PPubMed28 Aug 2018Cited 163×

Chimera: enabling hierarchy based multi-objective optimization for self-driving laboratories † †Electronic supplementary information (ESI) available. See DOI: 10.1039/c8sc02239a

Florian Häse, Loïc M. Roch, Alán Aspuru-Guzik

Abstract

Chimera enables multi-target optimization for experimentation or expensive computations, where evaluations are the limiting factor.

A figure from Chimera: enabling hierarchy based multi-objective optimization for self-driving laboratories † †Electronic supplementary information (ESI) available. See DOI: 10.1039/c8sc02239a — fig. from the paper

§ The Valyu brief

The paper, in five minutes

Read from the full text via Valyu Contents

Chimera provides a hierarchy‑based scalarizing function that lets single‑objective optimizers efficiently find Pareto‑optimal solutions when objective evaluations are expensive.

What’s new

i.Introduced Chimera, a new achievement scalarizing function that combines multiple objectives using a user‑defined hierarchy and tolerances.
ii.Demonstrated Chimera’s compatibility with various single‑objective optimizers (grid search, CMA‑ES, Spearmint, Phoenics) across six analytic benchmark sets.
iii.Applied Chimera to two real‑world self‑driving laboratory problems: auto‑calibration of a virtual robotic HPLC platform and inverse‑design of a four‑pigment excitonic system.
iv.Showed that Chimera requires minimal prior knowledge of objective shapes and is largely insensitive to its smoothness hyperparameter τ.

How they did it

Chimera is formulated as an achievement scalarizing function that merges n objectives into a single smooth objective using a hierarchy, tolerance levels, and a smoothness hyperparameter τ. Benchmarking paired Chimera (and a comparative c‑ASF) with four gradient‑free optimizers on six analytic multi‑objective test functions, measuring performance after 100 function evaluations over 25 independent runs. Additional case studies used a Bayesian neural‑network model of an HPLC robot (trained on 1500 experiments) and a parametrized excitonic system with ten variables to test Chimera in virtual experiments.

What they found

Chimera achieved the smallest relative deviation to Pareto‑optimal points after only 100 objective evaluations across benchmark sets (Fonseca and Viennet) (Fig. 3).
Benchmarking employed 25 independent optimization runs per algorithm.
The virtual robot model was trained on 1500 autonomous experiments.
The excitonic inverse‑design problem involved 10 independent parameters.
The recommended smoothness hyperparameter is τ = 10⁻³, with performance insensitive across several orders of magnitude.

Caveats

✕Performance may depend on the choice of hierarchy and tolerance values, requiring user expertise to set them appropriately.
✕Pareto‑optimal solutions are not guaranteed for highly non‑convex objective spaces.
✕The virtual‑robot benchmark was performed on a simulated model rather than on physical hardware, leaving real‑world robustness untested.
✕Chimera’s scalability may be limited by the cubic scaling of underlying Bayesian optimization methods with the number of observations.

§ The Valyu brief

Reading the full paper and taking notes. This takes a few seconds…

§ Ask this paper

Ask a question about this paper

Valyu reads the full text and answers from what the paper actually says.

§ Related work in other archives

Chimera: enabling hierarchy based multi-objective optimization for self-driving laboratories † †Electronic supplementary information (ESI) available. See DOI: 10.1039/c8sc02239a

The paper, in five minutes

Ask a question about this paper

Similar papers in all four archives

Similar papers in all four archives

The paper, in five minutes