Channel
Digests from Stanford Online — Stanford Online technical and educational videos.
This lecture examines test-time compute scaling, showing how repeated sampling and advanced architectures like Arkon can boost model performance without retraining. It covers power laws and optimizing inference pipelines.
Read article
The lecture explores advanced LLM verification techniques, moving from simple outcome checks (GSM8K) to sophisticated process-based reward models (PRMs). Learn how ensembling weak verifiers can dramatically close the generation-verification gap.
Read article
Dive into train-time scaling and RL techniques that enable self-improving AI agents. Learn how STaR, DeepSeekMath, and DAPO boost reasoning in LLMs using closed feedback loops.
Read article
This seminar explores the interplay between text and data visualization, detailing how Multimodal LLMs process combined inputs using cross-attention mechanisms.
Read article
The lecture details advanced RL techniques for LLMs, covering Policy Gradient derivations and the stabilization mechanism of Proximal Policy Optimization (PPO). It shows how to model Chain-of-Thought prompting as an MDP problem.
Read article
This session covers Reinforcement Learning, detailing the Markov Decision Process (MDP) framework and the Policy Gradient algorithm. Key concepts include defining states, actions, transition dynamics, and using the Bellman equation for optimal value estimation.
Read article
The lecture reviews advanced Transformer optimizations, covering Grouped Query Attention (GQA) to reduce KV cache memory and Mixture of Experts (MoE) for scaling parameters while controlling compute. It also details In-Context Learning and SFT.
Read article
This deep dive covers LLMs, focusing on the Transformer architecture. Key topics include subword tokenization, auto-regressive modeling, and the self-attention mechanism using Q, K, V vectors.
Read article
This session covers Diffusion Models, detailing the loss function $\mathcal{L}_{t-1}$ derived from lower bounds. It also explores Foundation Model paradigms, zero-shot learning, and parameter efficiency techniques like Low-Rank Adaptation (LoRA).
Read article
Diffusion models are generative frameworks used to generate realistic images, videos, and actions. The process involves a fixed forward noising step ($X_0 o X_T$) and a learned reverse denoising step $p_ heta(X_{t-1}|X_t)$, optimized using the Evidence Lower Bound (ELBO).
Read article
This lecture covers GMM via EM and PCA. It details how EM uses Jensen's inequality to construct a lower bound for optimization, while PCA finds maximal variance directions using Eigen Decomposition of the covariance matrix.
Read article
This lecture covers unsupervised learning, contrasting K-Means clustering with its probabilistic extension, Gaussian Mixture Models (GMM). The core technique for solving GMM is the Expectation-Maximization (EM) algorithm, which relies on convex relaxation and Jensen's inequality.
Read article
This lecture explores Automatic Differentiation, proving that the gradient computation (backward pass) in neural networks is efficient. By viewing networks as differentiable circuits, backpropagation can be computed in time proportional to the number of parameters.
Read article
Master the Bias-Variance Tradeoff: Learn how to select robust ML models by balancing model complexity, regularization techniques like Ridge Regression, and advanced methods such as Hyperband for efficient hyperparameter tuning.
Read article
The Exponential Family unifies diverse statistical models—from Gaussian errors to Softmax classification—under one common mathematical form. This framework shows that core ML procedures like inference and learning are fundamentally consistent across all these types of data.
Read article
The lecture establishes Maximum Likelihood Estimation (MLE) as a foundational principle to generalize modeling from continuous regression to discrete classification. It derives Logistic Regression and compares optimization techniques like SGD vs. Newton's method.
Read article
The core of supervised learning is minimizing error using $\text{L}_2$ loss. Optimization relies on Gradient Descent, which scales to massive datasets using Stochastic Gradient Descent (SGD) and mini-batching.
Read article
The course emphasizes understanding the mathematical foundations of ML, covering Supervised, Unsupervised, and Reinforcement Learning. A key focus is optimizing algorithms for hardware compatibility.
Read article
AI is fundamentally changing computing economics, moving from low-cost software distribution to compute-intensive inference. Learn how architectural breakthroughs and specialized hardware like Groq are driving a parabolic explosion in token generation capacity.
Read article
The talk details designing LLM coaches that promote user agency by eliciting qualitative context and navigating uncertainty. Techniques include zero-shot Bayesian Adaptive Planning.
Read article