arXiv subjects

Eric Xia

Publications and source records attributed to Eric Xia.

10 recordsLinked to original sources

arXiv2026

Classification Imbalance as Transfer Learning

Classification imbalance arises when one class is much rarer than the other. We frame this setting as transfer learning under label (prior) shift between an imbalanced source distribution induced by the observed data and a balanced target distribution under which performance is evaluated. Within this framework, we study a family of oversampling procedures that augment the training data by generating synthetic samples from an estimated minority-class distribution to roughly balance the classes, among which the celebrated SMOTE algorithm is a canonical example. We show that the excess risk decomposes into the rate achievable under balanced training (as if the data had been drawn from the balanced target distribution) and an additional term, the cost of transfer, which quantifies the discrepancy between the estimated and true minority-class distributions. In particular, we show that the cost of transfer for SMOTE dominates that of bootstrapping (random oversampling) in moderately high dimensions, suggesting that we should expect bootstrapping to have better performance than SMOTE in general. We corroborate these findings with experimental evidence. More broadly, our results provide guidance for choosing among augmentation strategies for imbalanced classification.

stat.ML↗

arXiv2025

Task Matrices: Linear Maps for Cross-Model Finetuning Transfer

Results in interpretability suggest that large vision and language models learn implicit linear encodings when models are biased by in-context prompting. However, the existence of similar linear representations in more general adaptation regimes has not yet been demonstrated. In this work, we develop the concept of a task matrix, a linear transformation from a base to finetuned embedding state. We demonstrate that for vision and text models and ten different datasets, a base model augmented with a task matrix achieves results surpassing linear probes, sometimes approaching finetuned levels. Our results validate the existence of cross-layer linear encodings between pretrained and finetuned architectures. Moreover, we show that a data-based approximation for such encodings is both efficient and generalizable to multiple domains. We make our implementation publicly available.

cs.LG↗

arXiv2025

Linear Relational Decoding of Morphology in Language Models

A two-part affine approximation has been found to be a good approximation for transformer computations over certain subject object relations. Adapting the Bigger Analogy Test Set, we show that the linear transformation Ws, where s is a middle layer representation of a subject token and W is derived from model derivatives, is also able to accurately reproduce final object states for many relations. This linear technique is able to achieve 90% faithfulness on morphological relations, and we show similar findings multi-lingually and across models. Our findings indicate that some conceptual relationships in language models, such as morphology, are readily interpretable from latent space, and are sparsely encoded by cross-layer linear transformations.

cs.CL↗

arXiv2024

Prediction Aided by Surrogate Training

We study a class of prediction problems in which relatively few observations have associated responses, but all observations include both standard covariates as well as additional "helper" covariates. While the end goal is to make high-quality predictions using only the standard covariates, helper covariates can be exploited during training to improve prediction. Helper covariates arise in many applications, including forecasting in time series; incorporation of biased or mis-calibrated predictions from foundation models; and sharing information in transfer learning. We propose "prediction aided by surrogate training" ($\texttt{PAST}$), a class of methods that exploit labeled data to construct a response estimator based on both the standard and helper covariates; and then use the full dataset with pseudo-responses to train a predictor based only on standard covariates. We establish guarantees on the prediction error of this procedure, with the response estimator allowed to be constructed in an arbitrary way, and the final predictor fit by empirical risk minimization over an arbitrary function class. These upper bounds involve the risk associated with the oracle data set (all responses available), plus an overhead that measures the accuracy of the pseudo-responses. This theory characterizes both regimes in which $\texttt{PAST}$ accuracy is comparable to the oracle accuracy, as well as more challenging regimes where it behaves poorly. We demonstrate its empirical performance across a range of applications, including forecasting of societal ills over time with future covariates as helpers; prediction of cardiovascular risk after heart attacks with prescription data as helpers; and diagnosing pneumonia from chest X-rays using machine-generated predictions as helpers.

math.ST↗

arXiv2024

Inference under Staggered Adoption: Case Study of the Affordable Care Act

Panel data consists of a collection of $N$ units that are observed over $T$ units of time. A policy or treatment is subject to staggered adoption if different units take on treatment at different times and remains treated (or never at all). Assessing the effectiveness of such a policy requires estimating the treatment effect, corresponding to the difference between outcomes for treated versus untreated units. We develop inference procedures that build upon a computationally efficient matrix estimator for treatment effects in panel data. Our routines return confidence intervals (CIs) both for individual treatment effects, as well as for more general bilinear functionals of treatment effects, with prescribed coverage guarantees. We apply these inferential methods to analyze the effectiveness of Medicaid expansion portion of the Affordable Care Act. Based on our analysis, Medicaid expansion has led to substantial reductions in uninsurance rates, has reduced infant mortality rates, and has had no significant effects on healthcare expenditures.

stat.ME↗

arXiv2024

Instrumental variables: A non-asymptotic viewpoint

We provide a non-asymptotic analysis of the linear instrumental variable estimator allowing for the presence of exogeneous covariates. In addition, we introduce a novel measure of the strength of an instrument that can be used to derive non-asymptotic confidence intervals. For strong instruments, these non-asymptotic intervals match the asymptotic ones exactly up to higher order corrections; for weaker instruments, our intervals involve adaptive adjustments to the instrument strength, and thus remain valid even when asymptotic predictions break down. We illustrate our results via an analysis of the effect of PM2.5 pollution on various health conditions, using wildfire smoke exposure as an instrument. Our analysis shows that exposure to PM2.5 pollution leads to statistically significant increases in incidence of health conditions such as asthma, heart disease, and strokes.

math.ST↗

arXiv2022

Krylov-Bellman boosting: Super-linear policy evaluation in general state spaces

We present and analyze the Krylov-Bellman Boosting (KBB) algorithm for policy evaluation in general state spaces. It alternates between fitting the Bellman residual using non-parametric regression (as in boosting), and estimating the value function via the least-squares temporal difference (LSTD) procedure applied with a feature set that grows adaptively over time. By exploiting the connection to Krylov methods, we equip this method with two attractive guarantees. First, we provide a general convergence bound that allows for separate estimation errors in residual fitting and LSTD computation. Consistent with our numerical experiments, this bound shows that convergence rates depend on the restricted spectral structure, and are typically super-linear. Second, by combining this meta-result with sample-size dependent guarantees for residual fitting and LSTD computation, we obtain concrete statistical guarantees that depend on the sample size along with the complexity of the function class used to fit the residuals. We illustrate the behavior of the KBB algorithm for various types of policy evaluation problems, and typically find large reductions in sample complexity relative to the standard approach of fitted value iterationn.

stat.ML↗

arXiv2022

Instance-Dependent Confidence and Early Stopping for Reinforcement Learning

Various algorithms for reinforcement learning (RL) exhibit dramatic variation in their convergence rates as a function of problem structure. Such problem-dependent behavior is not captured by worst-case analyses and has accordingly inspired a growing effort in obtaining instance-dependent guarantees and deriving instance-optimal algorithms for RL problems. This research has been carried out, however, primarily within the confines of theory, providing guarantees that explain \textit{ex post} the performance differences observed. A natural next step is to convert these theoretical guarantees into guidelines that are useful in practice. We address the problem of obtaining sharp instance-dependent confidence regions for the policy evaluation problem and the optimal value estimation problem of an MDP, given access to an instance-optimal algorithm. As a consequence, we propose a data-dependent stopping rule for instance-optimal algorithms. The proposed stopping rule adapts to the instance-specific difficulty of the problem and allows for early termination for problems with favorable structure.

stat.ML↗

arXiv2021

Instance-optimality in optimal value estimation: Adaptivity via variance-reduced Q-learning

Various algorithms in reinforcement learning exhibit dramatic variability in their convergence rates and ultimate accuracy as a function of the problem structure. Such instance-specific behavior is not captured by existing global minimax bounds, which are worst-case in nature. We analyze the problem of estimating optimal $Q$-value functions for a discounted Markov decision process with discrete states and actions and identify an instance-dependent functional that controls the difficulty of estimation in the $\ell_\infty$-norm. Using a local minimax framework, we show that this functional arises in lower bounds on the accuracy on any estimation procedure. In the other direction, we establish the sharpness of our lower bounds, up to factors logarithmic in the state and action spaces, by analyzing a variance-reduced version of $Q$-learning. Our theory provides a precise way of distinguishing "easy" problems from "hard" ones in the context of $Q$-learning, as illustrated by an ensemble with a continuum of difficulty.

stat.ML↗

arXiv2019

Posterior Distribution for the Number of Clusters in Dirichlet Process Mixture Models

Dirichlet process mixture models (DPMM) play a central role in Bayesian nonparametrics, with applications throughout statistics and machine learning. DPMMs are generally used in clustering problems where the number of clusters is not known in advance, and the posterior distribution is treated as providing inference for this number. Recently, however, it has been shown that the DPMM is inconsistent in inferring the true number of components in certain cases. This is an asymptotic result, and it would be desirable to understand whether it holds with finite samples, and to more fully understand the full posterior. In this work, we provide a rigorous study for the posterior distribution of the number of clusters in DPMM under different prior distributions on the parameters and constraints on the distributions of the data. We provide novel lower bounds on the ratios of probabilities between $s+1$ clusters and $s$ clusters when the prior distributions on parameters are chosen to be Gaussian or uniform distributions.

stat.ML↗