In the European Center of Excellence in Exascale computing "Research on AI- and Simulation-Based Engineering at Exascale" (CoE RAISE), researchers develop novel, scalable AI technologies towards Exascale. This work exercises High Performance Computing resources to perform large-scale hyperparameter optimization using distributed training on multiple compute nodes. This is part of RAISE's work on data-driven use cases which leverages AI- and HPC cross-methods developed within the project. In response to the demand for parallelizable and resource efficient hyperparameter optimization methods, advanced hyperparameter search algorithms are benchmarked and compared. The evaluated algorithms, including Random Search, Hyperband and ASHA, are tested and compared in terms of both accuracy and accuracy per compute resources spent. As an example use case, a graph neural network model known as MLPF, developed for the task of Machine-Learned Particle-Flow reconstruction in High Energy Physics, acts as the base model for optimization. Results show that hyperparameter optimization significantly increased the performance of MLPF and that this would not have been possible without access to large-scale High Performance Computing resources. It is also shown that, in the case of MLPF, the ASHA algorithm in combination with Bayesian optimization gives the largest performance increase per compute resources spent out of the investigated algorithms.
Visual speech recognition (VSR) aims to recognize the content of speech based on lip movements, without relying on the audio stream. Advances in deep learning and the availability of large audio-visual datasets have led to the development of much more accurate and robust VSR models than ever before. However, these advances are usually due to the larger training sets rather than the model design. Here we demonstrate that designing better models is equally as important as using larger training sets. We propose the addition of prediction-based auxiliary tasks to a VSR model, and highlight the importance of hyperparameter optimization and appropriate data augmentations. We show that such a model works for different languages and outperforms all previous methods trained on publicly available datasets by a large margin. It even outperforms models that were trained on non-publicly available datasets containing up to to 21 times more data. We show, furthermore, that using additional training data, even in other languages or with automatically generated transcriptions, results in further improvement.
We present two novel hyperparameter optimization strategies for optimization of deep learning models with a modular architecture constructed of multiple subnetworks. As complex networks with multiple subnetworks become more frequently applied in machine learning, hyperparameter optimization methods are required to efficiently optimize their hyperparameters. Existing hyperparameter searches are general, and can be used to optimize such networks, however, by exploiting the multi-subnetwork architecture, these searches can be sped up substantially. The proposed methods offer faster convergence to a better-performing final model. To demonstrate this, we propose 2 independent approaches to enhance these prior algorithms: 1) a divide-and-conquer approach, in which the best subnetworks of top-performing models are combined, allowing for more rapid sampling of the hyperparameter search space. 2) A subnetwork adaptive approach that distributes computational resources based on the importance of each subnetwork, allowing more intelligent resource allocation. These approaches can be flexibily applied to many hyperparameter optimization algorithms. To illustrate this, we combine our approaches with the commonly-used Bayesian optimization method. Our approaches are then tested against both synthetic examples and real-world examples and applied to multiple network types including convolutional neural networks and dense feed forward neural networks. Our approaches show an increased optimization efficiency of up to 23.62x, and a final performance boost of up to 3.5% accuracy for classification and 4.4 MSE for regression, when compared to comparable BO approach.
We investigate the effectiveness of ensembles of pretrained transformer-based language models on short answer questions using the Kaggle Automated Short Answer Scoring dataset. We fine-tune a collection of popular small, base, and large pretrained transformer-based language models, and train one feature-base model on the dataset with the aim of testing ensembles of these models. We used an early stopping mechanism and hyperparameter optimization in training. We observe that generally that the larger models perform slightly better, however, they still fall short of state-of-the-art results one their own. Once we consider ensembles of models, there are ensembles of a number of large networks that do produce state-of-the-art results, however, these ensembles are too large to realistically be put in a production environment.
Multi-fidelity (gray-box) hyperparameter optimization techniques (HPO) have recently emerged as a promising direction for tuning Deep Learning methods. However, existing methods suffer from a sub-optimal allocation of the HPO budget to the hyperparameter configurations. In this work, we introduce DyHPO, a Bayesian Optimization method that learns to decide which hyperparameter configuration to train further in a dynamic race among all feasible configurations. We propose a new deep kernel for Gaussian Processes that embeds the learning curve dynamics, and an acquisition function that incorporates multi-budget information. We demonstrate the significant superiority of DyHPO against state-of-the-art hyperparameter optimization methods through large-scale experiments comprising 50 datasets (Tabular, Image, NLP) and diverse architectures (MLP, CNN/NAS, RNN).
Topic models are statistical methods that extract underlying topics from document collections. When performing topic modeling, a user usually desires topics that are coherent, diverse between each other, and that constitute good document representations for downstream tasks (e.g. document classification). In this paper, we conduct a multi-objective hyperparameter optimization of three well-known topic models. The obtained results reveal the conflicting nature of different objectives and that the training corpus characteristics are crucial for the hyperparameter selection, suggesting that it is possible to transfer the optimal hyperparameter configurations between datasets.
Reservoir computers (RC) have proven useful as surrogate models in forecasting and replicating systems of chaotic dynamics. The quality of surrogate models based on RCs is crucially dependent on their optimal implementation that involves selecting optimal reservoir topology and hyperparameters. By systematically applying Bayesian hyperparameter optimization and using ensembles of reservoirs of various topology we show that connectednes of reservoirs is of significance only in forecasting and replication of chaotic system of sufficient complexity. By applying RCs of different topology in forecasting and replicating the Lorenz system, a coupled Wilson-Cowan system, and the Kuramoto-Sivashinsky system, we show that simple reservoirs of unconnected nodes (RUN) outperform reservoirs of connected nodes for target systems whose estimated fractal dimension dimension is $d \lesssim 5.5$ and that linked reservoirs are better for systems with $d > 5.5$. This finding is highly important for evaluation of reservoir computing methods and on selecting a method for prediction of signals measured on nonlinear systems.
We analyse a general class of bilevel problems, in which the upper-level problem consists in the minimization of a smooth objective function and the lower-level problem is to find the fixed point of a smooth contraction map. This type of problems include instances of meta-learning, equilibrium models, hyperparameter optimization and data poisoning adversarial attacks. Several recent works have proposed algorithms which warm-start the lower-level problem, i.e.~they use the previous lower-level approximate solution as a staring point for the lower-level solver. This warm-start procedure allows one to improve the sample complexity in both the stochastic and deterministic settings, achieving in some cases the order-wise optimal sample complexity. However, there are situations, e.g., meta learning and equilibrium models, in which the warm-start procedure is not well-suited or ineffective. In this work we show that without warm-start, it is still possible to achieve order-wise (near) optimal sample complexity. In particular, we propose a simple method which uses (stochastic) fixed point iterations at the lower-level and projected inexact gradient descent at the upper-level, that reaches an $\epsilon$-stationary point using $O(\epsilon^{-2})$ and $\tilde{O}(\epsilon^{-1})$ samples for the stochastic and the deterministic setting, respectively. Finally, compared to methods using warm-start, our approach yields a simpler analysis that does not need to study the coupled interactions between the upper-level and lower-level iterates.
Time series forecasting is fundamental for various use cases in different domains such as energy systems and economics. Creating a forecasting model for a specific use case requires an iterative and complex design process. The typical design process includes the five sections (1) data pre-processing, (2) feature engineering, (3) hyperparameter optimization, (4) forecasting method selection, and (5) forecast ensembling, which are commonly organized in a pipeline structure. One promising approach to handle the ever-growing demand for time series forecasts is automating this design process. The present paper, thus, analyzes the existing literature on automated time series forecasting pipelines to investigate how to automate the design process of forecasting models. Thereby, we consider both Automated Machine Learning (AutoML) and automated statistical forecasting methods in a single forecasting pipeline. For this purpose, we firstly present and compare the proposed automation methods for each pipeline section. Secondly, we analyze the automation methods regarding their interaction, combination, and coverage of the five pipeline sections. For both, we discuss the literature, identify problems, give recommendations, and suggest future research. This review reveals that the majority of papers only cover two or three of the five pipeline sections. We conclude that future research has to holistically consider the automation of the forecasting pipeline to enable the large-scale application of time series forecasting.
Background: Embedded feature selection in high-dimensional data with very small sample sizes requires optimized hyperparameters for the model building process. For this hyperparameter optimization, nested cross-validation must be applied to avoid a biased performance estimation. The resulting repeated training with high-dimensional data leads to very long computation times. Moreover, it is likely to observe a high variance in the individual performance evaluation metrics caused by outliers in tiny validation sets. Therefore, early stopping applying standard pruning algorithms to save time risks discarding promising hyperparameter sets. Result: To speed up feature selection for high-dimensional data with tiny sample size, we adapt the use of a state-of-the-art asynchronous successive halving pruner. In addition, we combine it with two complementary pruning strategies based on domain or prior knowledge. One pruning strategy immediately stops computing trials with semantically meaningless results for the selected hyperparameter combinations. The other is a new extrapolating threshold pruning strategy suitable for nested-cross-validation with a high variance of performance evaluation metrics. In repeated experiments, our combined pruning strategy keeps all promising trials. At the same time, the calculation time is substantially reduced compared to using a state-of-the-art asynchronous successive halving pruner alone. Up to 81.3\% fewer models were trained achieving the same optimization result. Conclusion: The proposed combined pruning strategy accelerates data analysis or enables deeper searches for hyperparameters within the same computation time. This leads to significant savings in time, money and energy consumption, opening the door to advanced, time-consuming analyses.
Computing a Gaussian process (GP) posterior has a computational cost cubical in the number of historical points. A reformulation of the same GP posterior highlights that this complexity mainly depends on how many \emph{unique} historical points are considered. This can have important implication in active learning settings, where the set of historical points is constructed sequentially by the learner. We show that sequential black-box optimization based on GPs (GP-Opt) can be made efficient by sticking to a candidate solution for multiple evaluation steps and switch only when necessary. Limiting the number of switches also limits the number of unique points in the history of the GP. Thus, the efficient GP reformulation can be used to exactly and cheaply compute the posteriors required to run the GP-Opt algorithms. This approach is especially useful in real-world applications of GP-Opt with high switch costs (e.g. switching chemicals in wet labs, data/model loading in hyperparameter optimization). As examples of this meta-approach, we modify two well-established GP-Opt algorithms, GP-UCB and GP-EI, to switch candidates as infrequently as possible adapting rules from batched GP-Opt. These versions preserve all the theoretical no-regret guarantees while improving practical aspects of the algorithms such as runtime, memory complexity, and the ability of batching candidates and evaluating them in parallel.
The choices of hyperparameters have critical effects on the performance of machine learning models. In this paper, we present a general framework that is able to construct an adaptive optimizer, which automatically adjust the appropriate algorithm and parameters in the process of optimization. Examining the method of adaptive optimizer, we product an example of using genetic algorithm to construct an adaptive optimizer based on Bayesian Optimizer and compared effectiveness with original optimizer. Especially, It has great advantages in parallel optimization.
Despite the introduction of vaccines, Coronavirus disease (COVID-19) remains a worldwide dilemma, continuously developing new variants such as Delta and the recent Omicron. The current standard for testing is through polymerase chain reaction (PCR). However, PCRs can be expensive, slow, and/or inaccessible to many people. X-rays on the other hand have been readily used since the early 20th century and are relatively cheaper, quicker to obtain, and typically covered by health insurance. With a careful selection of model, hyperparameters, and augmentations, we show that it is possible to develop models with 83% accuracy in binary classification and 64% in multi-class for detecting COVID-19 infections from chest x-rays.
This paper demonstrates how Automated Machine Learning (AutoML) methods can be used as effective surrogate models in engineering design problems. To do so, we consider the challenging problem of structurally-performant bicycle frame design and demonstrate across-the-board dominance by AutoML in regression and classification surrogate modeling tasks. We also introduce FRAMED -- a parametric dataset of 4500 bicycle frames based on bicycles designed by practitioners and enthusiasts worldwide. Accompanying these frame designs, we provide ten structural performance values such as weight, displacements under load, and safety factors computed using finite element simulations for all the bicycle frame designs. We formulate two challenging test problems: a performance-prediction regression problem and a feasibility-prediction classification problem. We then systematically search for optimal surrogate models using Bayesian hyperparameter tuning and neural architecture search. Finally, we show how a state-of-the-art AutoML method can be effective for both regression and classification problems. We demonstrate that the proposed AutoML models outperform the strongest gradient boosting and neural network surrogates identified through Bayesian optimization by an improved F1 score of 24\% for classification and reduced mean absolute error by 12.5\% for regression. Our work introduces a dataset for bicycle design practitioners, provides two benchmark problems for surrogate modeling researchers, and demonstrates the advantages of AutoML in machine learning tasks. The dataset and code are provided at \url{http://decode.mit.edu/projects/framed/}.
This manuscript introduces an autotuned algorithm for searching nearest neighbors based on neighbor graphs and optimization metaheuristics to produce Pareto-optimal searches for quality and search speed automatically; the same strategy is also used to produce indexes that achieve a minimum quality. Our approach is described and benchmarked with other state-of-the-art similarity search methods, showing convenience and competitiveness.
Machine learning (ML) methods are used in most technical areas such as image recognition, product recommendation, financial analysis, medical diagnosis, and predictive maintenance. An important aspect of implementing ML methods involves controlling the learning process for the ML method so as to maximize the performance of the method under consideration. Hyperparameter tuning is the process of selecting a suitable set of ML method parameters that control its learning process. In this work, we demonstrate the use of discrete simulation optimization methods such as ranking and selection (R&S) and random search for identifying a hyperparameter set that maximizes the performance of a ML method. Specifically, we use the KN R&S method and the stochastic ruler random search method and one of its variations for this purpose. We also construct the theoretical basis for applying the KN method, which determines the optimal solution with a statistical guarantee via solution space enumeration. In comparison, the stochastic ruler method asymptotically converges to global optima and incurs smaller computational overheads. We demonstrate the application of these methods to a wide variety of machine learning models, including deep neural network models used for time series prediction and image classification. We benchmark our application of these methods with state-of-the-art hyperparameter optimization libraries such as $hyperopt$ and $mango$. The KN method consistently outperforms $hyperopt$'s random search (RS) and Tree of Parzen Estimators (TPE) methods. The stochastic ruler method outperforms the $hyperopt$ RS method and offers statistically comparable performance with respect to $hyperopt$'s TPE method and the $mango$ algorithm.
In recent years, people from all over the world are suffering from one of the most severe diseases in history, known as Coronavirus disease 2019, COVID-19 for short. When the virus reaches the lungs, it has a higher probability to cause lung pneumonia and sepsis. X-ray image is a powerful tool in identifying the typical features of the infection for COVID-19 patients. The radiologists and pathologists observe that ground-glass opacity appears in the chest X-ray for infected patient \cite{cozzi2021ground}, and it could be used as one of the criteria during the diagnosis process. In the past few years, deep learning has proven to be one of the most powerful methods in the field of image classification. Due to significant differences in Chest X-Ray between normal and infected people \cite{rousan2020chest}, deep models could be used to identify the presence of the disease given a patient's Chest X-Ray. Many deep models are complex, and it evolves with lots of input parameters. Designers sometimes struggle with the tuning process for deep models, especially when they build up the model from scratch. Genetic Algorithm, inspired by the biological evolution process, plays a key role in solving such complex problems. In this paper, I proposed a genetic-based approach to optimize the Convolutional Neural Network(CNN) for the Chest X-Ray classification task.
Accurate uncertainty estimates are important in sequential model-based decision-making tasks such as Bayesian optimization. However, these estimates can be imperfect if the data violates assumptions made by the model (e.g., Gaussianity). This paper studies which uncertainties are needed in model-based decision-making and in Bayesian optimization, and argues that uncertainties can benefit from calibration -- i.e., an 80% predictive interval should contain the true outcome 80% of the time. Maintaining calibration, however, can be challenging when the data is non-stationary and depends on our actions. We propose using simple algorithms based on online learning to provably maintain calibration on non-i.i.d. data, and we show how to integrate these algorithms in Bayesian optimization with minimal overhead. Empirically, we find that calibrated Bayesian optimization converges to better optima in fewer steps, and we demonstrate improved performance on standard benchmark functions and hyperparameter optimization tasks.
Image analysis tasks in computational pathology are commonly solved using convolutional neural networks (CNNs). The selection of a suitable CNN architecture and hyperparameters is usually done through exploratory iterative optimization, which is computationally expensive and requires substantial manual work. The goal of this article is to evaluate how generic tools for neural network architecture search and hyperparameter optimization perform for common use cases in computational pathology. For this purpose, we evaluated one on-premises and one cloud-based tool for three different classification tasks for histological images: tissue classification, mutation prediction, and grading. We found that the default CNN architectures and parameterizations of the evaluated AutoML tools already yielded classification performance on par with the original publications. Hyperparameter optimization for these tasks did not substantially improve performance, despite the additional computational effort. However, performance varied substantially between classifiers obtained from individual AutoML runs due to non-deterministic effects. Generic CNN architectures and AutoML tools could thus be a viable alternative to manually optimizing CNN architectures and parametrizations. This would allow developers of software solutions for computational pathology to focus efforts on harder-to-automate tasks such as data curation.
We introduce a machine-learning (ML) framework for high-throughput benchmarking of diverse representations of chemical systems against datasets of materials and molecules. The guiding principle underlying the benchmarking approach is to evaluate raw descriptor performance by limiting model complexity to simple regression schemes while enforcing best ML practices, allowing for unbiased hyperparameter optimization, and assessing learning progress through learning curves along series of synchronized train-test splits. The resulting models are intended as baselines that can inform future method development, next to indicating how easily a given dataset can be learnt. Through a comparative analysis of the training outcome across a diverse set of physicochemical, topological and geometric representations, we glean insight into the relative merits of these representations as well as their interrelatedness.