Searcharxiv⌕ Search

arXiv subjects

Jeff M. Phillips

Publications and source records attributed to Jeff M. Phillips.

At least 19 recordsLinked to original sources
arXiv2026

Orthogonal and Parametrized Multi-Attribute Steering of LLMs

Activation steering offers an effective and efficient way to modify an LLM's behavior at inference time. However, it remains unclear how to apply steering such that the signal is consistently strong and the steering strength has a universally interpretable meaning. We address this first by identifying the most steerable interval for each steering vector and rescaling the steering range accordingly. Building on this, we introduce a more interpretable method for steering, called parameterized steering which instead of pushing activations in a direction, it sets them at the desired trait strength. Next, with multi-trait steering, we find that naively combining vectors causes them to interfere with one another. In response, we develop a method to orthogonalize the vectors; it decouples these traits, allowing a model to exhibit one behavior without inducing the others, thus enabling multiple traits to be steered simultaneously.

cs.LG↗
arXiv2026

Simulation-Based Inference with Convolutional Maximum Mean Discrepancy for Noisy Data

Ignoring measurement noise degrades statistical inference. Existing correction techniques are often computationally costly and inefficient. Kernel methods based on maximum mean discrepancy have enabled efficient, likelihood-free inference, but they typically assume clean data. We introduce a framework for inference with samples corrupted by potentially heteroscedastic noise from a known distribution. Central to our method is the convolutional maximum mean discrepancy, which compares distributions after noise convolution and retains metric validity under standard kernel conditions. We establish finite-sample deviation bounds that are unaffected by measurement error and an equivalence between testing under noise and kernel smoothing. Based on this discrepancy, we propose an estimator, establish its consistency and asymptotic normality, and provide an efficient implementation using stochastic gradient descent. We demonstrate the practical effectiveness of our method through simulations and applications in astronomy, anthropometry, and the social sciences. Our method provides a unified, likelihood-free, simulation-based framework for efficient and robust inference for broad classes of problems in noisy data.

stat.ME↗
arXiv2026

A Classifier Framework for ALMA Bandpass Calibration Anomalies

Bandpass calibration corrects instrumental responses across observing frequencies in interferometric radio observations, so the measured calibration solutions inherently contain spectral variation that varies with instrumental configurations and noise conditions. However, some signals contain anomalies, which if not caught can lead to poorly constructed science products. Observatory pipelines often use scalar metrics with individual, linear decision boundaries. Constructing a simple parametric model can be effective for observations with a relatively limited parameter space, but it becomes progressively less accurate as the parameter space expands. To address this challenge, we design a machine learning framework that treats anomaly identification as a supervised classification task and learns complex, non-linear decision boundaries across multiple features optimized for predictive performance. In particular, we use an XGBoost model that sits on top of several features and specialized scan statistic models to classify anomalies in amplitude solutions from ALMA bandpass calibrations. Trained on an extensive, quality-controlled ALMA dataset, this classifier framework leverages features derived from kernel-based scan statistics to characterize deviating spectral intervals, while remaining extensible to additional feature sets. This method substantially reduces misclassifications in bandpass anomaly detection compared to the current pipeline heuristics. By improving flagging decisions and calibration quality, this classifier framework provides a scalable supplement for observatory operations and is generalizable to next-generation radio facilities.

astro-ph.IM↗
arXiv2026

Judging by the Cover: Cleaning LLM Truthfulness Benchmarks to Avoid Surface-Level Feature Leakage

Binary-choice truth benchmarks ask models to choose between a correct and an incorrect answer, but if the two answers differ systematically in surface-level features, models can exceed chance without performing the intended reasoning. We show that this failure mode is detectable and can be exploited by downstream classifiers. In TruthfulQA, a simple six-feature logistic classifier achieves substantial accuracy in separating correct from incorrect answers. We further show that similar surface-level artifacts are present in additional benchmarks. To counteract this, we developed a general mechanism to clean them by removing the most leakage-reinforcing pairs. We release a version of TruthfulQA with surface-feature leakage reduced close to chance and provide a mechanism, Audit-Prune, so that the datasets can be cleaned before release.

cs.CL↗
arXiv2026

Efficient Regression Models for Scan Statistics

We introduce a new class of regression models for scan statistics on real-valued signals. These allow for improved fitting of non-stationary signals to contrast with the interval anomalies identified by the scan statistics. Our models can represent generalized likelihood ratio statistics. While these methods naively require $O(n^4)$ for a length $n$ signal, we provide algorithmic improvements which lead to linear time algorithms (with assumptions on max interval width). Our methods, especially ones based on Nadaraya-Watson kernel regression, are demonstrated as especially effective in detecting both synthetically planted anomalies, and for identifying a real ``platforming'' issue in interferometric astronomy.

stat.ME↗
arXiv2026

Sampling for Region-Aggregated Spatial Scan Statistics

Anomaly detection in geospatial data is a crucial tool in geographic information science (GIS), with applications ranging from national security to public-health surveillance to the study of societal disparities. This work focuses on spatial scan statistics and addresses a key mismatch: spatial counts are typically aggregated into predefined regions (census tracts, zip codes, counties), whereas the most efficient scan algorithms operate on spatial point data. The standard remedy -- collapsing each region to its centroid, as in widely used tools such as SaTScan -- is convenient but, as we show, discards the region's spatial extent and causes a significant loss in statistical power. To resolve this, we propose a simple yet scalable fix: replace each spatial region with 20-50 points sampled uniformly from its geometry, and divide the region's measured and baseline counts evenly among them. This approach improves statistical power while maintaining computational tractability. A convergence analysis explains why so few samples per region suffice. We recommend this sampling-based conversion as the default way to apply point-based spatial scan statistics to region-aggregated data for anomaly detection.

stat.AP↗
arXiv2026

The Expiration Streaming Model: Diameter, $k$-Center, Counting, Sampling, and Friends

An important thread in the study of data-stream algorithms focuses on settings where stream items are active only for a limited time. We introduce a new expiration model, where each item arrives with its own arbitrary expiration time. The special case where items expire in the order that they arrive, which we call consistent expirations, contains the classical sliding-window model of Datar, Gionis, Indyk, and Motwani [SICOMP 2002] and its timestamp-based variant of Braverman and Ostrovsky [FOCS 2007]. Our first set of results explores the expiration streaming model and presents algorithms for several fundamental problems, including approximate counting, uniform sampling, and weighted sampling by efficiently tracking active items without explicitly storing them all. Naturally, these algorithms have many immediate applications, e.g., to range counting. Our second and main set of results for the expiration model designs algorithms for the diameter and k-center problems, where items are points in a metric space. Our results significantly extend those known for the special case of sliding-window streams by Cohen-Addad, Schwiegelshohn, and Sohler [ICALP 2016], and obtain a strictly better approximation factor for the diameter in the important special case of high-dimensional Euclidean metrics. We develop new decomposition and coordination techniques along with a geometric dominance framework to filter out redundant points based on both temporal and spatial proximity.

cs.DS↗
arXiv2026

Efficient and Stable Multi-Dimensional Kolmogorov-Smirnov Distance

We revisit extending the Kolmogorov-Smirnov distance between probability distributions to the multi-dimensional setting, and make new arguments about the proper way to approach this generalization. Our proposed formulation maximizes the difference over orthogonal dominating rectangular ranges (d-sided rectangles in R^d), and is an integral probability metric. We also prove that the distance between a distribution and a sample from the distribution converges to 0 as the sample size grows, and bound this rate. Moreover, we show that one can, up to this same approximation error, compute the distance efficiently in 4 or fewer dimensions; specifically, the runtime is near-linear in the size of the sample needed for that error. With this, we derive a delta-precision two-sample hypothesis test using this distance. Finally, we show these metrics and approximation properties do not hold for other popular variants.

stat.CO↗
arXiv2026

Rotation-Invariant Vectorized Shape Representations

We introduce a rotation-invariant representation of planar shapes. In particular, this representation encodes shapes as vectors such that the Euclidean distance between them serves as a valid shape distance. For standardized, star-shaped objects, we can deterministically create a sketched vector of dimension $O(1/\varepsilon)$ in $O((1/\varepsilon) \log (1/\varepsilon))$ time that approximates this shape distance to within $\varepsilon$. Moreover, because the representation is a standard Euclidean vector, we can directly and efficiently perform various data analyses, such as nearest neighbor search and clustering, in shape space, inherently invariant to the rotation of the shapes. We demonstrate this through a series of simple experiments. The key technical contribution operates on functions over $\mathbb{S}^1$, which we use to encode standardized objects. The most general rotation-invariant representation of these functions works through a map to an infinite-dimensional function space, parameterized by an offset parameter. By analyzing special discretized cases of these functions, we show that the representation is strictly injective up to the desired rotation and a mirror-flip-type operation we call \emph{reverse of complement} (RoC). While RoC status can be controlled by how the function is defined, it is inherent to the representation and required to be handled in the analysis. Regardless, the vectorized representation is robust to small shape perturbations, and hence discretizing the angles leads to the efficient approximation and algorithm.

cs.CG↗
arXiv2026

Optimal Dimension-Free Sampling for Regularized Classification

We prove optimal sampling bounds achieving $(1\pm\varepsilon)$-relative error for a broad class of Lipschitz continuous classification loss functions under various regularization terms. This includes important functions such as logistic and sigmoid loss, hinge loss, and ReLU loss, as prominent and popular representative examples. In particular, we prove $k^2/\varepsilon^2$ upper and lower bounds for $\|\cdot\|_2/k$ regularization, and $k/\varepsilon^2$ upper and lower bounds for $\|\cdot\|_1/k$ regularization. For $\|\cdot\|_2^2/k$ regularization, the sampling complexity depends mainly on a bounded derivative property: if $|g'(x)|\leq g(x)$, and $g(0)>0$, and $g$ is monotonic or convex, then it admits linear in $k$ sampling complexity; otherwise the general bound is $k^2/\varepsilon^2$. However, if $g(0)=0$, our results indicate that no dimension-free bounds are possible, and even sublinear bounds are ruled out. All upper bounds are complemented by matching lower bounds up to polylogarithmic terms. Moreover, our work relies conceptually and algorithmically on simple uniform or (squared) norm sampling and hereby improves over recent cubic $k^3/\varepsilon^2$ sensitivity sampling bounds of (Alishahi and Phillips, ICML'24). This is achieved by refined arguments involving higher moment bounds and empirical process analyses to avoid overcounting that appears in the de-facto standard VC-dimension and sensitivity framework.

cs.LG↗
arXiv2026

Optimizing Computational-Statistical Runtime for Wasserstein Distance Estimation

Squared Wasserstein distance is a frequently used tool to measure discrepancy between probability distributions. This distance is typically computed between empirical measures of size $n$ from two underlying random samples. Unfortunately, even in lower dimensional Euclidean space problems $\left( d \in \{2,3\} \right)$, algorithms for Wasserstein distance computation with approximate or exact precision guarantees scale poorly in the runtime as a function of $n$ and the desired precision. In response, we consider the computational-statistical runtime, where the goal is to estimate from samples the Wasserstein distance between potentially smooth measures up to $ε$-additive error in expectation with respect to the sampling; we allow $O(1)$ computational cost for collecting a sample. Towards this, we develop a Sample-Sketch-Solve paradigm where we introduce a regular cartesian grid sketch of the samples. We show that (especially under $α$-Hölder smooth distributions) this can compress the data without increasing asymptotic error, and also regularizes the structure which enables faster exact algorithms. Ultimately, we approximate $W_2^2(P,Q)$ within $ε$ error in $ε^{-\max(2,\frac{d+1+o(1)}{1+α})}$ time for $0 < α< 1$ Hölder smooth distributions $P,Q$ on $(0,1)^{d}$; an optimal $Θ(ε^{-2})$ for $α> 1/2$ when $d=2$ and nearly optimal as $α\to 1$ when $d = 3$.

stat.ML↗
arXiv2026

TabKDE: Simple and Scalable Tabular Data Generation with Kernel Density Estimates

Tabular data generation considers a large table with multiple columns -- each column comprised of numerical, categorical, or sometimes ordinal values. The goal is to produce new rows for the table that replicate the distribution of rows from the original data -- without just copying those initial rows. The last 4 years have seen enormous progress on this problem, mostly using computational expensive methods that employ one-hot encoding, VAEs, and diffusion. This paper describes a new approach to the problem of tabular data generation. By employing copula transformations and modeling the distribution as a kernel density estimate we can nearly match the accuracy and leakage-avoidance achievements of the previous methods, but with almost no training time. Our method is very scalable, and can be run on data sets orders of magnitude larger than prior state-of-the-art on a simple laptop. Moreover, because we employ kernel density estimates, we can store the model as a coreset of the original data -- we believe the first for generative modeling -- and as a result, require significantly less space as well. Our code is available here: \url{https://github.com/tabkde/tabkde-main}

cs.LG↗
arXiv2026

Computing Planar Convex Hulls with a Promise

Computing the convex hull of a planar $n$-point set $P$ is one of the most fundamental problems in computational geometry. It has an $Ω(n \log n)$ lower bound in the algebraic computation tree model, and many convex hull algorithms match this bound. Classical results show that, under special input assumptions, sub-$O(n \log n)$ algorithms are possible. For instance, when the points are given in lexicographic or angular order, the convex hull can be computed in linear time. Even under the weaker assumption that the sequence of points corresponds to the ordered vertices of a simple polygonal chain, linear-time algorithms exist. This naturally raises the question: can the convex hull of a point set be computed in sub-$O(n \log n)$ time under weaker input assumptions? We answer this positively. Under the promise that the input sequence contains the convex hull as a subsequence, we give a deterministic $O(n \sqrt{\log n})$-time algorithm to compute the convex hull of $P$. With randomisation, we achieve expected running time $O(n \log^{\varepsilon} n)$ for any constant $\varepsilon > 0$. We find this surprising, as points not on the convex hull may behave adversarially toward our convex hull construction algorithm. Yet the promise that \emph{only} the hull points are sorted suffices for $o(n \log n)$-time algorithms. Finally, we show that this promise is tight: if it is even slightly broken, i.e., allowing just one hull point to appear out of order, we prove an adversarial $Ω(n \log n)$-time lower bound. Consequently, the promise cannot be verified with fewer than $Ω(n \log n)$ comparisons. This also negatively resolves an open problem of Löffler and Raichel, who conjectured sub-$O(n \log n)$-time algorithms for computing the convex hull of a supersequence containing the hull as a subsequence.

cs.CG↗
arXiv2026

Relative Error Embeddings for the Gaussian Kernel Distance

A reproducing kernel can define an embedding of a data point into an infinite dimensional reproducing kernel Hilbert space (RKHS). The norm in this space describes a distance, which we call the kernel distance. The random Fourier features (of Rahimi and Recht) describe an oblivious approximate mapping into finite dimensional Euclidean space that behaves similar to the RKHS. We show in this paper that for the Gaussian kernel the Euclidean norm between these mapped to features has $(1+\varepsilon)$-relative error with respect to the kernel distance. When there are $n$ data points, we show that $O((1/\varepsilon^2) \log(n))$ dimensions of the approximate feature space are sufficient and necessary. Without a bound on $n$, but when the original points lie in $\mathbb{R}^d$ and have diameter bounded by $\mathcal{M}$, then we show that $O((d/\varepsilon^2) \log(\mathcal{M}))$ dimensions are sufficient, and that this many are required, up to $\log(1/\varepsilon)$ factors.

cs.LG↗
arXiv2026

Curveball Steering: The Right Direction To Steer Isn't Always Linear

Activation steering is a widely used approach for controlling large language model (LLM) behavior by intervening on internal representations. Existing methods largely rely on the Linear Representation Hypothesis, assuming behavioral attributes can be manipulated using global linear directions. In practice, however, such linear interventions often behave inconsistently. We question this assumption by analyzing the intrinsic geometry of LLM activation spaces. Measuring geometric distortion via the ratio of geodesic to Euclidean distances, we observe substantial and concept-dependent distortions, indicating that activation spaces are not well-approximated by a globally linear geometry. Motivated by this, we propose "Curveball steering", a nonlinear steering method based on polynomial kernel PCA that performs interventions in a feature space, better respecting the learned activation geometry. Curveball steering consistently outperforms linear PCA-based steering, particularly in regimes exhibiting strong geometric distortion, suggesting that geometry-aware, nonlinear steering provides a principled alternative to global, linear interventions.

cs.AI↗
arXiv2026

Hardness of High-Dimensional Linear Classification

We establish new exponential in dimension lower bounds for the Maximum Halfspace Discrepancy problem, which models linear classification. Both are fundamental problems in computational geometry and machine learning in their exact and approximate forms. However, only $O(n^d)$ and respectively $\tilde O(1/\varepsilon^d)$ upper bounds are known and complemented by polynomial lower bounds that do not support the exponential in dimension dependence. We close this gap up to polylogarithmic terms by reduction from widely-believed hardness conjectures for Affine Degeneracy testing and $k$-Sum problems. Our reductions yield matching lower bounds of $\tildeΩ(n^d)$ and respectively $\tildeΩ(1/\varepsilon^d)$ based on Affine Degeneracy testing, and $\tildeΩ(n^{d/2})$ and respectively $\tildeΩ(1/\varepsilon^{d/2})$ conditioned on $k$-Sum. The first bound also holds unconditionally if the computational model is restricted to make sidedness queries, which corresponds to a widely spread setting implemented and optimized in many contemporary algorithms and computing paradigms.

cs.CG↗
arXiv2026

Understanding and Mitigating Dataset Corruption in LLM Steering

Contrastive steering has been shown as a simple and effective method to adjust the generative behavior of LLMs at inference time. It uses examples of prompt responses with and without a trait to identify a direction in an intermediate activation layer, and then shifts activations in this 1-dimensional subspace. However, despite its growing use in AI safety applications, the robustness of contrastive steering to noisy or adversarial data corruption is poorly understood. We initiate a study of the robustness of this process with respect to corruption of the dataset of examples used to train the steering direction. Our first observation is that contrastive steering is quite robust to a moderate amount of corruption, but unwanted side effects can be clearly and maliciously manifested when a non-trivial fraction of the training data is altered. Second, we analyze the geometry of various types of corruption, and identify some safeguards. Notably, a key step in learning the steering direction involves high-dimensional mean computation, and we show that replacing this step with a recently developed robust mean estimator often mitigates most of the unwanted effects of malicious corruption.

cs.LG↗
arXiv2025

Designing a Secure and Resilient Distributed Smartphone Participant Data Collection System

Real-world health studies require continuous and secure data collection from mobile and wearable devices. We introduce MotionPI, a smartphone-based system designed to collect behavioral and health data through sensors and surveys with minimal interaction from participants. The system integrates passive data collection (such as GPS and wristband motion data) with Ecological Momentary Assessment (EMA) surveys, which can be triggered randomly or based on physical activity. MotionPI is designed to work under real-life constraints, including limited battery life, weak or intermittent cellular connection, and minimal user supervision. It stores data both locally and on a secure cloud server, with encrypted transmission and storage. It integrates through Bluetooth Low Energy (BLE) into wristband devices that store raw data and communicate motion summaries and trigger events. MotionPI demonstrates a practical solution for secure and scalable mobile data collection in cyber-physical health studies.

cs.CR↗