Troubleshooting deep-learner training data problems using an evolutionary algorithm on Summit

Coletti, Mark; Fafard, Alexander; Page, David L.

doi:10.1147/jrd.2019.2960225

Cited by 5 publications

(3 citation statements)

References 21 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The Dask library was designed to enable parallel computing in Python by distributing scientific calculations such as linear algebra operations and singular-value decomposition [23,24]; it can also be used purely as a distributed workflow manager via Python subprocess calls. On Summit, Dask for the former use case was tested on small numbers of nodes [25], and deployed on larger numbers of nodes for parallel, distributed dataframe and database query processing tasks [19]; it was deployed as a workflow manager for large-scale evolutionary algorithms at the scale of 500 Summit nodes [26,27].…”

Section: Leadership-scale Hpc Workflowsmentioning

confidence: 99%

Proteome-scale Deployment of Protein Structure Prediction Workflows on the Summit Supercomputer

Gao¹,

Coletti²,

Davidson³

et al. 2022

Preprint

View full text Add to dashboard Cite

Deep learning has contributed to major advances in the prediction of protein structure from sequence, a fundamental problem in structural bioinformatics. With predictions now approaching the accuracy of crystallographic resolution in some cases, and with accelerators like GPUs and TPUs making inference using large models rapid, fast genome-level structure prediction becomes an obvious aim. Leadership-class computing resources can be used to perform genome-scale protein structure prediction using state-of-the-art deep learning models, providing a wealth of new data for systems biology applications. Here we describe our efforts to efficiently deploy the AlphaFold2 program, for full-proteome structure prediction, at scale on the Oak Ridge Leadership Computing Facility's resources, including the Summit supercomputer. We performed inference to produce the predicted structures for 35,634 protein sequences, corresponding to three prokaryotic proteomes and one plant proteome, using under 4,000 total Summit node hours, equivalent to using the majority of the supercomputer for one hour. We also designed an optimized structure refinement that reduced the time for the relaxation stage of the AlphaFold pipeline by over 10X for longer sequences. We demonstrate the types of analyses that can be performed on proteome-scale collections of sequences, including a search for novel quaternary structures and implications for functional annotation.

show abstract

Section: Leadership-scale Hpc Workflowsmentioning

confidence: 99%

Proteome-scale Deployment of Protein Structure Prediction Workflows on the Summit Supercomputer

Gao¹,

Coletti²,

Davidson³

et al. 2022

Preprint

View full text Add to dashboard Cite

show abstract

“…In particular, asynchronous steady-state evolutionary algorithms (ASSEAs) can often reach nearperfect resource utilization throughout an experiment (see Figure 1), particularly when the algorithm's computational cost is dominated by fitness evaluation on the workers as is often the case when, say, tuning the parameters or behaviours of an expensive simulation (D'Auria et al, 2020;Gunaratne & Garibay, 2020). Asynchronous EAs are growing in popularity, and have most recently been applied to a variety computationally challenging problems such as deep neural network hyperparameter tuning (Jaderberg et al, 2017;Coletti et al, 2019), evolutionary reinforcement learning (Lee et al, 2020), and simulation problems in air traffic management (Pellegrini et al, 2020).…”

Section: Introductionmentioning

confidence: 99%

Avoiding excess computation in asynchronous evolutionary algorithms

et al. 2022

View full text Add to dashboard Cite

Asynchronous evolutionary algorithms are becoming increasingly popular as a means of making full use of many processors while solving computationally expensive search and optimization problems. These algorithms excel at keeping large clusters fully utilized, but may sometimes inefficiently sample an excess of fast-evaluating solutions at the expense of higher-quality, slow-evaluating ones. We have previously introduced a steady-state parent selection strategy, SWEET ("Selection whilE Evalu-aTing"), that sometimes selects individuals that are still being evaluated and allows them to reproduce early. We perform a takeover-time analysis that confirms that this strategy gives slow-evaluating individuals that have higher fitnesses an increased ability to multiply in the population. We also find that SWEET appears effective at improving optimization performance on problems in which solution quality is positively correlated with evaluation time. We evaluate our approach on six simulated real-valued optimization problems and three real-world applications: an autonomous vehicle controller problem that involves tuning a spiking neural network and two adversarial EA problems. We further evaluate SWEET versus a basic asynchronous process in a simulated setting. We present evidence that SWEET outperforms basic asynchronous processes in a use-case in which performance is positively correlated with evaluation time, and performs comparably (and often better) than basic asynchronous processes in several use-cases where performance is negatively correlated with evaluation time. That said, in the cases where performance and evaluation time are negatively correlated the variance of outcomes for SWEET is notably high.

show abstract

“…1), particularly when the algorithm's computational cost is dominated by fitness evaluation on the workers (as is often the case when, say, tuning the parameters or behaviors of an expensive simulation [7,11]). Asynchronous EAs are growing in popularity, and have most recently been applied to a variety computationally challenging problems such as deep neural network hyperparameter tuning [5,13], evolutionary reinforcement learning [15], and simulation problems in air traffic management [19].…”

Section: Introductionmentioning

confidence: 99%