Don't let low-order statistics fool you

Virtually every ML paper is a method comparison, where the authors almost always have a "horse in the race". This is bad for science. We need to diversify and embrace empirical studies that are not incentivized to push a particular narrative or make a specific method look good.

Happy to share our recent work titled “Robust counterfactual explanations for neural networks with probabilistic guarantees” (to appear at ICML 2023) arxiv.org/abs/2305.11997

1/🧵New Research on Language Models! Language models (LMs) often "memorize" data, leading to privacy risks. This paper explores ways to reduce that! Paper: arxiv.org/pdf/2410.02159 Code: github.com/msakarvadia/me… Blog: mansisak.com/memorization/