Yuetian Luo
Rutgers
Room: M3 3127
Algorithmic Stability in Statistical Inference
Algorithmic stability is a central concept in statistics and learning theory that measures how sensitive an algorithm's output is to small changes in the training data. In this talk, we consider two different perspectives of algorithmic stability in statistical inference. In the first part, we investigate the problem of distribution-free algorithm risk evaluation, uncovering fundamental limitations for answering these questions with limited amounts of data. To navigate the challenge, we will discuss how incorporating an assumption about algorithmic stability might help. In this second part, we reveal that while stability entails desirable properties, it is typically not sufficient on its own for statistical learning. For example, an algorithm that always outputs a constant function is perfectly stable but statistically meaningless. Then we will discuss the trade-offs between stability and accuracy in statistical inference.