arXiv subjects

Benjamin Letham

Publications and source records attributed to Benjamin Letham.

At least 19 recordsLinked to original sources

arXiv2026

Offline-to-Online Creative Optimization with Generative Models and Adaptive Testing

Ad creative optimization is increasingly constrained by evaluation rather than generation. Generative models can produce many plausible creatives, but reliable evaluation requires online experiments, in which only a limited slate can be tested. We study how to use data from historical A/B tests to generate and select the candidates in that slate. We developed and deployed a performance-driven offline-to-online workflow that guides creative generation with a predictive model as an inference-time critic. In the offline phase, we use a predictive model trained on historical experiments to rank and refine variants created by a generative model. A final test slate is then deployed in an online adaptive experiment. In a 50-arm field experiment, we found that the best creative generated with this method yielded 45.1% higher engagement than the best human-authored creative. Two additional experiments showed the same upper-tail pattern, with lifts of 46.7% and 36.2%. We found that despite the predictive model being too noisy to directly identify the best creative offline, it effectively guides the generative model toward creating strong candidates that can be efficiently evaluated in an adaptive experiment. The results suggest a design principle for creative optimization with generative models: use predictive models to guide generation of a slate to test, judge the slate by whether it contains high-performing candidates at a feasible test size, and use adaptive experiments to select among candidates while limiting traffic lost to weak arms.

cs.AI↗

arXiv2026

Embedding by Elicitation: Dynamic Representations for Bayesian Optimization of System Prompts

System prompts are a central control mechanism in modern AI systems, shaping behavior across conversations, tasks, and user populations. Yet they are difficult to tune when feedback is available only as aggregate metrics rather than per-example labels, failures, or critiques. We study this aggregate feedback setting as sample-constrained black-box optimization over discrete, variable-length text. We introduce ReElicit, a Bayesian optimization framework based on \emph{embedding by elicitation}. Given a task description, previously evaluated prompts, and scalar scores, an LLM elicits a compact, interpretable feature space and maps prompts into it. Leveraging a probabilistic Gaussian process surrogate, an acquisition function then selects target feature vectors, which the LLM realizes and refines into deployable system prompts. Re-eliciting the feature space as new evaluations arrive lets the representation adapt to the observed prompt-score history. We evaluate the setting using offline benchmark accuracy as a controlled aggregate proxy: the optimizer observes one scalar score per prompt and no per-example labels, errors, or critiques. Across ten system prompt optimization tasks with a 30 total evaluation budget, ReElicit achieves the strongest aggregate performance profile among representative aggregate-only prompt-optimization baselines. These results suggest that LLMs can serve as adaptive semantic representation builders, not only prompt generators, for Bayesian optimization over natural-language artifacts.

cs.AI↗

arXiv2025

LILO: Bayesian Optimization with Natural Language Feedback

Many real-world optimization problems are guided by complex, subjective preferences that are difficult to express as explicit closed-form objectives. In response, we introduce Language-in-the-Loop Optimization (LILO), a Bayesian optimization (BO) framework that employs a large language model (LLM) to translate free-form natural language feedback and prior knowledge from a decision maker into structured preference signals, going beyond the restrictive scalar or pairwise feedback formats typically assumed in preferential BO. The LLM-derived preferences are integrated by a Gaussian process proxy model, enabling principled acquisition-driven exploration with calibrated uncertainty. By placing the LLM in a supporting role rather than as the optimizer itself, LILO preserves the sample efficiency and stability of BO while providing a flexible and expressive feedback interface. Across synthetic and real-world benchmarks, LILO consistently outperforms both conventional preference-based BO methods and LLM-only optimizers, with particularly strong gains in feedback-limited regimes.

cs.LG↗

arXiv2025

Experimenting, Fast and Slow: Bayesian Optimization of Long-term Outcomes with Online Experiments

Online experiments in internet systems, also known as A/B tests, are used for a wide range of system tuning problems, such as optimizing recommender system ranking policies and learning adaptive streaming controllers. Decision-makers generally wish to optimize for long-term treatment effects of the system changes, which often requires running experiments for a long time as short-term measurements can be misleading due to non-stationarity in treatment effects over time. The sequential experimentation strategies--which typically involve several iterations--can be prohibitively long in such cases. We describe a novel approach that combines fast experiments (e.g., biased experiments run only for a few hours or days) and/or offline proxies (e.g., off-policy evaluation) with long-running, slow experiments to perform sequential, Bayesian optimization over large action spaces in a short amount of time.

cs.LG↗

arXiv2025

Mixed Likelihood Variational Gaussian Processes

Gaussian processes (GPs) are powerful models for human-in-the-loop experiments due to their flexibility and well-calibrated uncertainty. However, GPs modeling human responses typically ignore auxiliary information, including a priori domain expertise and non-task performance information like user confidence ratings. We propose mixed likelihood variational GPs to leverage auxiliary information, which combine multiple likelihoods in a single evidence lower bound to model multiple types of data. We demonstrate the benefits of mixing likelihoods in three real-world experiments with human participants. First, we use mixed likelihood training to impose prior knowledge constraints in GP classifiers, which accelerates active learning in a visual perception task where users are asked to identify geometric errors resulting from camera position errors in virtual reality. Second, we show that leveraging Likert scale confidence ratings by mixed likelihood training improves model fitting for haptic perception of surface roughness. Lastly, we show that Likert scale confidence ratings improve human preference learning in robot gait optimization. The modeling performance improvements found using our framework across this diverse set of applications illustrates the benefits of incorporating auxiliary information into active learning and preference learning by using mixed likelihoods to jointly model multiple inputs.

cs.LG↗

arXiv2024

Active Learning for Derivative-Based Global Sensitivity Analysis with Gaussian Processes

We consider the problem of active learning for global sensitivity analysis of expensive black-box functions. Our aim is to efficiently learn the importance of different input variables, e.g., in vehicle safety experimentation, we study the impact of the thickness of various components on safety objectives. Since function evaluations are expensive, we use active learning to prioritize experimental resources where they yield the most value. We propose novel active learning acquisition functions that directly target key quantities of derivative-based global sensitivity measures (DGSMs) under Gaussian process surrogate models. We showcase the first application of active learning directly to DGSMs, and develop tractable uncertainty reduction and information gain acquisition functions for these measures. Through comprehensive evaluation on synthetic and real-world problems, our study demonstrates how these active learning acquisition strategies substantially enhance the sample efficiency of DGSM estimation, particularly with limited evaluation budgets. Our work paves the way for more efficient and accurate sensitivity analysis in various scientific and engineering applications.

cs.LG↗

arXiv2023

Response Time Improves Choice Prediction and Function Estimation for Gaussian Process Models of Perception and Preferences

Models for human choice prediction in preference learning and psychophysics often consider only binary response data, requiring many samples to accurately learn preferences or perceptual detection thresholds. The response time (RT) to make each choice captures additional information about the decision process, however existing models incorporating RTs for choice prediction do so in fully parametric settings or over discrete stimulus sets. This is in part because the de-facto standard model for choice RTs, the diffusion decision model (DDM), does not admit tractable, differentiable inference. The DDM thus cannot be easily integrated with flexible models for continuous, multivariate function approximation, particularly Gaussian process (GP) models. We propose a novel differentiable approximation to the DDM likelihood using a family of known, skewed three-parameter distributions. We then use this new likelihood to incorporate RTs into GP models for binary choices. Our RT-choice GPs enable both better latent value estimation and held-out choice prediction relative to baselines, which we demonstrate on three real-world multivariate datasets covering both human psychophysics and preference learning applications.

q-bio.NC↗

arXiv2023

Perceptual Requirements for World-Locked Rendering in AR and VR

Stereoscopic, head-tracked display systems can show users realistic, world-locked virtual objects and environments. However, discrepancies between the rendering pipeline and physical viewing conditions can lead to perceived instability in the rendered content resulting in reduced immersion and, potentially, visually-induced motion sickness. Precise requirements to achieve perceptually stable world-locked rendering (WLR) are unknown due to the challenge of constructing a wide field of view, distortion-free display with highly accurate head and eye tracking. We present a system capable of rendering virtual objects over real-world references without perceivable drift under such constraints. This platform is used to study acceptable errors in render camera position for WLR in augmented and virtual reality scenarios, where we find an order of magnitude difference in perceptual sensitivity. We conclude with an analytic model which examines changes to apparent depth and visual direction in response to camera displacement errors.

cs.GR↗

arXiv2023

A Semi-Parametric Model for Decision Making in High-Dimensional Sensory Discrimination Tasks

Psychometric functions typically characterize binary sensory decisions along a single stimulus dimension. However, real-life sensory tasks vary along a greater variety of dimensions (e.g. color, contrast and luminance for visual stimuli). Approaches to characterizing high-dimensional sensory spaces either require strong parametric assumptions about these additional contextual dimensions, or fail to leverage known properties of classical psychometric curves. We overcome both limitations by introducing a semi-parametric model of sensory discrimination that applies traditional psychophysical models along a stimulus intensity dimension, but puts Gaussian process (GP) priors on the parameters of these models with respect to the remaining dimensions. By combining the flexibility of the GP with the deep literature on parametric psychophysics, our semi-parametric models achieve good performance with much less data than baselines on both synthetic and real-world high-dimensional psychophysics datasets. We additionally show strong performance in a Bayesian active learning setting, and present a novel active learning paradigm for the semi-parametric model.

q-bio.NC↗

arXiv2022

Look-Ahead Acquisition Functions for Bernoulli Level Set Estimation

Level set estimation (LSE) is the problem of identifying regions where an unknown function takes values above or below a specified threshold. Active sampling strategies for efficient LSE have primarily been studied in continuous-valued functions. Motivated by applications in human psychophysics where common experimental designs produce binary responses, we study LSE active sampling with Bernoulli outcomes. With Gaussian process classification surrogate models, the look-ahead model posteriors used by state-of-the-art continuous-output methods are intractable. However, we derive analytic expressions for look-ahead posteriors of sublevel set membership, and show how these lead to analytic expressions for a class of look-ahead LSE acquisition functions, including information-based methods. Benchmark experiments show the importance of considering the global look-ahead impact on the entire posterior. We demonstrate a clear benefit to using this new class of acquisition functions on benchmark problems, and on a challenging real-world task of estimating a high-dimensional contrast sensitivity function.

stat.ML↗

arXiv2022

Sparse Bayesian Optimization

Bayesian optimization (BO) is a powerful approach to sample-efficient optimization of black-box objective functions. However, the application of BO to areas such as recommendation systems often requires taking the interpretability and simplicity of the configurations into consideration, a setting that has not been previously studied in the BO literature. To make BO useful for this setting, we present several regularization-based approaches that allow us to discover sparse and more interpretable configurations. We propose a novel differentiable relaxation based on homotopy continuation that makes it possible to target sparsity by working directly with $L_0$ regularization. We identify failure modes for regularized BO and develop a hyperparameter-free method, sparsity exploring Bayesian optimization (SEBO) that seeks to simultaneously maximize a target objective and sparsity. SEBO and methods based on fixed regularization are evaluated on synthetic and real-world problems, and we show that we are able to efficiently optimize for sparsity.

cs.LG↗

arXiv2021

Adaptive Nonparametric Psychophysics

We introduce a new set of models and adaptive psychometric testing methods for multidimensional psychophysics. In contrast to traditional adaptive staircase methods like PEST and QUEST, the method is multi-dimensional and does not require a grid over contextual dimensions, retaining sub-exponential scaling in the number of stimulus dimensions. In contrast to more recent multi-dimensional adaptive methods, our underlying model does not require a parametric assumption about the interaction between intensity and the additional dimensions. In addition, we introduce a new active sampling policy that explicitly targets psychometric detection threshold estimation and does so substantially faster than policies that attempt to estimate the full psychometric function (though it still provides estimates of the function, albeit with lower accuracy). Finally, we introduce AEPsych, a user-friendly open-source package for nonparametric psychophysics that makes these technically-challenging methods accessible to the broader community.

stat.ME↗

arXiv2020

Re-Examining Linear Embeddings for High-Dimensional Bayesian Optimization

Bayesian optimization (BO) is a popular approach to optimize expensive-to-evaluate black-box functions. A significant challenge in BO is to scale to high-dimensional parameter spaces while retaining sample efficiency. A solution considered in existing literature is to embed the high-dimensional space in a lower-dimensional manifold, often via a random linear embedding. In this paper, we identify several crucial issues and misconceptions about the use of linear embeddings for BO. We study the properties of linear embeddings from the literature and show that some of the design choices in current approaches adversely impact their performance. We show empirically that properly addressing these issues significantly improves the efficacy of linear embeddings for BO on a range of problems, including learning a gait policy for robot locomotion.

stat.ML↗

arXiv2019

BoTorch: A Framework for Efficient Monte-Carlo Bayesian Optimization

Bayesian optimization provides sample-efficient global optimization for a broad range of applications, including automatic machine learning, engineering, physics, and experimental design. We introduce BoTorch, a modern programming framework for Bayesian optimization that combines Monte-Carlo (MC) acquisition functions, a novel sample average approximation optimization approach, auto-differentiation, and variance reduction techniques. BoTorch's modular design facilitates flexible specification and optimization of probabilistic models written in PyTorch, simplifying implementation of new acquisition functions. Our approach is backed by novel theoretical convergence results and made practical by a distinctive algorithmic foundation that leverages fast predictive distributions, hardware acceleration, and deterministic optimization. We also propose a novel "one-shot" formulation of the Knowledge Gradient, enabled by a combination of our theoretical and software contributions. In experiments, we demonstrate the improved sample efficiency of BoTorch relative to other popular libraries.

cs.LG↗

arXiv2019

Bayesian Optimization for Policy Search via Online-Offline Experimentation

Online field experiments are the gold-standard way of evaluating changes to real-world interactive machine learning systems. Yet our ability to explore complex, multi-dimensional policy spaces - such as those found in recommendation and ranking problems - is often constrained by the limited number of experiments that can be run simultaneously. To alleviate these constraints, we augment online experiments with an offline simulator and apply multi-task Bayesian optimization to tune live machine learning systems. We describe practical issues that arise in these types of applications, including biases that arise from using a simulator and assumptions for the multi-task kernel. We measure empirical learning curves which show substantial gains from including data from biased offline experiments, and show how these learning curves are consistent with theoretical results for multi-task Gaussian process generalization. We find that improved kernel inference is a significant driver of multi-task generalization. Finally, we show several examples of Bayesian optimization efficiently tuning a live machine learning system by combining offline and online experiments.

stat.ML↗

arXiv2018

Practical Transfer Learning for Bayesian Optimization

When hyperparameter optimization of a machine learning algorithm is repeated for multiple datasets it is possible to transfer knowledge to an optimization run on a new dataset. We develop a new hyperparameter-free ensemble model for Bayesian optimization that is a generalization of two existing transfer learning extensions to Bayesian optimization and establish a worst-case bound compared to vanilla Bayesian optimization. Using a large collection of hyperparameter optimization benchmark problems, we demonstrate that our contributions substantially reduce optimization time compared to standard Gaussian process-based Bayesian optimization and improve over the current state-of-the-art for transfer hyperparameter optimization.

stat.ML↗

arXiv2017

Constrained Bayesian Optimization with Noisy Experiments

Randomized experiments are the gold standard for evaluating the effects of changes to real-world systems. Data in these tests may be difficult to collect and outcomes may have high variance, resulting in potentially large measurement error. Bayesian optimization is a promising technique for efficiently optimizing multiple continuous parameters, but existing approaches degrade in performance when the noise level is high, limiting its applicability to many randomized experiments. We derive an expression for expected improvement under greedy batch optimization with noisy observations and noisy constraints, and develop a quasi-Monte Carlo approximation that allows it to be efficiently optimized. Simulations with synthetic functions show that optimization performance on noisy, constrained problems outperforms existing methods. We further demonstrate the effectiveness of the method with two real-world experiments conducted at Facebook: optimizing a ranking system, and optimizing server compiler flags.

stat.ML↗

arXiv2015

Prediction uncertainty and optimal experimental design for learning dynamical systems

Dynamical systems are frequently used to model biological systems. When these models are fit to data it is necessary to ascertain the uncertainty in the model fit. Here we present prediction deviation, a new metric of uncertainty that determines the extent to which observed data have constrained the model's predictions. This is accomplished by solving an optimization problem that searches for a pair of models that each provide a good fit for the observed data, yet have maximally different predictions. We develop a method for estimating a priori the impact that additional experiments would have on the prediction deviation, allowing the experimenter to design a set of experiments that would most reduce uncertainty. We use prediction deviation to assess uncertainty in a model of interferon-alpha inhibition of viral infection, and to select a sequence of experiments that reduces this uncertainty. Finally we prove a theoretical result which shows that prediction deviation provides bounds on the trajectories of the underlying true model. These results show that prediction deviation is a meaningful metric of uncertainty that can be used for optimal experimental design.

stat.AP↗