covariances, based on these probabilities).

5. Convexity and Jensen's Inequality 30:00

   The EM algorithm relies on convex analysis. Convex functions are those where the line segment connecting any two points lies above the function graph (e.g., $x^2$). Jensen's inequality is used to derive a lower bound for the log-likelihood function, allowing the complex optimization problem to be solved iteratively.

• Stanford Online
• Unsupervised Learning
• Clustering Algorithms
• Gaussian Mixture Models (GMM)
• Expectation-Maximization (EM) Algorithm
• Convex Optimization
• Latent Variables
• Maximum Likelihood Estimation
• Stanford CS229 Machine Learning Course Website
• Stanford AI Programs Information

Watch on YouTube Full article

Jul 30, 2026 · 1:02:13

Stanford CS229 Machine Learning | Spring 2026 | Lecture 8: Neural Networks 2 (Backprop)

This lecture provides a deep theoretical dive into Backpropagation and Automatic Differentiation, establishing it as an efficient method for computing gradients in complex neural networks. The core principle is that any differentiable network can be viewed as a 'differentiable circuit' or computational graph. This allows the gradient (the backward pass) to be computed with a time complexity proportional to the number of parameters ($O(N)$), matching the efficiency of the forward pass, regardless of how complex the underlying function is.

Key takeaways

1. Automatic Differentiation and Computational Graphs 2:00

   The gradient computation (backward pass) can be viewed as traversing a differentiable circuit. This method allows for efficient calculation because it reuses intermediate results, unlike expanding the function into a traditional mathematical formula.

2. Efficiency of Gradient Computation 2:40

   The fundamental theorem states that if a differentiable circuit of size $N$ computes a real-valued function, its gradient can also be computed in time complexity $O(N)$. This means the forward pass (evaluating loss) and the backward pass (calculating gradients) have similar computational costs.

3. Chain Rule Application for Backprop 5:50

   Backpropagation is an application of the Chain Rule. By knowing the gradient with respect to an intermediate variable ($U$), one can compute the gradient with respect to its input ($Z$) using matrix multiplication involving the Jacobian (or transpose of the Jacobian). This process allows computation to proceed layer by layer.

• Stanford Online
• Automatic Differentiation
• Backpropagation
• Neural Networks
• Computational Graphs
• Chain Rule
• Optimization Theory
• Stanford CS229 Machine Learning
• Stanford AI Programs Information

Watch on YouTube Full article

Jul 30, 2026 · 1:18:27

Stanford CS229 Machine Learning | Spring 2026 | Lecture 6: Dataset Split, ML Advice

This lecture provides a deep dive into the fundamental challenge of machine learning: generalization. It systematically analyzes how models trained on finite, noisy samples can be selected to perform reliably on unseen data. Key concepts include decomposing test error into bias (model class limitation), variance (sensitivity to training data), and noise (measurement error). The discussion covers classical techniques like regularization (e.g., Ridge Regression) for reducing variance at the cost of slight bias, modern phenomena like Double Descent in overparameterized models, and practical model selection methods such as K-fold cross validation and Hyperband.

Key takeaways

1. Bias-Variance Decomposition 20:05

   Test error is decomposed into three components: noise (intrinsic data error), squared bias (how far the average prediction is from the true function, dependent on model class), and variance (how much the prediction jumps around across different training sets). Minimizing test error requires balancing these three sources of error.

2. Regularization as Variance Reduction 30:05

   Regularization techniques, such as adding an $L_2$ penalty (Ridge Regression), constrain the model weights ($ heta$) to prevent them from becoming too large. This stabilizes the model, significantly reducing variance while accepting a small increase in bias.

3. Modern ML Phenomena: Double Descent 1:03:25

   The classical Bias-Variance curve suggests that test error must rise after a certain model complexity threshold. However, modern models can exhibit 'Double Descent,' where the test error decreases again in the massively overparameterized regime (i.e., having more parameters than data points).

4. Model Selection and Hyperparameter Tuning 50:50

   To prevent information leakage from the test set, techniques like K-fold cross validation are used. For compute efficiency in tuning hyperparameters (e.g., regularization strength $ ho$), algorithms like Hyperband efficiently allocate computational resources to promising model configurations.

• Stanford Online
• Bias-Variance Tradeoff
• Regularization (L1/L2)
• Model Selection
• Cross-Validation
• Hyperparameter Optimization
• Double Descent
• Overfitting and Underfitting
• Stanford CS229 Machine Learning
• General AI Information

Watch on YouTube Full article

Jul 29, 2026 · 1:14:12

Stanford CS229 Machine Learning | Spring 2026 | Lecture 4: Exponential Family, GLMs classification

The lecture introduces the Exponential Family as a unifying framework for numerous statistical models, including linear regression, logistic regression, Gaussian error modeling, and multinomial classification (Softmax). By showing that these diverse distributions share a common mathematical form, the theory demonstrates that core machine learning procedures—such as inference (calculating expected values) and parameter estimation (learning)—can be generalized across all members of this family. This foundational understanding is critical for modern AI architectures, particularly those utilizing Softmax in large language models.

Key takeaways

1. The Exponential Family Unification 2:00

   Many common distributions (Bernoulli, Gaussian, Multinomial) can be written into a single functional form. This allows for the generalization of model building and analysis across different data types.

2. Inference and Learning are Generalized 3:30

   The structure of the exponential family ensures that key statistical operations, such as computing the expected value (inference) and performing maximum likelihood estimation (learning), follow a consistent mathematical pattern regardless of the specific distribution used.

3. Softmax in Multiclass Classification 6:10

   The Softmax function is presented as the standard mechanism for multiclass classification, allowing prediction over $K$ discrete values. Mathematically, it fits within the exponential family structure and generalizes logistic regression (the two-class case).

4. Generalized Linear Models (GLMs) 10:50

   The GLM framework ties the abstract error models to practical supervised learning. The model predicts a distribution over $Y$ given features $X$, and the loss function is derived directly from maximizing the log-likelihood of this distribution.

• Stanford Online
• Exponential Family
• Generalized Linear Models (GLMs)
• Softmax Activation
• Multiclass Classification
• Maximum Likelihood Estimation (MLE)
• Cross-Entropy Loss
• Stanford CS229 Machine Learning
• Stanford AI Programs Information

Watch on YouTube Full article

Jul 29, 2026 · 1:02:14

Stanford CS229 Machine Learning | Spring 2026 | Lecture 3: Weighted Least Squares

This lecture provides a deep theoretical dive into classification and model building using the Maximum Likelihood Estimation (MLE) framework. The core principle demonstrated is that MLE allows generalizing modeling techniques from continuous regression (Least Squares) to discrete classification problems by introducing probabilistic interpretations. Key derivations include establishing Logistic Regression via the log-likelihood function, which transforms the complex product of probabilities into a numerically stable additive sum suitable for optimization using Stochastic Gradient Descent (SGD). Furthermore, the lecture compares various optimization methods—Gradient Descent, SGD, and Newton's Method—highlighting that while Newton's method is theoretically efficient, SGD remains the workhorse due to its scalability with massive datasets ($N$ and $D$).

Key takeaways

1. Maximum Likelihood Principle (MLE) 1:35

   The MLE framework dictates setting up a probabilistic model (a forward model) of how data is generated. This principle generalizes across different problem types (continuous, discrete, etc.) by maximizing the likelihood function, which measures how probable the observed data is under a given set of parameters ($ heta$).

2. Logistic Regression Derivation 5:50

   Classification problems are modeled using a link function (e.g., the sigmoid function) applied to a linear combination of features ($ ext{logit}(p)$). By applying MLE, the resulting log-likelihood function for binary classification leads directly to the standard form used in logistic regression.

3. Optimization Method Comparison 17:50

   While Newton's method offers fast convergence and high precision, Stochastic Gradient Descent (SGD) is preferred in modern Machine Learning because it scales efficiently when dealing with massive datasets ($N$ and $D$), making it computationally feasible where full gradient computation is impossible.

• Stanford Online
• Maximum Likelihood Estimation
• Classification Theory
• Logistic Regression
• Gaussian Distributions
• Optimization Methods (SGD, Newton's Method)
• Link Functions
• Stanford Online
• CS229 Machine Learning Course Website

Watch on YouTube Full article

Jul 29, 2026 · 1:18:10

Stanford CS229 Machine Learning | Spring 2026 | Lecture 2: Supervised Learning Setup

This lecture establishes the foundational mathematical framework for supervised machine learning, focusing on linear regression as a canonical example. Key concepts include defining a hypothesis function ($h: X o Y$), minimizing error using $ ext{L}_2$ loss (Mean Squared Error), and implementing optimization via Gradient Descent. The discussion emphasizes that modern ML success relies heavily on scaling models and computational efficiency, particularly through techniques like mini-batching and understanding the relationship between batch size and generalization.

Key takeaways

1. Supervised Learning Setup 17:12

   Supervised learning requires a training set of paired examples $(X_i, Y_i)$, where $X$ is the input feature space (e.g., images, house data) and $Y$ is the corresponding label or value. The goal is to learn a hypothesis function $h$ that generalizes well to unseen data.

2. Regression vs. Classification 22:00

   If the output $Y$ is a continuous real number (e.g., house price), it is a regression problem. If $Y$ is discrete or categorical (e.g., cat/dog label, positive/negative review), it is a classification problem.

3. Optimization via $ ext{L}_2$ Loss 30:40

   The model parameters ($ heta$) are chosen by minimizing the empirical risk, typically using the squared error (the $ ext{L}_2$ norm). The objective function $J( heta)$ is minimized to find the best fit for all training points.

4. Stochastic Gradient Descent (SGD) 59:50

   To handle massive datasets, optimization uses SGD, which updates parameters by calculating gradients over small subsets of data called mini-batches. This approach is computationally efficient and critical for scaling large models.

• Stanford Online
• Supervised Learning
• Linear Regression
• Gradient Descent
• Stochastic Gradient Descent (SGD)
• Loss Functions ($ ext{L}_2$)
• Mini-Batching
• Machine Learning Theory
• CS229 Machine Learning Course Notes
• Stanford TA Sessions

Watch on YouTube Full article

Jul 29, 2026 · 36:59

Stanford CS229 Machine Learning | Spring 2026 | Lecture 1: Introduction

This lecture provides a high-level introduction to Machine Learning fundamentals (Supervised, Unsupervised, and Reinforcement Learning) within the context of modern AI. The course emphasizes understanding the mathematical foundations and core techniques behind algorithms rather than focusing on programming implementation. Key topics include model training using large datasets, advanced concepts like embeddings, and the architectural differences between traditional ML tasks and general-purpose Large Language Models (LLMs). A critical focus for system builders is placed on the necessity of optimizing ML systems for hardware compatibility and speed.

Key takeaways

1. ML Paradigm Shift 6:11

   The field has moved from specific, task-oriented models to general-purpose agents (LLMs). While traditional methods still apply, the focus is on tuning fundamental model capabilities rather than building complex data pipelines for every single use case.

2. The Importance of ML Systems 33:00

   A planned lecture will cover 'ML system,' which addresses the critical need to make software and hardware compatible. Optimizing algorithms for speed (e.g., making them run 2x faster) is crucial due to the high cost associated with AI computation.

3. Reinforcement Learning (RL) in LLMs 24:50

   RL can be used to train models dealing with stochastic sampling, such as text generation. Techniques like Policy Gradient and using human feedback (e.g., RLHF/RAG) are necessary because the generation process is not differentiable.

• Stanford Online
• Machine Learning Fundamentals
• Supervised Learning (SFT)
• Unsupervised Learning (Clustering, LSA)
• Reinforcement Learning (RL)
• Large Language Models (LLMs)
• Diffusion Models
• ML System Optimization
• CS229 Machine Learning
• Stanford AI Programs Info

Watch on YouTube Full article

Jul 23, 2026 · 56:01

Stanford MS&E435 Economics of the AI Supercycle | Spring 2026 | The GPU Economy

The video provides an in-depth analysis of the economic and technical shifts driven by AI, arguing that unlike previous software cycles with near-zero distribution costs, modern AI requires massive compute resources. The discussion highlights how the shift from pre-training to inference time reasoning is causing a parabolic explosion in token consumption. Hardware innovation (e.g., Groq's architecture) and architectural breakthroughs—such as decoupling prefill and decode stages and utilizing high-bandwidth SRAMM—are critical for maintaining efficiency, leading to an expected deflationary trend in the unit cost of intelligence.

Key takeaways

1. AI Compute is Not Zero Marginal Cost

   Unlike previous software where distribution costs were near zero, AI applications require significant compute power. The increasing demand for tokens means that computing resources are a primary economic constraint and driver of value.

2. Inference Time Reasoning is the New Frontier 34:33

   The industry is shifting focus from pre-training models to inference time reasoning. This shift dramatically increases token consumption, with predictions suggesting a potential 1 billionx increase in required compute cycles.

3. Architectural Innovation Drives Efficiency 38:25

   Efficiency gains are achieved by architectural breakthroughs, such as Groq's design which utilizes high-bandwidth SRAMM and a deterministic compiler. Combining different systems (e.g., NVLink Fusion) allows for significantly higher token output per unit of power.

4. The Value Proposition is Democratizing Intelligence 50:15

   AI's value lies in democratizing access to high-level capabilities (e.g., specialized tutoring, concierge medicine), making previously exclusive functions available globally. The economic shift suggests that the unit cost of intelligence will continue to plummet.

• Stanford Online
• Artificial Intelligence (AI)
• Compute Economics
• Hardware Acceleration
• System Architecture
• Digital Transformation
• Stanford Online
• MSE435 Course Schedule

Watch on YouTube Full article

Jul 22, 2026 · 44:21

Stanford CS547 HCI Seminar | Spring 2026 | Promoting Agency in Human-AI Interaction

This seminar explores designing AI systems that promote user agency when LLMs are used as personal advisers (coaches/counselors), rather than mere assistants. The core thesis is that successful health behavior change requires systems to elicit qualitative context and navigate uncertainty to provide non-prescriptive support. Practical implementations, such as the GPT coach chatbot and the Bloom iOS application, demonstrate how integrating motivational interviewing strategies with wearable data can improve user mindset and sense of control. Algorithmically, the work proposes 'zero-shot Bayesian Adaptive Planning' using LLMs to strategically balance asking informative questions versus acting on known information by modeling latent uncertainty over the user's state.

Key takeaways

1. Shift from Assistant to Adviser 2:00

   LLM usage is shifting toward deeply personal advice (e.g., health, relationships), requiring a design paradigm that augments the user rather than automating tasks. This necessitates non-prescriptive support.

2. The Role of Qualitative Context 4:00

   Effective coaching relies on eliciting qualitative context (goals, values, motivations) over quantitative data (step count, heart rate). This aligns with principles from Motivational Interviewing.

3. GPT coach Implementation 8:10

   The GPT coach chatbot uses three prompt chains—Dialogue State Chain, Motivational Interviewing Chain, and Tool Use Chain—to integrate qualitative coaching strategies with quantitative data from the Apple Health Kit API.

4. Algorithmic Solution: Bayesian Adaptive Planning 17:55

   To improve strategic decision-making, the proposed algorithm uses Reinforcement Learning (RL) theory to model latent uncertainty ($ heta$) over the user's state. The agent must learn to strategically trade off asking informative questions against acting on current knowledge.

• Stanford Online
• Human-AI Interaction
• Behavior Change Theory
• Large Language Models (LLMs)
• Reinforcement Learning (RL)
• Bayesian Inference
• Health Technology
• Apple Health Kit API
• Stanford HCI Seminar Schedule

Watch on YouTube Full article

Jul 17, 2026 · 34:23

Stanford MS&E435 Economics of the AI Supercycle | Spring 2026 | Economics of Generative AI

This lecture analyzes the economic structure of the Generative AI supercycle, arguing that the industry's value accrual is highly concentrated in the hardware and infrastructure layers (Semis). The speaker introduces a model—an inverted triangle—to contrast the current AI ecosystem with previous tech cycles (Internet, Mobile, Cloud). Key insights focus on the shift from software-driven marginal cost near zero to an inference workload that requires significant GPU burn. The lecture emphasizes understanding hyperscaler CapEx guidance and the competitive dynamics between training and inference workloads.

Key takeaways

1. AI Value Accrual is Concentrated in Semis 23:49

   The most profitable part of the stack, by a wide margin, is the Semiconductors (Semis) layer. The speaker notes that while Application Layer revenues are estimated between 0% and 30% gross margin, data center revenues from chip providers like NVIDIA can reach around 75% gross margin.

2. The AI Ecosystem is Modeled as an Inverted Triangle 18:16

   Unlike the Cloud ecosystem, which followed a pyramid shape, the current AI market structure is modeled as an inverted triangle. This suggests that value creation is currently bottlenecked by hardware capacity and compute power.

3. Inference vs. Training Workloads 26:12

   The economics are shifting toward inference, which involves burst usage (unlike the predictable high utilization of training). The speaker notes that understanding the share of inference in a hyperscaler's fleet is critical for predicting future market dynamics.

4. Monetization Challenge for Consumer AI 30:42

   The primary economic challenge for consumer AI applications (like ChatGPT and Gemini) is scaling monetization. The current model shows a low revenue per user ($10/user/year for ChatGPT), necessitating a shift toward ad-based models or achieving mandatory utility status to reach the $100/user/year mark.

• Stanford Online
• Generative AI Economics
• Semiconductor Market Analysis
• Tech Supercycles
• Cloud Computing Infrastructure
• Profitability Modeling (AI)
• Stanford Online Graduate Programs
• MSE435 Course Schedule

Watch on YouTube Full article

Jul 17, 2026 · 49:13

Stanford MS&E435 Economics of the AI Supercycle | Spring 2026 | Applications, AI in Life Sciences

The discussion details how Artificial Intelligence is poised to fundamentally transform drug discovery and life sciences R&D by creating an end-to-end acceleration platform. Speakers from Anthropic and Chai argue that AI models (like Claude and specialized foundation models) can dramatically compress the current 10–15 year timeline for drug development, addressing bottlenecks in target identification, molecular design, clinical trials, and regulatory processes. The value is shifting from merely selling drugs to building scalable, integrated AI tools and platforms.

Key takeaways

1. AI Accelerates Drug Development Timelines 23:49

   The current median time for drug development (from idea to market) is 10–15 years. AI has the potential to compress this timeline, with estimates suggesting a reduction to the five-year range or less by optimizing preclinical and clinical phases.

2. Value Shifts from Drugs to Tools 36:00

   The industry value is expected to shift from traditional drug sales (revenue stream) to the tools, platforms, and foundational models that enable discovery. This makes tool developers highly valuable.

3. AI's Role in Molecular Design 17:22

   Companies are building Computer-Aided Design (CAD) suites for molecules, aiming for 'zero shot' drug design—the ability to generate patient-ready molecules directly from the computer, bypassing much of the traditional trial-and-error process.

4. The Platform Approach 20:40

   Anthropic's vision is to train Claude for end-to-end life science R&D acceleration, covering basic research, drug development, clinical trials, and regulatory strategy (e.g., designing clinical protocols).

• Stanford Online
• Artificial Intelligence (AI)
• Drug Discovery
• Life Sciences Technology
• Large Language Models (LLMs)
• Platform Engineering
• Anthropic
• Chai
• Benchling

Watch on YouTube Full article