BoLT is a benchmark of surrogate models fitted to real LLM experiment data that enables evaluation of Bayesian and black-box optimization methods on multi-fidelity, multi-objective, high-dimensional LLM tasks.
hub Canonical reference
A Tutorial on Bayesian Optimization
Canonical reference. 90% of citing Pith papers cite this work as background.
68
Pith papers citing it
Background
90% of classified citations
abstract
Bayesian optimization is an approach to optimizing objective functions that take a long time (minutes or hours) to evaluate. It is best-suited for optimization over continuous domains of less than 20 dimensions, and tolerates stochastic noise in function evaluations. It builds a surrogate for the objective and quantifies the uncertainty in that surrogate using a Bayesian machine learning technique, Gaussian process regression, and then uses an acquisition function defined from this surrogate to decide where to sample. In this tutorial, we describe how Bayesian optimization works, including Gaussian process regression and three common acquisition functions: expected improvement, entropy search, and knowledge gradient. We then discuss more advanced techniques, including running multiple function evaluations in parallel, multi-fidelity and multi-information source optimization, expensive-to-evaluate constraints, random environmental conditions, multi-task Bayesian optimization, and the inclusion of derivative information. We conclude with a discussion of Bayesian optimization software and future research directions in the field. Within our tutorial material we provide a generalization of expected improvement to noisy evaluations, beyond the noise-free setting where it is more commonly applied. This generalization is justified by a formal decision-theoretic argument, standing in contrast to previous ad hoc modifications.
hub tools
citation-role summary
background 9
dataset 1
citation-polarity summary
representative citing papers
EFE-based planning is formulated as variational free energy minimization with epistemic priors, decomposing into expected plan costs plus a complexity term.
Conditional diffusion models trained with BO-aware strategies approximate the optimum distribution, enabling a Diffusion-based Mode Seeking acquisition function with a sub-optimality guarantee that outperforms baselines in experiments.
EFE-based active inference planning is characterized as VFE on an augmented model plus entropy and planning corrections, with a derived message-passing implementation and grid-world validation.
Dynamic reputation updates per objective-expert pair plus a three-arm counterfactual gate improve robustness over fixed LLM priors on synthetic tests and molecule benchmarks, but raw LLM confidence is not reliably helpful.
Graph-coupled causal Bayesian optimization couples intervention effects via shared causal parameters to produce a low-rank causal kernel, logarithmic information-gain bounds, and a regret bound separating optimization, estimation, and intervention-choice errors in linear Gaussian models.
First experimental realization of LEO-based adiabatic speedup on superconducting quantum processors, with enhanced fidelity using ideal pulses but limited gains from noise-model optimization.
BUP-TR completes underdetermined quadratic models via Bayesian projection in the prior precision norm, yielding fully linear hard-MAP models under stated conditions and attaining global first-order convergence with O(ε^{-2}) complexity.
CurveRL derives a quantile-coordinate reweighting rule from a utility functional on pass rates and shows it outperforms GRPO on reasoning benchmarks.
ShapeBench is a new benchmark suite for aerodynamic shape optimization with 103 tasks showing high variance in optimizer rankings across categories.
Bayesian optimization identifies cement-salt hydrate composites achieving up to five times higher specific energy than prior cement-based TCES materials, with LiCl-based formulations reaching 458 kJ/kg.
A cost-aware value-of-information acquisition function is derived to balance direct observations against noisy pairwise human comparisons in Bayesian optimization, approaching the convex hull of the individual information sources' performance trajectories.
Proposes a vector-valued RKHS framework for Bayesian optimization with structured measurements, deriving concentration bounds and UCB-based regret guarantees that recover sublinear rates.
A myopic MINMPC framework learns a value function offline via inverse optimization from expert data, allowing short horizons with near-optimal performance and strict integer feasibility online for hybrid systems.
PALM-Mean combines sign-aware piecewise-linear relaxations of locally important kernel terms with closed-form analytic bounds on the rest inside a reduced-space branch-and-bound framework, yielding valid lower bounds and ε-global convergence for GP posterior mean optimization.
CCBO enables collaborative contextual Bayesian optimization across clients with sublinear regret guarantees and shows substantial gains over non-collaborative methods in simulations and a hot rolling application even under heterogeneity.
Proposes a trajectory-oriented Bayesian optimization method using Gaussian process surrogates on parameters and seeds with adaptive Thompson sampling for efficient discovery of data-consistent trajectories in stochastic epidemic models.
Reformulates constrained black-box optimization as posterior inference in latent space of flow-based models amortized by outsourced diffusion models, claiming superior performance on synthetic and real tasks.
Introduces BOBa, a multi-armed bandit method for scalable surrogate optimization that adaptively allocates inference and evaluations to promising partitions of ultra-large chemical libraries.
PEM via residual bootstrapping improves fixed-budget noisy evolution strategies by reducing update dispersion while preserving conditional mean updates, yielding gains on noisy benchmarks.
MatFormBench introduces a synthetic data generator with five difficulty levels and MatFormScore metric to benchmark 39 inverse design algorithms for target-driven materials formulation.
A framework learns invariant symbolic reward functions from few demonstrations that generalize zero-shot to variations in robotic manipulation tasks.
Recasts sampling-based nonconvex optimization as smoothed gradient descent to obtain non-asymptotic convergence guarantees and introduces the DIDA annealed algorithm that converges to the global optimum.
SSLA approximates the posterior predictive distribution by refitting Bayesian models on self-predicted data, yielding a deterministic uncertainty measure that outperforms standard Laplace approximations in calibration on regression tasks.
citing papers explorer
-
Expected Free Energy-based Planning as Variational Inference
EFE-based planning is formulated as variational free energy minimization with epistemic priors, decomposing into expected plan costs plus a complexity term.
-
What Type of Inference is Active Inference?
EFE-based active inference planning is characterized as VFE on an augmented model plus entropy and planning corrections, with a derived message-passing implementation and grid-world validation.
-
Evidence-Gated LLM Priors for Multi-Objective Bayesian Optimization
Dynamic reputation updates per objective-expert pair plus a three-arm counterfactual gate improve robustness over fixed LLM priors on synthetic tests and molecule benchmarks, but raw LLM confidence is not reliably helpful.
-
Transferring Information Across Interventions in Causal Bayesian Optimization
Graph-coupled causal Bayesian optimization couples intervention effects via shared causal parameters to produce a low-rank causal kernel, logarithmic information-gain bounds, and a regret bound separating optimization, estimation, and intervention-choice errors in linear Gaussian models.
-
AI4S-SDS: A Neuro-Symbolic Solvent Design System via Sparse MCTS and Differentiable Physics Alignment
AI4S-SDS uses sparse MCTS and differentiable physics alignment to generate valid solvent mixtures and identifies a competitive photoresist developer formulation.