Likelihood Free Bayesian Optimization
Last edited: October 10, 2026High level motivation: black-box optimization
\begin{equation} x^{*} \in \arg\max_{x \in S} f\qty(x) \end{equation}
where \(x\) is some candidate design, \(f\) the objective, and generally we can only see evaluations of \(y = f\qty(x) + \varepsilon\).
Baysian Optimization
A basic method is to fit a Gaussian process; given:
\begin{equation} D_{t-1} = \qty {\qty(x_{i}, y_{i})}_{i=1}^{t-1} \end{equation}
The historical measurement we fit some kind of parametrized probability distribution, say, a guassian process:
\begin{equation} f\qty(x) \sim \mathcal{GP} \qty(u_{t}, k_{t}) \end{equation}
Machine Learning from Human Preferences Index
Last edited: October 10, 2026Lectures
query selection
Last edited: October 10, 2026Each query presents one item and records a binary response (did they like it?).
- Rasch: a person parameter \(\alpha\)
- Frozen encoder: reward-heads weights \(w\)
We don’t want to present too many queries to train this.
Fisher Information
Fisher recognizes fixed \(u\), and have expectation under different measures Y
New User, Known Object
Suppose you want to know the information gain from understanding \(\alpha\) the user parameter, conditioned on some known item you tested \(\beta\). Consider some random variable \(Y\) under some kind of Rasch model:
Shielding Safe AI Systems
Last edited: October 10, 2026Safe Leraning Problems
How do we make decision making agents abide by safety requirement at runtime? Given \(T\) and spec \(\phi\), find optimal \(\pi^{*}\) to solve \(T\) such that \(\pi^{*} \text{ SAT } \phi\) using some notion of satisfaction.
- What is the notion of safety?
- what are the requirements in temporal logic?
- How do we enforce compliance?
- How do we prove compliance?
why not just apply formal methods to AI?
- environments change…
- notion of safety for stochastic systems are random
- neural networks are not easily probable since they are approximations of what we are trying to model
MASA Library
MASA uses shielding as the key technique for ensuring safety.
Stanford UG Courses Index
Last edited: October 10, 2026Stanford UG Y1, Aut
Stanford UG Y1, Win
Stanford UG Y1, Spr
Stanford UG Y2, Aut
Stanford UG Y2, Win
Stanford UG Y2, Spr
Stanford UG Y3, Aut
Stanford UG Y3, Win
Stanford UG Y3, Spr
Stanford GR Y1, Aut
Stanford Talks
| Date | Topic | Presenter | Link |
|---|---|---|---|
| UG Research Program | Brian Thomas | Stanford UG Research Program | |
| Bld an Ecosystem, Not Monolith | Colin Raffel | Build a System | |
| Training Helpful CHatbots | Nazeen Rajani | Training Helpful Chatbots | |
| AI Intepretability for Bio | Gasper Begus | AI Intepretability | |
| PT Transformers on Long Seqs | Mike Lewis | Pretraining Long Transformers | |
| Transformers! | A. Vaswani | Transformers | |
| Towards Interactive Agents | Jessy Lin | Interactive Agent | |
| Dissociating Language and Thought | Anna Ivanova | Dissociating Language and Thought | |
| Language Agents | Karthik Narasimhan | Language Agents with Karthik | |
| Pretraining Data | |||
| value alignment | Been Kim | LM Alignment | |
| model editing | Peter Hase | Knowledge Editing | |
| Knowledge Localization | |||
| Presentations | Sydney Katz | Presentations | |
| Video Generation with Learned Prior | Meenakshi Sarkar | Priors | |
| Theoretical Drone Control | Sliding Mode UAV Control | ||
| VLM to Agents | Tao Yu | VLM to Agents | |
| Social RL | Natasha Jaques | Social Reinforcement Learning | |
| Model Predictive Control + Prompting | Gabriel Maher | LLM MPC | |
| Planning for Learning | |||
| Theorem Proving | Self-Play Conjection Generalization | ||
| Safety for Trucks | Safety for Autonomous Trucking | ||
| Collaborate Multiagent DM | Collaborative Multiagent DM | ||
| AI Safety Talks | AI Safety Annual Meeting | ||
| Pretraining under infinite compute | Limited Samples and Infinite Compute | ||
| Mel Krusniak | Decisions.jl | ||
| SISL Flash Talks | SISL Talks | ||
| Predicting Scaling Performance | |||
| mixed-autonomy traffic with LLMS | mixed-autonomy traffic with LLMs | ||
| AI Incidents Policy | AI Incidents Policy | ||
| Reliable RL | Reliable RL | ||
| Words to Concepts | Words to Concepts | ||
| Zen’s Defense | |||
| multi-agent LLM | Multi-Agent LLMs | ||
| Alex’s Defense | |||
| Yi Chen | gen ai safety model and datasets | ||
| Anastasia Koslova | From Security to Trustworthy AI | ||
| Problems with AI | |||
| Likelihood Free Bayesian Optimization |
