Martin T. Wells

Luca Candelori, Alexander G. Abanov, Jeffrey Berger et al.

We propose a new data representation method based on Quantum Cognition Machine Learning and apply it to manifold learning, specifically to the estimation of intrinsic dimension of data sets. The idea is to learn a representation of each data point as a quantum state, encoding both local properties of the point as well as its relation with the entire data. Inspired by ideas from quantum geometry, we then construct from the quantum states a point cloud equipped with a quantum metric. The metric exhibits a spectral gap whose location corresponds to the intrinsic dimension of the data. The proposed estimator is based on the detection of this spectral gap. When tested on synthetic manifold benchmarks, our estimates are shown to be robust with respect to the introduction of point-wise Gaussian noise. This is in contrast to current state-of-the-art estimators, which tend to attribute artificial ``shadow dimensions'' to noise artifacts, leading to overestimates. This is a significant advantage when dealing with real data sets, which are inevitably affected by unknown levels of noise. We show the applicability and robustness of our method on real data, by testing it on the ISOMAP face database, MNIST, and the Wisconsin Breast Cancer Dataset.

Haoxuan Wu, David S. Matteson, Martin T. Wells

We introduce a new version of deep state-space models (DSSMs) that combines a recurrent neural network with a state-space framework to forecast time series data. The model estimates the observed series as functions of latent variables that evolve non-linearly through time. Due to the complexity and non-linearity inherent in DSSMs, previous works on DSSMs typically produced latent variables that are very difficult to interpret. Our paper focus on producing interpretable latent parameters with two key modifications. First, we simplify the predictive decoder by restricting the response variables to be a linear transformation of the latent variables plus some noise. Second, we utilize shrinkage priors on the latent variables to reduce redundancy and improve robustness. These changes make the latent variables much easier to understand and allow us to interpret the resulting latent variables as random effects in a linear mixed model. We show through two public benchmark datasets the resulting model improves forecasting performances.

Sharan Sahu, Cameron J. Hogan, Martin T. Wells

In this paper, we provide a comprehensive theoretical analysis of Stochastic Gradient Descent (SGD) and its momentum variants (Polyak Heavy-Ball and Nesterov) for tracking time-varying optima under strong convexity and smoothness. Our finite-time bounds reveal a sharp decomposition of tracking error into transient, noise-induced, and drift-induced components. This decomposition exposes a fundamental trade-off: while momentum is often used as a gradient-smoothing heuristic, under distribution shift it incurs an explicit drift-amplification penalty that diverges as the momentum parameter $β$ approaches 1, yielding systematic tracking lag. We complement these upper bounds with minimax lower bounds under gradient-variation constraints, proving this momentum-induced tracking penalty is not an analytical artifact but an information-theoretic barrier: in drift-dominated regimes, momentum is unavoidably worse because stale-gradient averaging forces systematic lag. Our results provide theoretical grounding for the empirical instability of momentum in dynamic settings and precisely delineate regime boundaries where vanilla SGD provably outperforms its accelerated counterparts.

Sharan Sahu, Abir Sarkar, Cameron J. Hogan et al.

We provide a theoretical analysis of Adam under non-stationary stochastic objectives, separating two regimes: Euclidean tracking under adaptive strong monotonicity of the Adam-preconditioned mean-gradient operator, and high-probability projected stationarity guarantees under general $L$-smooth objectives. In the tracking regime, we derive finite-time expected and high-probability bounds that decompose sharply into four components: initialization, objective drift, a first-moment tracking error governed by $β_1$, and a preconditioner perturbation governed by $β_2$. We characterize the burn-in time to reach Adam's irreducible tracking floor under constant and step-decay schedules. We also prove a high-probability bound on the average projected stationarity gap for Adam under distribution shift. Across both analyses, our bounds reveal a noise--drift tradeoff: in noise-dominated regimes, first-moment averaging and adaptive preconditioning can improve the high-probability error, whereas in drift-dominated regimes, stale first-moment information and preconditioner perturbations can compound the cost of nonstationarity, allowing vanilla SGD to achieve a smaller tracking floor. Our explicit $(β_1,β_2,ε)$-dependent bounds delineate when adaptive step-sizing is beneficial versus harmful, and provide a theoretical mechanism for Adam's empirical instability and stabilization under distribution shift.

Keri Mallari, Julius Adebayo, Kori Inkpen et al.

Despite strong advisory against it, large generative models (LMs) are already being used for decision making tasks that were previously done by predictive models or humans. We put popular LMs to the test in a high-stakes decision making task: recidivism prediction. Studying three closed-access and open-source LMs, we analyze the LMs not exclusively in terms of accuracy, but also in terms of agreement with (imperfect, noisy, and sometimes biased) human predictions or existing predictive models. We conduct experiments that assess how providing different types of information, including distractor information such as photos, can influence LM decisions. We also stress test techniques designed to either increase accuracy or mitigate bias in LMs, and find that some to have unintended consequences on LM decisions. Our results provide additional quantitative evidence to the wisdom that current LMs are not the right tools for these types of tasks.

Skyler Seto, Martin T. Wells, Wenyu Zhang

Deep neural networks achieve state-of-the-art performance in a variety of tasks by extracting a rich set of features from unstructured data, however this performance is closely tied to model size. Modern techniques for inducing sparsity and reducing model size are (1) network pruning, (2) training with a sparsity inducing penalty, and (3) training a binary mask jointly with the weights of the network. We study different sparsity inducing penalties from the perspective of Bayesian hierarchical models and present a novel penalty called Hierarchical Adaptive Lasso (HALO) which learns to adaptively sparsify weights of a given network via trainable parameters. When used to train over-parametrized networks, our penalty yields small subnetworks with high accuracy without fine-tuning. Empirically, on image recognition tasks, we find that HALO is able to learn highly sparse network (only 5% of the parameters) with significant gains in performance over state-of-the-art magnitude pruning methods at the same level of sparsity. Code is available at https://github.com/skyler120/sparsity-halo.

Tejas Ramdas, Martin T. Wells

In this study, we leverage powerful non-linear machine learning methods to identify the characteristics of trades that contain valuable information. First, we demonstrate the effectiveness of our optimized neural network predictor in accurately predicting future market movements. Then, we utilize the information from this successful neural network predictor to pinpoint the individual trades within each data point (trading window) that had the most impact on the optimized neural network's prediction of future price movements. This approach helps us uncover important insights about the heterogeneity in information content provided by trades of different sizes, venues, trading contexts, and over time.

Daniel Andrew Coulson, Martin T. Wells

Bayesian neural networks (BNNs) offer a natural probabilistic formulation for inference in deep learning models. Despite their popularity, their optimality has received limited attention through the lens of statistical decision theory. In this paper, we study decision rules induced by deep, fully connected feedforward ReLU BNNs in the normal location model under quadratic loss. We show that, for fixed prior scales, the induced Bayes decision rule is not minimax. We then propose a hyperprior on the effective output variance of the BNN prior that yields a superharmonic square-root marginal density, establishing that the resulting decision rule is simultaneously admissible and minimax. We further extend these results from the quadratic loss setting to the predictive density estimation problem with Kullback--Leibler loss. Finally, we validate our theoretical findings numerically through simulation.

Giuseppe Di Caro, Vahagn Kirakosyan, Alexander G. Abanov et al.

The accurate prediction of chromosomal instability from the morphology of circulating tumor cells (CTCs) enables real-time detection of CTCs with high metastatic potential in the context of liquid biopsy diagnostics. However, it presents a significant challenge due to the high dimensionality and complexity of single-cell digital pathology data. Here, we introduce the application of Quantum Cognition Machine Learning (QCML), a quantum-inspired computational framework, to estimate morphology-predicted chromosomal instability in CTCs from patients with metastatic breast cancer. QCML leverages quantum mechanical principles to represent data as state vectors in a Hilbert space, enabling context-aware feature modeling, dimensionality reduction, and enhanced generalization without requiring curated feature selection. QCML outperforms conventional machine learning methods when tested on out of sample verification CTCs, achieving higher accuracy in identifying predicted large-scale state transitions (pLST) status from CTC-derived morphology features. These preliminary findings support the application of QCML as a novel machine learning tool with superior performance in high-dimensional, low-sample-size biomedical contexts. QCML enables the simulation of cognition-like learning for the identification of biologically meaningful prediction of chromosomal instability from CTC morphology, offering a novel tool for CTC classification in liquid biopsy.

Suvadip Sana, Jinzhou Wu, Martin T. Wells

Whose values should AI systems learn? Preference based alignment methods like RLHF derive their training signal from human raters, yet these rater pools are typically convenience samples that systematically over represent some demographics and under represent others. We introduce Democratic Preference Optimization, or DemPO, a framework that applies algorithmic sortition, the same mechanism used to construct citizen assemblies, to preference based fine tuning. DemPO offers two training schemes. Hard Panel trains exclusively on preferences from a quota satisfying mini public sampled via sortition. Soft Panel retains all data but reweights each rater by their inclusion probability under the sortition lottery. We prove that Soft Panel weighting recovers the expected Hard Panel objective in closed form. Using a public preference dataset that pairs human judgments with rater demographics and a seventy five clause constitution independently elicited from a representative United States panel, we evaluate Llama models from one billion to eight billion parameters fine tuned under each scheme. Across six aggregation methods, the Hard Panel consistently ranks first and the Soft Panel consistently outperforms the unweighted baseline, with effect sizes growing as model capacity increases. These results demonstrate that enforcing demographic representativeness at the preference collection stage, rather than post hoc correction, yields models whose behavior better reflects values elicited from representative publics.

Sharan Sahu, Martin T. Wells

Reinforcement learning with human feedback (RLHF) has become crucial for aligning Large Language Models (LLMs) with human intent. However, existing offline RLHF approaches suffer from overoptimization, where models overfit to reward misspecification and drift from preferred behaviors observed during training. We introduce DRO-REBEL, a unified family of robust REBEL updates with type-$p$ Wasserstein, KL, and $χ^2$ ambiguity sets. Using Fenchel duality, each update reduces to a simple relative-reward regression, preserving scalability and avoiding PPO-style clipping or auxiliary value networks. Under standard linear-reward and log-linear policy classes with a data-coverage condition, we establish $O(n^{-1/4})$ estimation bounds with tighter constants than prior DRO-DPO approaches, and recover the minimax-optimal $O(n^{-1/2})$ rate via a localized Rademacher complexity analysis. The same analysis closes the gap for Wasserstein-DPO and KL-DPO, showing both also attain optimal parametric rates. We derive practical SGD algorithms for all three divergences: gradient regularization (Wasserstein), importance weighting (KL), and a fast 1-D dual solve ($χ^2$). Experiments on Emotion Alignment, the large-scale ArmoRM multi-objective benchmark, and HH-Alignment demonstrate strong worst-case robustness across unseen preference mixtures, model sizes, and data scales, with $χ^2$-REBEL showing consistently strong empirical performance. A controlled radius--coverage study validates a no-free-lunch trade-off: radii shrinking faster than empirical divergence concentration rates achieve minimax-optimal parametric rates but forfeit coverage, while coverage-guaranteeing radii incur $O(n^{-1/4})$ rates.

Alexander G. Abanov, Luca Candelori, Harold C. Steinacker et al.

We demonstrate how Quantum Cognition Machine Learning (QCML) encodes data as quantum geometry. In QCML, features of the data are represented by learned Hermitian matrices, and data points are mapped to states in Hilbert space. The quantum geometry description endows the dataset with rich geometric and topological structure - including intrinsic dimension, quantum metric, and Berry curvature - derived directly from the data. QCML captures global properties of data, while avoiding the curse of dimensionality inherent in local methods. We illustrate this on a number of synthetic and real-world examples. Quantum geometric representation of QCML could advance our understanding of cognitive phenomena within the framework of quantum cognition.

Daniel Andrew Coulson, Martin T. Wells

In this paper we propose a novel problem called the ForeClassing problem where the loss of a classification decision is only observed at a future time point after the classification decision has to be made. To solve this problem, we propose an approximately Bayesian deep neural network architecture called ForeClassNet for time series forecasting and classification. This network architecture forces the network to consider possible future realizations of the time series, by forecasting future time points and their likelihood of occurring, before making its final classification decision. To facilitate this, we introduce two novel neural network layers, Welford mean-variance layers and Boltzmann convolutional layers. Welford mean-variance layers allow networks to iteratively update their estimates of the mean and variance for the forecasted time points for each inputted time series to the network through successive forward passes, which the model can then consider in combination with a learned representation of the observed realizations of the time series for its classification decision. Boltzmann convolutional layers are linear combinations of approximately Bayesian convolutional layers with different filter lengths, allowing the model to learn multitemporal resolution representations of the input time series, and which resolutions to focus on within a given Boltzmann convolutional layer through a Boltzmann distribution. Through several simulation scenarios and two real world applications we demonstrate ForeClassNet achieves superior performance compared with current state of the art methods including a near 30% improvement in test set accuracy in our financial example compared to the second best performing model.

Liao Zhu, Ningning Sun, Martin T. Wells

This paper builds the clustering model of measures of market microstructure features which are popular in predicting stock returns. In a 10-second time-frequency, we study the clustering structure of different measures to find out the best ones for predicting. In this way, we can predict more accurately with a limited number of predictors, which removes the noise and makes the model more interpretable.

Liao Zhu, Haoxuan Wu, Martin T. Wells

The paper proposes a new asset pricing model -- the News Embedding UMAP Selection (NEUS) model, to explain and predict the stock returns based on the financial news. Using a combination of various machine learning algorithms, we first derive a company embedding vector for each basis asset from the financial news. Then we obtain a collection of the basis assets based on their company embedding. After that for each stock, we select the basis assets to explain and predict the stock return with high-dimensional statistical methods. The new model is shown to have a significantly better fitting and prediction power than the Fama-French 5-factor model.

Liao Zhu, Robert A. Jarrow, Martin T. Wells

The purpose of this paper is to test the time-invariance of the beta coefficients estimated by the Adaptive Multi-Factor (AMF) model. The AMF model is implied by the generalized arbitrage pricing theory (GAPT), which implies constant beta coefficients. The AMF model utilizes a Groupwise Interpretable Basis Selection (GIBS) algorithm to identify the relevant factors from among all traded ETFs. We compare the AMF model with the Fama-French 5-factor (FF5) model. We show that for nearly all time periods with length less than 6 years, the beta coefficients are time-invariant for the AMF model, but not for the FF5 model. This implies that the AMF model with a rolling window (such as 5 years) is more consistent with realized asset returns than is the FF5 model.

Daqian Sun, Martin T. Wells

We consider the matrix completion problem of recovering a structured low rank matrix with partially observed entries with mixed data types. Vast majority of the solutions have proposed computationally feasible estimators with strong statistical guarantees for the case where the underlying distribution of data in the matrix is continuous. A few recent approaches have extended using similar ideas these estimators to the case where the underlying distributions belongs to the exponential family. Most of these approaches assume that there is only one underlying distribution and the low rank constraint is regularized by the matrix Schatten Norm. We propose a computationally feasible statistical approach with strong recovery guarantees along with an algorithmic framework suited for parallelization to recover a low rank matrix with partially observed entries for mixed data types in one step. We also provide extensive simulation evidence that corroborate our theoretical results.

Robert A. Jarrow, Rinald Murataj, Martin T. Wells et al.

The paper provides a new explanation of the low-volatility anomaly. We use the Adaptive Multi-Factor (AMF) model estimated by the Groupwise Interpretable Basis Selection (GIBS) algorithm to find those basis assets significantly related to low and high volatility portfolios. These two portfolios load on very different factors, indicating that volatility is not an independent risk, but that it's related to existing risk factors. The out-performance of the low-volatility portfolio is due to the (equilibrium) performance of these loaded risk factors. The AMF model outperforms the Fama-French 5-factor model both in-sample and out-of-sample.

Benjamin R. Baer, Daniel E. Gilbert, Martin T. Wells

A substantial portion of the literature on fairness in algorithms proposes, analyzes, and operationalizes simple formulaic criteria for assessing fairness. Two of these criteria, Equalized Odds and Calibration by Group, have gained significant attention for their simplicity and intuitive appeal, but also for their incompatibility. This chapter provides a perspective on the meaning and consequences of these and other fairness criteria using graphical models which reveals Equalized Odds and related criteria to be ultimately misleading. An assessment of various graphical models suggests that fairness criteria should ultimately be case-specific and sensitive to the nature of the information the algorithm processes.

Liao Zhu, Sumanta Basu, Robert A. Jarrow et al.

The paper proposes a new algorithm for the high-dimensional financial data -- the Groupwise Interpretable Basis Selection (GIBS) algorithm, to estimate a new Adaptive Multi-Factor (AMF) asset pricing model, implied by the recently developed Generalized Arbitrage Pricing Theory, which relaxes the convention that the number of risk-factors is small. We first obtain an adaptive collection of basis assets and then simultaneously test which basis assets correspond to which securities, using high-dimensional methods. The AMF model, along with the GIBS algorithm, is shown to have a significantly better fitting and prediction power than the Fama-French 5-factor model.

Skyler Seto, Sarah Tan, Giles Hooker et al.

Non-negative matrix factorization (NMF) is a technique for finding latent representations of data. The method has been applied to corpora to construct topic models. However, NMF has likelihood assumptions which are often violated by real document corpora. We present a double parametric bootstrap test for evaluating the fit of an NMF-based topic model based on the duality of the KL divergence and Poisson maximum likelihood estimation. The test correctly identifies whether a topic model based on an NMF approach yields reliable results in simulated and real data.

Sarah Tan, Matvey Soloviev, Giles Hooker et al.

Ensembles of decision trees perform well on many problems, but are not interpretable. In contrast to existing approaches in interpretability that focus on explaining relationships between features and predictions, we propose an alternative approach to interpret tree ensemble classifiers by surfacing representative points for each class -- prototypes. We introduce a new distance for Gradient Boosted Tree models, and propose new, adaptive prototype selection methods with theoretical guarantees, with the flexibility to choose a different number of prototypes in each class. We demonstrate our methods on random forests and gradient boosted trees, showing that the prototypes can perform as well as or even better than the original tree ensemble when used as a nearest-prototype classifier. In a user study, humans were better at predicting the output of a tree ensemble classifier when using prototypes than when using Shapley values, a popular feature attribution method. Hence, prototypes present a viable alternative to feature-based explanations for tree ensembles.

Irina Gaynanova, James Booth, Martin T. Wells

We investigate the difference between using an $\ell_1$ penalty versus an $\ell_1$ constraint in generalized eigenvalue problems, such as principal component analysis and discriminant analysis. Our main finding is that an $\ell_1$ penalty may fail to provide very sparse solutions; a severe disadvantage for variable selection that can be remedied by using an $\ell_1$ constraint. Our claims are supported both by empirical evidence and theoretical analysis. Finally, we illustrate the advantages of an $\ell_1$ constraint in the context of discriminant analysis and principal component analysis.

Irina Gaynanova, James G. Booth, Martin T. Wells

This article considers the problem of sparse estimation of canonical vectors in linear discriminant analysis when $p\gg N$. Several methods have been proposed in the literature that estimate one canonical vector in the two-group case. However, $G-1$ canonical vectors can be considered if the number of groups is $G$. In the multi-group context, it is common to estimate canonical vectors in a sequential fashion. Moreover, separate prior estimation of the covariance structure is often required. We propose a novel methodology for direct estimation of canonical vectors. In contrast to existing techniques, the proposed method estimates all canonical vectors at once, performs variable selection across all the vectors and comes with theoretical guarantees on the variable selection and classification consistency. First, we highlight the fact that in the $N>p$ setting the canonical vectors can be expressed in a closed form up to an orthogonal transformation. Secondly, we propose an extension of this form to the $p\gg N$ setting and achieve feature selection by using a group penalty. The resulting optimization problem is convex and can be solved using a block-coordinate descent algorithm. The practical performance of the method is evaluated through simulation studies as well as real data applications.

Irina Gaynanova, James G. Booth, Martin T. Wells

It is well known that in a supervised classification setting when the number of features is smaller than the number of observations, Fisher's linear discriminant rule is asymptotically Bayes. However, there are numerous modern applications where classification is needed in the high-dimensional setting. Naive implementation of Fisher's rule in this case fails to provide good results because the sample covariance matrix is singular. Moreover, by constructing a classifier that relies on all features the interpretation of the results is challenging. Our goal is to provide robust classification that relies only on a small subset of important features and accounts for the underlying correlation structure. We apply a lasso-type penalty to the discriminant vector to ensure sparsity of the solution and use a shrinkage type estimator for the covariance matrix. The resulting optimization problem is solved using an iterative coordinate ascent algorithm. Furthermore, we analyze the effect of nonconvexity on the sparsity level of the solution and highlight the difference between the penalized and the constrained versions of the problem. The simulation results show that the proposed method performs favorably in comparison to alternatives. The method is used to classify leukemia patients based on DNA methylation features.