Tutorial 5.2: Tomer Ullman - Church Programming Language Part 2

TL;DR

Probabilistic programming allows for planning, inference, and understanding implicatures by using generative models and sampling techniques.

Transcript

The following content is provided under a Creative Commons license. Your support will help MIT OpenCourseWare continue to offer high quality educational resources for free. To make a donation or view additional materials from hundreds of MIT courses, visit MIT OpenCourseWare at ocw.mit.edu. TOMER ULLMAN: So so far, we've talked about just examples ... Read More

Key Insights

• 👻 Probabilistic programming allows for planning and inference by conditioning generative models on observed data or goals.
• ❓ Sampling and rejection methods can be used to explore different hypotheses and estimate posterior distributions.

Questions & Answers

Q: How does rejection query work?

Rejection query involves sampling from a simple distribution and checking if it satisfies the desired condition. If it does, the sample is accepted; otherwise, it is rejected and a new sample is drawn.

Q: What is Metropolis Hastings and how does it relate to probabilistic programming?

Metropolis Hastings is an algorithm used for Markov Chain Monte Carlo (MCMC) sampling, which is a common technique in probabilistic programming. MCMC methods are used to sample from complex posterior distributions when direct sampling is not possible.

Q: What is implicature and how can it be modeled using probabilistic programming?

Implicature is a pragmatic inference that a listener makes based on what a speaker says. It can be modeled using probabilistic programming by defining a generative model of the speaker and listener and reasoning about their goals, beliefs, and actions.

Summary & Key Takeaways

• Probabilistic programming involves writing generative models that describe how we think the world works.

• Inference and planning can be done by conditioning the generative model on observed data or goals, and sampling from the posterior distribution.

• Implicature can also be modeled using probabilistic programming, where the speaker and listener reason about each other's intentions and beliefs.