From the earliest years of our lives, humans use language to express our beliefs and desires. Being able to talk to artificial agents about our preferences would thus fulfill a central goal of value alignment. Y et today, we lack computational models explaining such language use.

In particular, we present a novel "squeeze-and-span" technique to distill knowledge from a generator

Unfortunately, Sparse VI and PSVI rely on access to exact samples from the model posterior and closed form gradients of it to be evaluated and updated.

In the upper-level, the fair predictor is updated to be close to all subgroup specific predictors. We further prove that such a bilevel objective can effectively control the group sufficiency and generalization error. We evaluate the proposed framework on real-world datasets.

We theoretically show that our self-supervised strategy has a provable ability to simulate test-agnostic class distributions.

One common answer is that an agent should be rewarded for attaining "novel" states in the environment, but naive measures of novelty have

The standard metric to quantify an online learning algorithm's performance is the regret.

Since weak learners perform only marginally better than random guesses, such subroutines constitute a weaker assumption than the availability of an accurate supervised learning oracle.