arXiv subjects

Iain Carmichael

Publications and source records attributed to Iain Carmichael.

12 recordsLinked to original sources

arXiv2026

Mitigating Batch Effects in Histopathology via Language-Mediated Robust Embedding Generation

Pathology foundation models (PFMs) have demonstrated strong potential across clinical and scientific applications, yet their performance is often hindered by batch effects, which are non-biological variations across tissue source institutions (TSIs) that distort learned feature representations and impair generalization. Conventional mitigation strategies, such as stain normalization, offer limited success in addressing these high-dimensional, complex artifacts. We present GLMP (General-purpose LLM-Mediated Pathology model), a novel framework that generates robust numerical embeddings from histology image patches through an intermediate textual representation. By leveraging pretrained general-purpose multimodal large language models (MLLMs) and text encoders, GLMP effectively prioritizes biologically meaningful signals over TSI-specific artifacts, thereby improving cross-institutional generalization. To our knowledge, GLMP is the first pathology model to use text descriptions of histological features as an intermediate representation for generating numerical embeddings from histology images. Our results highlight the untapped potential of broad-domain, non-specialized MLLMs in computational pathology and introduce a new paradigm for building versatile, generalizable, and robust pathology models.

cs.CV↗

arXiv2023

Measure of Strength of Evidence for Visually Observed Differences between Subpopulations

For measuring the strength of visually-observed subpopulation differences, the Population Difference Criterion is proposed to assess the statistical significance of visually observed subpopulation differences. It addresses the following challenges: in high-dimensional contexts, distributional models can be dubious; in high-signal contexts, conventional permutation tests give poor pairwise comparisons. We also make two other contributions: Based on a careful analysis we find that a balanced permutation approach is more powerful in high-signal contexts than conventional permutations. Another contribution is the quantification of uncertainty due to permutation variation via a bootstrap confidence interval. The practical usefulness of these ideas is illustrated in the comparison of subpopulations of modern cancer data.

stat.ME↗

arXiv2022

Incorporating intratumoral heterogeneity into weakly-supervised deep learning models via variance pooling

Supervised learning tasks such as cancer survival prediction from gigapixel whole slide images (WSIs) are a critical challenge in computational pathology that requires modeling complex features of the tumor microenvironment. These learning tasks are often solved with deep multi-instance learning (MIL) models that do not explicitly capture intratumoral heterogeneity. We develop a novel variance pooling architecture that enables a MIL model to incorporate intratumoral heterogeneity into its predictions. Two interpretability tools based on representative patches are illustrated to probe the biological signals captured by these models. An empirical study with 4,479 gigapixel WSIs from the Cancer Genome Atlas shows that adding variance pooling onto MIL frameworks improves survival prediction performance for five cancer types.

eess.IV↗

arXiv2022

Fluctuation Bounds for Continuous Time Branching Processes and Evolution of Growing Trees With a Change Point

We consider dynamic random trees constructed using an attachment function $f : \mathbb{N} \to \mathbb{R}_+$ where, at each step of the evolution, a new vertex attaches to an existing vertex $v$ in the current tree with probability proportional to $f$(degree(v)). We explore the effect of a change point in the system; the dynamics are initially driven by a function f until the tree reaches size $τ(n) \in (0,n)$, at which point the attachment function switches to another function, $g$, until the tree reaches size $n$. Two change point time scales are considered, namely the standard model where $τ(n) = γn$, and the quick big bang model where $τ(n) = n^γ$, for some $0 < γ< 1$. In the former case, we obtain deterministic approximations for the evolution of the empirical degree distribution (EDF) in sup-norm and use these to devise a provably consistent non-parametric estimator for the change point $γ$. In the latter case, we show that the effect of pre-change point dynamics asymptotically vanishes in the EDF, although this effect persists in functionals such as the maximal degree. Our proofs rely on embedding the discrete time tree dynamics in an associated (time) inhomogeneous continuous time branching process (CTBP). In the course of proving the above results, we develop novel mathematical techniques to analyze both homogeneous and inhomogeneous CTBPs and obtain rates of convergence for functionals of such processes, which are of independent interest.

math.PR↗

arXiv2021

yaglm: a Python package for fitting and tuning generalized linear models that supports structured, adaptive and non-convex penalties

The yaglm package aims to make the broader ecosystem of modern generalized linear models accessible to data analysts and researchers. This ecosystem encompasses a range of loss functions (e.g. linear, logistic, quantile regression), constraints (e.g. positive, isotonic) and penalties. Beyond the basic lasso/ridge, the package supports structured penalties such as the nuclear norm as well as the group, exclusive, fused, and generalized lasso. It also supports more accurate adaptive and non-convex (e.g. SCAD) versions of these penalties that often come with strong statistical guarantees at limited additional computational expense. yaglm comes with a variety of tuning parameter selection methods including: cross-validation, information criteria that have favorable model selection properties, and degrees of freedom estimators. While several solvers are built in (e.g. FISTA), a key design choice allows users to employ their favorite state of the art optimization algorithms. Designed to be user friendly, the package automatically creates tuning parameter grids, supports tuning with fast path algorithms along with parallelization, and follows a unified scikit-learn compatible API.

stat.CO↗

arXiv2021

The folded concave Laplacian spectral penalty learns block diagonal sparsity patterns with the strong oracle property

Structured sparsity is an important part of the modern statistical toolkit. We say a set of model parameters has block diagonal sparsity up to permutations if its elements can be viewed as the edges of a graph that has multiple connected components. For example, a block diagonal correlation matrix with K blocks of variables corresponds to a graph with K connected components whose nodes are the variables and whose edges are the correlations. This type of sparsity captures clusters of model parameters. To learn block diagonal sparsity patterns we develop the folded concave Laplacian spectral penalty and provide a majorization-minimization algorithm for the resulting non-convex problem. We show this algorithm has the appealing computational and statistical guarantee of converging to the oracle estimator after two steps with high probability, even in high-dimensional settings. The theory is then demonstrated in several classical problems including covariance estimation, linear regression, and logistic regression.

math.ST↗

arXiv2021

mvlearn: Multiview Machine Learning in Python

As data are generated more and more from multiple disparate sources, multiview data sets, where each sample has features in distinct views, have ballooned in recent years. However, no comprehensive package exists that enables non-specialists to use these methods easily. mvlearn is a Python library which implements the leading multiview machine learning methods. Its simple API closely follows that of scikit-learn for increased ease-of-use. The package can be installed from Python Package Index (PyPI) and the conda package manager and is released under the MIT open-source license. The documentation, detailed examples, and all releases are available at https://mvlearn.github.io/.

stat.ML↗

arXiv2020

Learning Sparsity and Block Diagonal Structure in Multi-View Mixture Models

Scientific studies increasingly collect multiple modalities of data to investigate a phenomenon from several perspectives. In integrative data analysis it is important to understand how information is heterogeneously spread across these different data sources. To this end, we consider a parametric clustering model for the subjects in a multi-view data set (i.e. multiple sources of data from the same set of subjects) where each view marginally follows a mixture model. In the case of two views, the dependence between them is captured by a cluster membership matrix parameter and we aim to learn the structure of this matrix (e.g. the zero pattern). First, we develop a penalized likelihood approach to estimate the sparsity pattern of the cluster membership matrix. For the specific case of block diagonal structures, we develop a constrained likelihood formulation where this matrix is constrained to be block diagonal up to permutations of the rows and columns. To enforce block diagonal constraints we propose a novel optimization approach based on the symmetric graph Laplacian. We demonstrate the performance of these methods through both simulations and applications to data sets from cancer genetics and neuroscience. Both methods naturally extend to multiple views.

stat.ME↗

arXiv2020

Joint and individual analysis of breast cancer histologic images and genomic covariates

A key challenge in modern data analysis is understanding connections between complex and differing modalities of data. For example, two of the main approaches to the study of breast cancer are histopathology (analyzing visual characteristics of tumors) and genetics. While histopathology is the gold standard for diagnostics and there have been many recent breakthroughs in genetics, there is little overlap between these two fields. We aim to bridge this gap by developing methods based on Angle-based Joint and Individual Variation Explained (AJIVE) to directly explore similarities and differences between these two modalities. Our approach exploits Convolutional Neural Networks (CNNs) as a powerful, automatic method for image feature extraction to address some of the challenges presented by statistical analysis of histopathology image data. CNNs raise issues of interpretability that we address by developing novel methods to explore visual modes of variation captured by statistical algorithms (e.g. PCA or AJIVE) applied to CNN features. Our results provide many interpretable connections and contrasts between histopathology and genetics.

q-bio.QM↗

arXiv2018

Geometric Insights into Support Vector Machine Behavior using the KKT Conditions

The support vector machine (SVM) is a powerful and widely used classification algorithm. This paper uses the Karush-Kuhn-Tucker conditions to provide rigorous mathematical proof for new insights into the behavior of SVM. These insights provide perhaps unexpected relationships between SVM and two other linear classifiers: the mean difference and the maximal data piling direction. For example, we show that in many cases SVM can be viewed as a cropped version of these classifiers. By carefully exploring these connections we show how SVM tuning behavior is affected by characteristics including: balanced vs. unbalanced classes, low vs. high dimension, separable vs. non-separable data. These results provide further insights into tuning SVM via cross-validation by explaining observed pathological behavior and motivating improved cross-validation methodology. Finally, we also provide new results on the geometry of complete data piling directions in high dimensional space.

stat.ML↗

arXiv2018

An exposition of the false confidence theorem

A recent paper presents the "false confidence theorem" (FCT) which has potentially broad implications for statistical inference using Bayesian posterior uncertainty. This theorem says that with arbitrarily large (sampling/frequentist) probability, there exists a set which does \textit{not} contain the true parameter value, but which has arbitrarily large posterior probability. Since the use of Bayesian methods has become increasingly popular in applications of science, engineering, and business, it is critically important to understand when Bayesian procedures lead to problematic statistical inferences or interpretations. In this paper, we consider a number of examples demonstrating the paradoxical nature of false confidence to begin to understand the contexts in which the FCT does (and does not) play a meaningful role in statistical inference. Our examples illustrate that models involving marginalization to non-linear, not one-to-one functions of multiple parameters play a key role in more extreme manifestations of false confidence.

stat.ME↗

arXiv2018

Data Science vs. Statistics: Two Cultures?

Data science is the business of learning from data, which is traditionally the business of statistics. Data science, however, is often understood as a broader, task-driven and computationally-oriented version of statistics. Both the term data science and the broader idea it conveys have origins in statistics and are a reaction to a narrower view of data analysis. Expanding upon the views of a number of statisticians, this paper encourages a big-tent view of data analysis. We examine how evolving approaches to modern data analysis relate to the existing discipline of statistics (e.g. exploratory analysis, machine learning, reproducibility, computation, communication and the role of theory). Finally, we discuss what these trends mean for the future of statistics by highlighting promising directions for communication, education and research.

stat.OT↗