will wolf

Lately, I've been diving into agentic RL for LLMs—a fascinating topic that sits neatly at the intersection of many of my technical interests.

Introduction For the past 12 months, I've been building a research platform for training AI algorithms to learn Monopoly Deal via self-play.

What is a Language Model?

Bayesian inference is the task of quantifying a posterior belief over parameters \(\boldsymbol{\theta}\) given observed data \(\mathbf{x}\)—where \(\mathbf{x}\) was generated from a model \(p(\mathbf{x}\mid{\boldsymbol{\theta}})\)—via Bayes' Theorem: $$ p…

To "explore vs. exploit" is a classic dilemma in decision-making.

I joined ASAPP in August 2017 as its 78th hire.

Halford J.

Nations exert influence in two key ways: hard power, and soft power.

Last week, Houthi rebels—a militant political group in Yemen—claimed credit for exploding an oil refinery in Saudi Arabia.

Machine learning and artificial intelligence have long been passions; to date, I've spent many years—my entire professional career included—developing relevant expertise.

"Mean-Field Variational Bayes" (MFVB), is similar to expectation-maximization (EM) yet distinct in two key ways: We do not minimize \(\text{KL}\big(q(\mathbf{Z})\Vert p(\mathbf{Z}\vert\mathbf{X}, \theta)\big)\), i.e. perform the E-step, as [in the problems

Consider a model with parameters \(\theta\) and latent variables \(\mathbf{Z}\); the expectation-maximization algorithm (EM) is a mechanism for computing the values of \(\theta\) that, under this model, maximize the likelihood of some observed data \(\mat…

In the previous post we introduced Boltzmann machines and the infeasibility of computing the gradient of its log-partition function \(\nabla\log{Z}\).

The principal task of machine learning is to fit a model to some data.

In the previous post, we covered the following topics: A Gaussian process (GP) defines a distribution over functions (i.e. function evaluations) Marginalizing a Gaussian over a subset of its elements gives another Gaussian (just pluck out the pieces of in…