Bayesian Optimization
A hyperparameter tuning method that models historical experimental results using a probabilistic surrogate model. It intelligently balances exploration and exploitation to identify near-optimal parameter configurations with a minimal number of experiments.
Timeline (last 90 days)
AI (especially reinforcement learning and Bayesian optimization) has been used to automatically search for optimal pulse shapes and timings to maximize quantum gate fidelity, and can continuously adjust online when hardware characteristics drift over time.
In practical deployment, Bayesian optimization typically cares only about the quality of the single recommended point that the algorithm finally returns, rather than the intermediate performance of the entire optimization process.
On synthetic benchmarks, a hyperparameter optimization task and a control task, after introducing the exploration strategy, the final recommendation quality of existing Bayesian optimization algorithms was improved in all cases.
This exploration query replacement scheme does not change the main structure of the original algorithm and can be seamlessly integrated into existing Bayesian optimization frameworks.
The exploration strategy proposed in this study is: in any Bayesian optimization procedure, occasionally replace the acquisition query with an exploration query.
This study proposes that the existing cumulative regret guarantees of Bayesian optimization do not necessarily characterize the quality of the algorithm's final output.
Expected improvement (EI) is a classic acquisition function in Bayesian optimization that computes the expected amount of improvement possibly obtained relative to the incumbent after sampling at a candidate point, and it usually has a closed-form expression.
LlMBO (Large language Model Bayesian Optimization) introduces Bayesian optimization into LLM fine-tuning.
The computational complexity of Bayesian optimization typically grows exponentially with parameter dimensionality, and it is traditionally only applicable to low-dimensional problems (generally within a few dozen dimensions).
AI agents do not try to exhaustively enumerate all combinations, but instead use strategies such as heuristic search, machine learning predictive models or Bayesian optimization to quickly locate high-potential regions in a vast space, compressing an unmanageable search space to a scale that can be experimentally verified.
4 more timeline events
All Facts (15)
AI (especially reinforcement learning and Bayesian optimization) has been used to automatically search for optimal pulse shapes and timings to maximize quantum gate fidelity, and can continuously adjust online when hardware characteristics drift over time.
50%UnverifiedOn synthetic benchmarks, a hyperparameter optimization task and a control task, after introducing the exploration strategy, the final recommendation quality of existing Bayesian optimization algorithms was improved in all cases.
50%UnverifiedThis exploration query replacement scheme does not change the main structure of the original algorithm and can be seamlessly integrated into existing Bayesian optimization frameworks.
50%UnverifiedIn practical deployment, Bayesian optimization typically cares only about the quality of the single recommended point that the algorithm finally returns, rather than the intermediate performance of the entire optimization process.
50%UnverifiedThe exploration strategy proposed in this study is: in any Bayesian optimization procedure, occasionally replace the acquisition query with an exploration query.
50%UnverifiedThis study proposes that the existing cumulative regret guarantees of Bayesian optimization do not necessarily characterize the quality of the algorithm's final output.
50%UnverifiedExpected improvement (EI) is a classic acquisition function in Bayesian optimization that computes the expected amount of improvement possibly obtained relative to the incumbent after sampling at a candidate point, and it usually has a closed-form expression.
50%UnverifiedThe computational complexity of Bayesian optimization typically grows exponentially with parameter dimensionality, and it is traditionally only applicable to low-dimensional problems (generally within a few dozen dimensions).
50%UnverifiedAI agents do not try to exhaustively enumerate all combinations, but instead use strategies such as heuristic search, machine learning predictive models or Bayesian optimization to quickly locate high-potential regions in a vast space, compressing an unmanageable search space to a scale that can be experimentally verified.
50%UnverifiedStandard Gaussian process Bayesian optimization is not suitable for spatio-temporal point process problems because its assumption of approximately stationary, independent observation noise is inconsistent with the data-generating mechanism of a doubly stochastic Cox process.
50%UnverifiedBayesian optimization is a sequential experimental design method for expensive black-box functions that typically uses Gaussian processes as a probabilistic surrogate model, and has long been regarded as a strong baseline in the field of experimental design
50%UnverifiedIn reaction optimization tasks, RSI-Anything achieved an average best-observed yield of 96.15% under a budget of 40 experiments, outperforming Bayesian optimization's 82.16%
50%UnverifiedProcess optimization uses Design of Experiments (DOE) or Bayesian optimization methods to search across multiple process parameters for an operating window that maximizes yield.
50%UnverifiedAlphaEvolve更适合一次优化长期使用的场景，因为进化算法需反复评估大量候选方案导致算力成本可观
50%UnverifiedLlMBO (Large language Model Bayesian Optimization) introduces Bayesian optimization into LLM fine-tuning.
50%