Hyperparameter optimization (HPO) is concerned with the automated search for the most appropriate hyperparameter configuration (HPC) of a parameterized machine learning algorithm. A state-of-the-art HPO method is Hyperband, which, however, has its own parameters that influence its performance. One of these parameters, the maximal budget, is especially problematic: If chosen too small, the budget needs to be increased in hindsight and, as Hyperband is not incremental by design, the entire algorithm must be re-run. This is not only costly but also comes with a loss of valuable knowledge already accumulated. In this paper, we propose incremental variants of Hyperband that eliminate these drawbacks, and show that these variants satisfy theoretical guarantees qualitatively similar to those for the original Hyperband with the "right" budget. Moreover, we demonstrate their practical utility in experiments with benchmark data sets.
When data is scarce, software analytics can make many mistakes. For example, consider learning predictors for open source project health (e.g. the number of closed pull requests in twelve months time). The training data for this task may be very small (e.g. five years of data, collected every month means just 60 rows of training data). The models generated from such tiny data sets can make many prediction errors. Those errors can be tamed by a {\em landscape analysis} that selects better learner control parameters. Our niSNEAK tool (a)~clusters the data to find the general landscape of the hyperparameters; then (b)~explores a few representatives from each part of that landscape. niSNEAK is both faster and more effective than prior state-of-the-art hyperparameter optimization algorithms (e.g. FLASH, HYPEROPT, OPTUNA). The configurations found by niSNEAK have far less error than other methods. For example, for project health indicators such as $C$= number of commits; $I$=number of closed issues, and $R$=number of closed pull requests, niSNEAK's 12 month prediction errors are \{I=0\%, R=33\%\,C=47\%\} Based on the above, we recommend landscape analytics (e.g. niSNEAK) especially when learning from very small data sets. This paper only explores the application of niSNEAK to project health. That said, we see nothing in principle that prevents the application of this technique to a wider range of problems. To assist other researchers in repeating, improving, or even refuting our results, all our scripts and data are available on GitHub at https://github.com/zxcv123456qwe/niSneak
Class-incremental learning (CIL) aims to train a classification model while the number of classes increases phase-by-phase. An inherent challenge of CIL is the stability-plasticity tradeoff, i.e., CIL models should keep stable to retain old knowledge and keep plastic to absorb new knowledge. However, none of the existing CIL models can achieve the optimal tradeoff in different data-receiving settings--where typically the training-from-half (TFH) setting needs more stability, but the training-from-scratch (TFS) needs more plasticity. To this end, we design an online learning method that can adaptively optimize the tradeoff without knowing the setting as a priori. Specifically, we first introduce the key hyperparameters that influence the trade-off, e.g., knowledge distillation (KD) loss weights, learning rates, and classifier types. Then, we formulate the hyperparameter optimization process as an online Markov Decision Process (MDP) problem and propose a specific algorithm to solve it. We apply local estimated rewards and a classic bandit algorithm Exp3 to address the issues when applying online MDP methods to the CIL protocol. Our method consistently improves top-performing CIL methods in both TFH and TFS settings, e.g., boosting the average accuracy of TFH and TFS by 2.2 percentage points on ImageNet-Full, compared to the state-of-the-art.
In this letter, we propose a model parameter identification method via a hyperparameter optimization scheme (MI-HPO). Our method adopts an efficient explore-exploit strategy to identify the parameters of dynamic models in a data-driven optimization manner. We utilize our method for model parameter identification of the AV-21, a full-scaled autonomous race vehicle. We then incorporate the optimized parameters for the design of model-based planning and control systems of our platform. In experiments, MI-HPO exhibits more than 13 times faster convergence than traditional parameter identification methods. Furthermore, the parametric models learned via MI-HPO demonstrate good fitness to the given datasets and show generalization ability in unseen dynamic scenarios. We further conduct extensive field tests to validate our model-based system, demonstrating stable obstacle avoidance and high-speed driving up to 217 km/h at the Indianapolis Motor Speedway and Las Vegas Motor Speedway. The source code for our work and videos of the tests are available at https://github.com/hynkis/MI-HPO.
Nearly all jurisdictions in the United States require a professional license exam, commonly referred to as "the Bar Exam," as a precondition for law practice. To even sit for the exam, most jurisdictions require that an applicant completes at least seven years of post-secondary education, including three years at an accredited law school. In addition, most test-takers also undergo weeks to months of further, exam-specific preparation. Despite this significant investment of time and capital, approximately one in five test-takers still score under the rate required to pass the exam on their first try. In the face of a complex task that requires such depth of knowledge, what, then, should we expect of the state of the art in "AI?" In this research, we document our experimental evaluation of the performance of OpenAI's `text-davinci-003` model, often-referred to as GPT-3.5, on the multistate multiple choice (MBE) section of the exam. While we find no benefit in fine-tuning over GPT-3.5's zero-shot performance at the scale of our training data, we do find that hyperparameter optimization and prompt engineering positively impacted GPT-3.5's zero-shot performance. For best prompt and parameters, GPT-3.5 achieves a headline correct rate of 50.3% on a complete NCBE MBE practice exam, significantly in excess of the 25% baseline guessing rate, and performs at a passing rate for both Evidence and Torts. GPT-3.5's ranking of responses is also highly-correlated with correctness; its top two and top three choices are correct 71% and 88% of the time, respectively, indicating very strong non-entailment performance. While our ability to interpret these results is limited by nascent scientific understanding of LLMs and the proprietary nature of GPT, we believe that these results strongly suggest that an LLM will pass the MBE component of the Bar Exam in the near future.
We introduce an end-to-end computational framework that allows for hyperparameter optimization using the DeepHyper library, accelerated model training, and interpretable AI inference. The framework is based on state-of-the-art AI models including CGCNN, PhysNet, SchNet, MPNN, MPNN-transformer, and TorchMD-NET. We employ these AI models along with the benchmark QM9, hMOF, and MD17 datasets to showcase how the models can predict user-specified material properties within modern computing environments. We demonstrate transferable applications in the modeling of small molecules, inorganic crystals and nanoporous metal organic frameworks with a unified, standalone framework. We have deployed and tested this framework in the ThetaGPU supercomputer at the Argonne Leadership Computing Facility, and in the Delta supercomputer at the National Center for Supercomputing Applications to provide researchers with modern tools to conduct accelerated AI-driven discovery in leadership-class computing environments. We release these digital assets as open source scientific software in GitLab, and ready-to-use Jupyter notebooks in Google Colab.
Although Reinforcement Learning (RL) has shown impressive results in games and simulation, real-world application of RL suffers from its instability under changing environment conditions and hyperparameters. We give a first impression of the extent of this instability by showing that the hyperparameters found by automatic hyperparameter optimization (HPO) methods are not only dependent on the problem at hand, but even on how well the state describes the environment dynamics. Specifically, we show that agents in contextual RL require different hyperparameters if they are shown how environmental factors change. In addition, finding adequate hyperparameter configurations is not equally easy for both settings, further highlighting the need for research into how hyperparameters influence learning and generalization in RL.
Bilevel optimization plays an essential role in many machine learning tasks, ranging from hyperparameter optimization to meta-learning. Existing studies on bilevel optimization, however, focus on either centralized or synchronous distributed setting. The centralized bilevel optimization approaches require collecting massive amount of data to a single server, which inevitably incur significant communication expenses and may give rise to data privacy risks. Synchronous distributed bilevel optimization algorithms, on the other hand, often face the straggler problem and will immediately stop working if a few workers fail to respond. As a remedy, we propose Asynchronous Distributed Bilevel Optimization (ADBO) algorithm. The proposed ADBO can tackle bilevel optimization problems with both nonconvex upper-level and lower-level objective functions, and its convergence is theoretically guaranteed. Furthermore, it is revealed through theoretic analysis that the iteration complexity of ADBO to obtain the $\epsilon$-stationary point is upper bounded by $\mathcal{O}(\frac{1}{{{\epsilon ^2}}})$. Thorough empirical studies on public datasets have been conducted to elucidate the effectiveness and efficiency of the proposed ADBO.
Hyperparameter optimization (HPO) is a vital step in improving performance in deep learning (DL). Practitioners are often faced with the trade-off between multiple criteria, such as accuracy and latency. Given the high computational needs of DL and the growing demand for efficient HPO, the acceleration of multi-objective (MO) optimization becomes ever more important. Despite the significant body of work on meta-learning for HPO, existing methods are inapplicable to MO tree-structured Parzen estimator (MO-TPE), a simple yet powerful MO-HPO algorithm. In this paper, we extend TPE's acquisition function to the meta-learning setting using a task similarity defined by the overlap of top domains between tasks. We also theoretically analyze and address the limitations of our task similarity. In the experiments, we demonstrate that our method speeds up MO-TPE on tabular HPO benchmarks and attains state-of-the-art performance. Our method was also validated externally by winning the AutoML 2022 competition on "Multiobjective Hyperparameter Optimization for Transformers".
Modern machine learning models are often constructed taking into account multiple objectives, e.g., minimizing inference time while also maximizing accuracy. Multi-objective hyperparameter optimization (MHPO) algorithms return such candidate models, and the approximation of the Pareto front is used to assess their performance. In practice, we also want to measure generalization when moving from the validation to the test set. However, some of the models might no longer be Pareto-optimal which makes it unclear how to quantify the performance of the MHPO method when evaluated on the test set. To resolve this, we provide a novel evaluation protocol that allows measuring the generalization performance of MHPO methods and studying its capabilities for comparing two optimization experiments.
Hyperparameter optimization (HPO) is crucial for strong performance of deep learning algorithms and real-world applications often impose some constraints, such as memory usage, or latency on top of the performance requirement. In this work, we propose constrained TPE (c-TPE), an extension of the widely-used versatile Bayesian optimization method, tree-structured Parzen estimator (TPE), to handle these constraints. Our proposed extension goes beyond a simple combination of an existing acquisition function and the original TPE, and instead includes modifications that address issues that cause poor performance. We thoroughly analyze these modifications both empirically and theoretically, providing insights into how they effectively overcome these challenges. In the experiments, we demonstrate that c-TPE exhibits the best average rank performance among existing methods with statistical significance on 81 expensive HPO with inequality constraints. Due to the lack of baselines, we only discuss the applicability of our method to hard-constrained optimization in Appendix D.
As a result of the ever increasing complexity of configuring and fine-tuning machine learning models, the field of automated machine learning (AutoML) has emerged over the past decade. However, software implementations like Auto-WEKA and Auto-sklearn typically focus on classical machine learning (ML) tasks such as classification and regression. Our work can be seen as the first attempt at offering a single AutoML framework for most problem settings that fall under the umbrella of multi-target prediction, which includes popular ML settings such as multi-label classification, multivariate regression, multi-task learning, dyadic prediction, matrix completion, and zero-shot learning. Automated problem selection and model configuration are achieved by extending DeepMTP, a general deep learning framework for MTP problem settings, with popular hyperparameter optimization (HPO) methods. Our extensive benchmarking across different datasets and MTP problem settings identifies cases where specific HPO methods outperform others.
In this work, we explore combining automatic hyperparameter tuning and optimization for federated learning (FL) in an online, one-shot procedure. We apply a principled approach on a method for adaptive client learning rate, number of local steps, and batch size. In our federated learning applications, our primary motivations are minimizing communication budget as well as local computational resources in the training pipeline. Conventionally, hyperparameter tuning methods involve at least some degree of trial-and-error, which is known to be sample inefficient. In order to address our motivations, we propose FATHOM (Federated AuTomatic Hyperparameter OptiMization) as a one-shot online procedure. We investigate the challenges and solutions of deriving analytical gradients with respect to the hyperparameters of interest. Our approach is inspired by the fact that, with the exception of local data, we have full knowledge of all components involved in our training process, and this fact can be exploited in our algorithm impactfully. We show that FATHOM is more communication efficient than Federated Averaging (FedAvg) with optimized, static valued hyperparameters, and is also more computationally efficient overall. As a communication efficient, one-shot online procedure, FATHOM solves the bottleneck of costly communication and limited local computation, by eliminating a potentially wasteful tuning process, and by optimizing the hyperparamters adaptively throughout the training procedure without trial-and-error. We show our numerical results through extensive empirical experiments with the Federated EMNIST-62 (FEMNIST) and Federated Stack Overflow (FSO) datasets, using FedJAX as our baseline framework.
The alfalfa crop is globally important as livestock feed, so highly efficient planting and harvesting could benefit many industries, especially as the global climate changes and traditional methods become less accurate. Recent work using machine learning (ML) to predict yields for alfalfa and other crops has shown promise. Previous efforts used remote sensing, weather, planting, and soil data to train machine learning models for yield prediction. However, while remote sensing works well, the models require large amounts of data and cannot make predictions until the harvesting season begins. Using weather and planting data from alfalfa variety trials in Kentucky and Georgia, our previous work compared feature selection techniques to find the best technique and best feature set. In this work, we trained a variety of machine learning models, using cross validation for hyperparameter optimization, to predict biomass yields, and we showed better accuracy than similar work that employed more complex techniques. Our best individual model was a random forest with a mean absolute error of 0.081 tons/acre and R{$^2$} of 0.941. Next, we expanded this dataset to include Wisconsin and Mississippi, and we repeated our experiments, obtaining a higher best R{$^2$} of 0.982 with a regression tree. We then isolated our testing datasets by state to explore this problem's eligibility for domain adaptation (DA), as we trained on multiple source states and tested on one target state. This Trivial DA (TDA) approach leaves plenty of room for improvement through exploring more complex DA techniques in forthcoming work.
AnalogVNN, a simulation framework built on PyTorch which can simulate the effects of optoelectronic noise, limited precision, and signal normalization present in photonic neural network accelerators. We use this framework to train and optimize linear and convolutional neural networks with up to 9 layers and ~1.7 million parameters, while gaining insights into how normalization, activation function, reduced precision, and noise influence accuracy in analog photonic neural networks. By following the same layer structure design present in PyTorch, the AnalogVNN framework allows users to convert most digital neural network models to their analog counterparts with just a few lines of code, taking full advantage of the open-source optimization, deep learning, and GPU acceleration libraries available through PyTorch. Code is available at https://analogvnn.github.io
Hyperparameter tuning is a fundamental aspect of machine learning research. Setting up the infrastructure for systematic optimization of hyperparameters can take a significant amount of time. Here, we present PyHopper, a black-box optimization platform designed to streamline the hyperparameter tuning workflow of machine learning researchers. PyHopper's goal is to integrate with existing code with minimal effort and run the optimization process with minimal necessary manual oversight. With simplicity as the primary theme, PyHopper is powered by a single robust Markov-chain Monte-Carlo optimization algorithm that scales to millions of dimensions. Compared to existing tuning packages, focusing on a single algorithm frees the user from having to decide between several algorithms and makes PyHopper easily customizable. PyHopper is publicly available under the Apache-2.0 license at https://github.com/PyHopper/PyHopper.
Sensitive datasets are often underutilized in research and industry due to privacy concerns, limiting the potential of valuable data-driven insights. Synthetic data generation presents a promising solution to address this challenge by balancing privacy protection with data utility. This paper introduces a new approach to mitigate privacy risks associated with outlier observations in synthetic datasets: the Local Resampler (LR). The LR leverages the $k$-nearest neighbors algorithm to generate synthetic data while minimizing disclosure risks by underrepresenting outliers, even when they are not detectable in marginal distributions. Theoretical and empirical analyses demonstrate that the LR effectively mitigates outlier-driven disclosure risks, and accurately replicates multimodal, skewed, and non-convex support distributions. The semiparametric nature of the LR ensures a low computational burden and works efficiently even with small samples. By parameterizing the balance between privacy risks and data utility, this approach promotes broader access to sensitive datasets for research.
For autonomous skill acquisition, robots have to learn about the physical rules governing the 3D world dynamics from their own past experience to predict and reason about plausible future outcomes. To this end, we propose a transformation-based 3D video prediction (T3VIP) approach that explicitly models the 3D motion by decomposing a scene into its object parts and predicting their corresponding rigid transformations. Our model is fully unsupervised, captures the stochastic nature of the real world, and the observational cues in image and point cloud domains constitute its learning signals. To fully leverage all the 2D and 3D observational signals, we equip our model with automatic hyperparameter optimization (HPO) to interpret the best way of learning from them. To the best of our knowledge, our model is the first generative model that provides an RGB-D video prediction of the future for a static camera. Our extensive evaluation with simulated and real-world datasets demonstrates that our formulation leads to interpretable 3D models that predict future depth videos while achieving on-par performance with 2D models on RGB video prediction. Moreover, we demonstrate that our model outperforms 2D baselines on visuomotor control. Videos, code, dataset, and pre-trained models are available at http://t3vip.cs.uni-freiburg.de.
The popularity of automated machine learning (AutoML) tools in different domains has increased over the past few years. Machine learning (ML) practitioners use AutoML tools to automate and optimize the process of feature engineering, model training, and hyperparameter optimization and so on. Recent work performed qualitative studies on practitioners' experiences of using AutoML tools and compared different AutoML tools based on their performance and provided features, but none of the existing work studied the practices of using AutoML tools in real-world projects at a large scale. Therefore, we conducted an empirical study to understand how ML practitioners use AutoML tools in their projects. To this end, we examined the top 10 most used AutoML tools and their respective usages in a large number of open-source project repositories hosted on GitHub. The results of our study show 1) which AutoML tools are mostly used by ML practitioners and 2) the characteristics of the repositories that use these AutoML tools. Also, we identified the purpose of using AutoML tools (e.g. model parameter sampling, search space management, model evaluation/error-analysis, Data/ feature transformation, and data labeling) and the stages of the ML pipeline (e.g. feature engineering) where AutoML tools are used. Finally, we report how often AutoML tools are used together in the same source code files. We hope our results can help ML practitioners learn about different AutoML tools and their usages, so that they can pick the right tool for their purposes. Besides, AutoML tool developers can benefit from our findings to gain insight into the usages of their tools and improve their tools to better fit the users' usages and needs.
We expect the generalization error to improve with more samples from a similar task, and to deteriorate with more samples from an out-of-distribution (OOD) task. In this work, we show a counter-intuitive phenomenon: the generalization error of a task can be a non-monotonic function of the number of OOD samples. As the number of OOD samples increases, the generalization error on the target task improves before deteriorating beyond a threshold. In other words, there is value in training on small amounts of OOD data. We use Fisher's Linear Discriminant on synthetic datasets and deep networks on computer vision benchmarks such as MNIST, CIFAR-10, CINIC-10, PACS and DomainNet to demonstrate and analyze this phenomenon. In the idealistic setting where we know which samples are OOD, we show that these non-monotonic trends can be exploited using an appropriately weighted objective of the target and OOD empirical risk. While its practical utility is limited, this does suggest that if we can detect OOD samples, then there may be ways to benefit from them. When we do not know which samples are OOD, we show how a number of go-to strategies such as data-augmentation, hyper-parameter optimization, and pre-training are not enough to ensure that the target generalization error does not deteriorate with the number of OOD samples in the dataset.