Anay Mehrotra

Existing matrix completion methods are too blunt to estimate individual treatment effects, but this new estimator provides sharp, row-wise guarantees that unlock heterogeneous treatment effect estimation.

Even without remembering everything, you can still generate infinite languages, but hitting density targets or identifying the right language becomes much harder.

Forget uniform resampling – targeting high-entropy decision points in reasoning traces unlocks better performance than RL-trained models, all without extra training data or compute.

Differential privacy imposes fundamental limits on language *identification*, even when it doesn't preclude language *generation*, revealing a surprising divergence in their privacy costs.

Solved: the precise conditions for efficiently estimating a Gaussian mean from coarse, partitioned data, closing a key gap in our understanding of learning from limited information.