arXiv subjects

Tracy Ke

Publications and source records attributed to Tracy Ke.

4 recordsLinked to original sources

arXiv2024

Recite, Reconstruct, Recollect: Memorization in LMs as a Multifaceted Phenomenon

Memorization in language models is typically treated as a homogenous phenomenon, neglecting the specifics of the memorized data. We instead model memorization as the effect of a set of complex factors that describe each sample and relate it to the model and corpus. To build intuition around these factors, we break memorization down into a taxonomy: recitation of highly duplicated sequences, reconstruction of inherently predictable sequences, and recollection of sequences that are neither. We demonstrate the usefulness of our taxonomy by using it to construct a predictive model for memorization. By analyzing dependencies and inspecting the weights of the predictive model, we find that different factors influence the likelihood of memorization differently depending on the taxonomic category.

cs.CL↗

arXiv2023

Semi-supervised Community Detection via Structural Similarity Metrics

Motivated by social network analysis and network-based recommendation systems, we study a semi-supervised community detection problem in which the objective is to estimate the community label of a new node using the network topology and partially observed community labels of existing nodes. The network is modeled using a degree-corrected stochastic block model, which allows for severe degree heterogeneity and potentially non-assortative communities. We propose an algorithm that computes a `structural similarity metric' between the new node and each of the $K$ communities by aggregating labeled and unlabeled data. The estimated label of the new node corresponds to the value of $k$ that maximizes this similarity metric. Our method is fast and numerically outperforms existing semi-supervised algorithms. Theoretically, we derive explicit bounds for the misclassification error and show the efficiency of our method by comparing it with an ideal classifier. Our findings highlight, to the best of our knowledge, the first semi-supervised community detection algorithm that offers theoretical guarantees.

cs.SI↗

arXiv2014

Rare and Weak effects in Large-Scale Inference: methods and phase diagrams

Often when we deal with `Big Data', the true effects we are interested in are Rare and Weak (RW). Researchers measure a large number of features, hoping to find perhaps only a small fraction of them to be relevant to the research in question; the effect sizes of the relevant features are individually small so the true effects are not strong enough to stand out for themselves. Higher Criticism (HC) and Graphlet Screening (GS) are two classes of methods that are specifically designed for the Rare/Weak settings. HC was introduced to determine whether there are any relevant effects in all the measured features. More recently, HC was applied to classification, where it provides a method for selecting useful predictive features for trained classification rules. GS was introduced as a graph-guided multivariate screening procedure, and was used for variable selection. We develop a theoretic framework where we use an Asymptotic Rare and Weak (ARW) model simultaneously controlling the size and prevalence of useful/significant features among the useless/null bulk. At the heart of the ARW model is the so-called phase diagram, which is a way to visualize clearly the class of ARW settings where the relevant effects are so rare or weak that desired goals (signal detection, variable selection, etc.) are simply impossible to achieve. We show that HC and GS have important advantages over better known procedures and achieve the optimal phase diagrams in a variety of ARW settings. HC and GS are flexible ideas that adapt easily to many interesting situations. We review the basics of these ideas and some of the recent extensions, discuss their connections to existing literature, and suggest some new applications of these ideas.

math.ST↗

arXiv2013

Homogeneity in Regression

This paper explores the homogeneity of coefficients in high-dimensional regression, which extends the sparsity concept and is more general and suitable for many applications. Homogeneity arises when one expects regression coefficients corresponding to neighboring geographical regions or a similar cluster of covariates to be approximately the same. Sparsity corresponds to a special case of homogeneity with a known atom zero. In this article, we propose a new method called clustering algorithm in regression via data-driven segmentation (CARDS) to explore homogeneity. New mathematics are provided on the gain that can be achieved by exploring homogeneity. Statistical properties of two versions of CARDS are analyzed. In particular, the asymptotic normality of our proposed CARDS estimator is established, which reveals better estimation accuracy for homogeneous parameters than that without homogeneity exploration. When our methods are combined with sparsity exploration, further efficiency can be achieved beyond the exploration of sparsity alone. This provides additional insights into the power of exploring low-dimensional strucuture in high-dimensional regression: homogeneity and sparsity. The newly developed method is further illustrated by simulation studies and applications to real data.

stat.ME↗