arXiv subjects

Philip Greengard

Publications and source records attributed to Philip Greengard.

At least 19 recordsLinked to original sources

arXiv2026

The Squealer: Sensification of model exploration and model misfit

We introduce a method for visual and auditory feedback when exploring the fit of a model to data. Starting with a best-fit curve fit to data, the user can drag the curve to a new position and the computer will emit a squeal, becoming louder and more unpleasant as the discrepancy between curve and data increases. We demonstrate with four examples: a two-parameter curve fit to golf putting data, a four-parameter curve fit to dilution assays, a fit to cosmological data sensitive to the parameters of the Big Bang model, and a nonparametric Gaussian process fit to temperature readings.

physics.data-an↗

arXiv2026

Approximate posterior recalibration

Bayesian inference is often implemented using approximations, which can yield interval estimates that are too narrow, not fully capturing the uncertainty in the posterior distribution. We address the question of how to adjust these approximate posteriors so that they appropriately capture uncertainty. vWe introduce two methods that extend simulation-based calibration checking (SBC) to widen approximate posterior uncertainty intervals to aim for marginal calibration. We demonstrate these methods in several experimental settings, and we discuss the challenge of calibration using posterior inferences and the potential for posterior recalibration of hierarchical models.

stat.ME↗

arXiv2025

A Network of Biologically Inspired Rectified Spectral Units (ReSUs) Learns Hierarchical Features Without Error Backpropagation

We introduce a biologically inspired, multilayer neural architecture composed of Rectified Spectral Units (ReSUs). Each ReSU projects a recent window of its input history onto a canonical direction obtained via canonical correlation analysis (CCA) of previously observed past-future input pairs, and then rectifies either its positive or negative component. By encoding canonical directions in synaptic weights and temporal filters, ReSUs implement a local, self-supervised algorithm for progressively constructing increasingly complex features. To evaluate both computational power and biological fidelity, we trained a two-layer ReSU network in a self-supervised regime on translating natural scenes. First-layer units, each driven by a single pixel, developed temporal filters resembling those of Drosophila post-photoreceptor neurons (L1/L2 and L3), including their empirically observed adaptation to signal-to-noise ratio (SNR). Second-layer units, which pooled spatially over the first layer, became direction-selective -- analogous to T4 motion-detecting cells -- with learned synaptic weight patterns approximating those derived from connectomic reconstructions. Together, these results suggest that ReSUs offer (i) a principled framework for modeling sensory circuits and (ii) a biologically grounded, backpropagation-free paradigm for constructing deep self-supervised neural networks.

q-bio.NC↗

arXiv2025

Neurons as Detectors of Coherent Sets in Sensory Dynamics

We model sensory streams as observations from high-dimensional stochastic dynamical systems and conceptualize sensory neurons as self-supervised learners of compact representations of such dynamics. From prior experience, neurons learn coherent sets-regions of stimulus state space whose trajectories evolve cohesively over finite times-and assign membership indices to new stimuli. Coherent sets are identified via spectral clustering of the stochastic Koopman operator (SKO), where the sign pattern of a subdominant singular function partitions the state space into minimally coupled regions. For multivariate Ornstein-Uhlenbeck processes, this singular function reduces to a linear projection onto the dominant singular vector of the whitened state-transition matrix. Encoding this singular vector as a receptive field enables neurons to compute membership indices via the projection sign in a biologically plausible manner. Each neuron detects either a predictive coherent set (stimuli with common futures) or a retrospective coherent set (stimuli with common pasts), suggesting a functional dichotomy among neurons. Since neurons lack access to explicit dynamical equations, the requisite singular vectors must be estimated directly from data, for example, via past-future canonical correlation analysis on lag-vector representations-an approach that naturally extends to nonlinear dynamics. This framework provides a novel account of neuronal temporal filtering, the ubiquity of rectification in neural responses, and known functional dichotomies. Coherent-set clustering thus emerges as a fundamental computation underlying sensory processing and transferable to bio-inspired artificial systems.

q-bio.NC↗

arXiv2025

A Bayesian analysis of home advantage in professional squash

We estimate the effect of playing in one's home country in professional squash using a Bayesian hierarchical model applied to men's and women's Professional Squash Association matches from 2018-2024. The model incorporates players' world rankings and whether they are competing in their home country. Using margin of victory in games as our outcome, we estimate that home advantage adds 0.4 games for men and 0.3 games for women to the expected margin, with standard errors of 0.1. For evenly matched players, this effect corresponds to an increase in win probability from 50% to roughly 58% for men and 56% for women. We estimate particularly strong home effects in Egypt, where many major tournaments are held, though data limitations prevent precise estimation of country-specific effects in many other nations.

stat.AP↗

arXiv2025

A calibrated BISG for inferring race from surname and geolocation

Bayesian Improved Surname Geocoding (BISG) is a ubiquitous tool for predicting race and ethnicity using an individual's geolocation and surname. Here we demonstrate that statistical dependence of surname and geolocation within racial/ethnic categories in the United States results in biases for minority subpopulations, and we introduce a raking-based improvement. Our method augments the data used by BISG--distributions of race by geolocation and race by surname--with the distribution of surname by geolocation obtained from state voter files. We validate our algorithm on state voter registration lists that contain self-identified race/ethnicity.

stat.AP↗

arXiv2024

LoRA Learns Less and Forgets Less

Low-Rank Adaptation (LoRA) is a widely-used parameter-efficient finetuning method for large language models. LoRA saves memory by training only low rank perturbations to selected weight matrices. In this work, we compare the performance of LoRA and full finetuning on two target domains, programming and mathematics. We consider both the instruction finetuning (approximately 100K prompt-response pairs) and continued pretraining (20B unstructured tokens) data regimes. Our results show that, in the standard low-rank settings, LoRA substantially underperforms full finetuning. Nevertheless, LoRA better maintains the base model's performance on tasks outside the target domain. We show that LoRA mitigates forgetting more than common regularization techniques such as weight decay and dropout; it also helps maintain more diverse generations. Finally, we show that full finetuning learns perturbations with a rank that is 10-100X greater than typical LoRA configurations, possibly explaining some of the reported gaps. We conclude by proposing best practices for finetuning with LoRA.

cs.LG↗

arXiv2024

LQ-LoRA: Low-rank Plus Quantized Matrix Decomposition for Efficient Language Model Finetuning

We propose a simple approach for memory-efficient adaptation of pretrained language models. Our approach uses an iterative algorithm to decompose each pretrained matrix into a high-precision low-rank component and a memory-efficient quantized component. During finetuning, the quantized component remains fixed and only the low-rank component is updated. We present an integer linear programming formulation of the quantization component which enables dynamic configuration of quantization parameters (e.g., bit-width, block size) for each matrix given an overall target memory budget. We further explore a data-aware version of the algorithm which uses an approximation of the Fisher information matrix to weight the reconstruction objective during matrix decomposition. Experiments on finetuning RoBERTa and LLaMA-2 (7B and 70B) demonstrate that our low-rank plus quantized matrix decomposition approach (LQ-LoRA) outperforms strong QLoRA and GPTQ-LoRA baselines and enables aggressive quantization to sub-3 bits with only minor performance degradations. When finetuned on a language modeling calibration dataset, LQ-LoRA can also be used for model compression; in this setting our 2.75-bit LLaMA-2-70B model (which has 2.85 bits on average when including the low-rank components and requires 27GB of GPU memory) performs respectably compared to the 16-bit baseline.

cs.CL↗

arXiv2024

The piranha problem: Large effects swimming in a small pond

In some scientific fields, it is common to have certain variables of interest that are of particular importance and for which there are many studies indicating a relationship with different explanatory variables. In such cases, particularly those where no relationships are known among the explanatory variables, it is worth asking under what conditions it is possible for all such claimed effects to exist simultaneously. This paper addresses this question by reviewing some theorems from multivariate analysis showing that, unless the explanatory variables also have sizable dependencies with each other, it is impossible to have many such large effects. We discuss implications for the replication crisis in social science.

math.ST↗

arXiv2024

Efficient Fourier representations of families of Gaussian processes

We introduce a class of algorithms for constructing Fourier representations of Gaussian processes in $1$ dimension that are valid over ranges of hyperparameter values. The scaling and frequencies of the Fourier basis functions are evaluated numerically via generalized quadratures. The representations introduced allow for $O(m^3)$ inference, independent of $N$, for all hyperparameters in the user-specified range after $O(N + m^2\log{m})$ precomputation where $N$, the number of data points, is usually significantly larger than $m$, the number of basis functions. Inference independent of $N$ for various hyperparameters is facilitated by generalized quadratures, and the $O(N + m^2\log{m})$ precomputation is achieved with the non-uniform FFT. Numerical results are provided for Matérn kernels with $ν\in [3/2, 7/2]$ and lengthscale $ρ\in [0.1, 0.5]$ and squared-exponential kernels with lengthscale $ρ\in [0.1, 0.5]$. The algorithms of this paper generalize mathematically to higher dimensions, though they suffer from the standard curse of dimensionality.

stat.CO↗

arXiv2024

Generalized prolate spheroidal functions: algorithms and analysis

Generalized prolate spheroidal functions (GPSFs) arise naturally in the study of bandlimited functions as the eigenfunctions of a certain truncated Fourier transform. In one dimension, the theory of GPSFs (typically referred to as prolate spheroidal wave functions) has a long history and is fairly complete. Furthermore, more recent work has led to the development of numerical algorithms for their computation and use in applications. In this paper we consider the more general problem, extending the one dimensional analysis and algorithms to the case of arbitrary dimension. Specifically, we introduce algorithms for efficient evaluation of GPSFs and their corresponding eigenvalues, quadrature rules for bandlimited functions, formulae for interpolation via GPSF expansion, and various analytical properties of GPSFs. We illustrate the numerical and analytical results with several numerical examples.

math.NA↗

arXiv2024

Hierarchical Bayesian Models to Mitigate Systematic Disparities in Prediction with Proxy Outcomes

Label bias occurs when the outcome of interest is not directly observable and instead, modeling is performed with proxy labels. When the difference between the true outcome and the proxy label is correlated with predictors, this can yield systematic disparities in predictions for different groups of interest. We propose Bayesian hierarchical measurement models to address these issues. When strong prior information about the measurement process is available, our approach improves accuracy and helps with algorithmic fairness. If prior knowledge is limited, our approach allows assessment of the sensitivity of predictions to the unknown specifications of the measurement process. This can help practitioners gauge if enough substantive information is available to guarantee the desired accuracy and avoid disparate predictions when using proxy outcomes. We demonstrate our approach through practical examples.

stat.ME↗

arXiv2023

Equispaced Fourier representations for efficient Gaussian process regression from a billion data points

We introduce a Fourier-based fast algorithm for Gaussian process regression in low dimensions. It approximates a translationally-invariant covariance kernel by complex exponentials on an equispaced Cartesian frequency grid of $M$ nodes. This results in a weight-space $M\times M$ system matrix with Toeplitz structure, which can thus be applied to a vector in ${\mathcal O}(M \log{M})$ operations via the fast Fourier transform (FFT), independent of the number of data points $N$. The linear system can be set up in ${\mathcal O}(N + M \log{M})$ operations using nonuniform FFTs. This enables efficient massive-scale regression via an iterative solver, even for kernels with fat-tailed spectral densities (large $M$). We provide bounds on both kernel approximation and posterior mean errors. Numerical experiments for squared-exponential and Matérn kernels in one, two and three dimensions often show 1-2 orders of magnitude acceleration over state-of-the-art rank-structured solvers at comparable accuracy. Our method allows 2D Matérn-$\mbox{$\frac{3}{2}$}$ regression from $N=10^9$ data points to be performed in 2 minutes on a standard desktop, with posterior mean accuracy $10^{-3}$. This opens up spatial statistics applications 100 times larger than previously possible.

stat.CO↗

arXiv2023

Uniform approximation of common Gaussian process kernels using equispaced Fourier grids

The high efficiency of a recently proposed method for computing with Gaussian processes relies on expanding a (translationally invariant) covariance kernel into complex exponentials, with frequencies lying on a Cartesian equispaced grid. Here we provide rigorous error bounds for this approximation for two popular kernels -- Matérn and squared exponential -- in terms of the grid spacing and size. The kernel error bounds are uniform over a hypercube centered at the origin. Our tools include a split into aliasing and truncation errors, and bounds on sums of Gaussians or modified Bessel functions over various lattices. For the Matérn case, motivated by numerical study, we conjecture a stronger Frobenius-norm bound on the covariance matrix error for randomly-distributed data points. Lastly, we prove bounds on, and study numerically, the ill-conditioning of the linear systems arising in such regression problems.

math.NA↗

arXiv2023

Learning to Grow Pretrained Models for Efficient Transformer Training

Scaling transformers has led to significant breakthroughs in many domains, leading to a paradigm in which larger versions of existing models are trained and released on a periodic basis. New instances of such models are typically trained completely from scratch, despite the fact that they are often just scaled-up versions of their smaller counterparts. How can we use the implicit knowledge in the parameters of smaller, extant models to enable faster training of newer, larger models? This paper describes an approach for accelerating transformer training by learning to grow pretrained transformers, where we learn to linearly map the parameters of the smaller model to initialize the larger model. For tractable learning, we factorize the linear transformation as a composition of (linear) width- and depth-growth operators, and further employ a Kronecker factorization of these growth operators to encode architectural knowledge. Extensive experiments across both language and vision transformers demonstrate that our learned Linear Growth Operator (LiGO) can save up to 50% computational cost of training from scratch, while also consistently outperforming strong baselines that also reuse smaller pretrained models to initialize larger models.

cs.LG↗

arXiv2023

Federated Learning as Variational Inference: A Scalable Expectation Propagation Approach

The canonical formulation of federated learning treats it as a distributed optimization problem where the model parameters are optimized against a global loss function that decomposes across client loss functions. A recent alternative formulation instead treats federated learning as a distributed inference problem, where the goal is to infer a global posterior from partitioned client data (Al-Shedivat et al., 2021). This paper extends the inference view and describes a variational inference formulation of federated learning where the goal is to find a global variational posterior that well-approximates the true posterior. This naturally motivates an expectation propagation approach to federated learning (FedEP), where approximations to the global posterior are iteratively refined through probabilistic message-passing between the central server and the clients. We conduct an extensive empirical study across various algorithmic considerations and describe practical strategies for scaling up expectation propagation to the modern federated setting. We apply FedEP on standard federated learning benchmarks and find that it outperforms strong baselines in terms of both convergence speed and accuracy.

cs.LG↗

arXiv2022

Efficient reduced-rank methods for Gaussian processes with eigenfunction expansions

In this work we introduce a reduced-rank algorithm for Gaussian process regression. Our numerical scheme converts a Gaussian process on a user-specified interval to its Karhunen-Loève expansion, the $L^2$-optimal reduced-rank representation. Numerical evaluation of the Karhunen-Loève expansion is performed once during precomputation and involves computing a numerical eigendecomposition of an integral operator whose kernel is the covariance function of the Gaussian process. The Karhunen-Loève expansion is independent of observed data and depends only on the covariance kernel and the size of the interval on which the Gaussian process is defined. The scheme of this paper does not require translation invariance of the covariance kernel. We also introduce a class of fast algorithms for Bayesian fitting of hyperparameters, and demonstrate the performance of our algorithms with numerical experiments in one and two dimensions. Extensions to higher dimensions are mathematically straightforward but suffer from the standard curses of high dimensions.

stat.CO↗

arXiv2022

On a linearization of quadratic Wasserstein distance

This paper studies the problem of computing a linear approximation of quadratic Wasserstein distance $W_2$. In particular, we compute an approximation of the negative homogeneous weighted Sobolev norm whose connection to Wasserstein distance follows from a classic linearization of a general Monge-Ampére equation. Our contribution is threefold. First, we provide expository material on this classic linearization of Wasserstein distance including a quantitative error estimate. Second, we reduce the computational problem to solving an elliptic boundary value problem involving the Witten Laplacian, which is a Schrödinger operator of the form $H = -Δ+ V$, and describe an associated embedding. Third, for the case of probability distributions on the unit square $[0,1]^2$ represented by $n \times n$ arrays we present a fast code demonstrating our approach. Several numerical examples are presented.

math.NA↗