Machine Learning (cs.LG)

Abstract: To comprehend complex systems with multiple states, it is imperative to reveal the identity of these states by system outputs. Nevertheless, the mathematical models describing these systems often exhibit nonlinearity so that render the resolution of the parameter inverse problem from the observed spatiotemporal data a challenging endeavor. Starting from the observed data obtained from such systems, we propose a novel framework that facilitates the investigation of parameter identification for multi-state systems governed by spatiotemporal varying parametric partial differential equations. Our framework consists of two integral components: a constrained self-adaptive physics-informed neural network, encompassing a sub-network, as our methodology for parameter identification, and a finite mixture model approach to detect regions of probable parameter variations. Through our scheme, we can precisely ascertain the unknown varying parameters of the complex multi-state system, thereby accomplishing the inversion of the varying parameters. Furthermore, we have showcased the efficacy of our framework on two numerical cases: the 1D Burgers' equation with time-varying parameters and the 2D wave equation with a space-varying parameter.

Abstract: Heterogeneous federated multi-task learning (HFMTL) is a federated learning technique that combines heterogeneous tasks of different clients to achieve more accurate, comprehensive predictions. In real-world applications, visual and natural language tasks typically require large-scale models to extract high-level abstract features. However, large-scale models cannot be directly applied to existing federated multi-task learning methods. Existing HFML methods also disregard the impact of gradient conflicts on multi-task optimization during the federated aggregation process. In this work, we propose an innovative framework called FedBone, which enables the construction of large-scale models with better generalization from the perspective of server-client split learning and gradient projection. We split the entire model into two components: a large-scale general model (referred to as the general model) on the cloud server and multiple task-specific models (referred to as the client model) on edge clients, solving the problem of insufficient computing power on edge clients. The conflicting gradient projection technique is used to enhance the generalization of the large-scale general model between different tasks. The proposed framework is evaluated on two benchmark datasets and a real ophthalmic dataset. Comprehensive results demonstrate that FedBone efficiently adapts to heterogeneous local tasks of each client and outperforms existing federated learning algorithms in most dense prediction and classification tasks with off-the-shelf computational resources on the client side.

Abstract: Objective: To determine whether machine learning methods can generate useful potion recipes for research and teaching at Hogwarts School of Witchcraft and Wizardry. Design: Using deep neural networks to classify generated recipes into a standard drug classification system. Setting: Hogwarts School of Witchcraft and Wizardry. Data sources: 72 potion recipes from the Hogwarts curriculum, extracted from the Harry Potter Wiki. Results: Most generated recipes fall into the categories of psychoanaleptics and dermatologicals. The number of recipes predicted for each category reflected the number of training recipes. Predicted probabilities were often above 90% but some recipes were classified into 2 or more categories with similar probabilities which complicates anticipating the predicted effects. Conclusions: Machine learning powered methods are able to generate potentially useful potion recipes for teaching and research at Hogwarts. This corresponds to similar efforts in the non-magical world where such methods have been applied to identify potentially effective drug combinations.

Abstract: Graph Neural Networks (GNNs) have emerged as a powerful tool for learning from graph-structured data. However, even state-of-the-art architectures have limitations on what structures they can distinguish, imposing theoretical limits on what the networks can achieve on different datasets. In this paper, we provide a new tool called Graphtester for a comprehensive analysis of the theoretical capabilities of GNNs for various datasets, tasks, and scores. We use Graphtester to analyze over 40 different graph datasets, determining upper bounds on the performance of various GNNs based on the number of layers. Further, we show that the tool can also be used for Graph Transformers using positional node encodings, thereby expanding its scope. Finally, we demonstrate that features generated by Graphtester can be used for practical applications such as Graph Transformers, and provide a synthetic dataset to benchmark node and edge features, such as positional encodings. The package is freely available at the following URL: https://github.com/meakbiyik/graphtester.

Abstract: Goal-conditioned hierarchical reinforcement learning (GCHRL) decomposes long-horizon tasks into sub-tasks through a hierarchical framework and it has demonstrated promising results across a variety of domains. However, the high-level policy's action space is often excessively large, presenting a significant challenge to effective exploration and resulting in potentially inefficient training. Moreover, the dynamic variability of the low-level policy introduces non-stationarity to the high-level state transition function, significantly impeding the learning of the high-level policy. In this paper, we design a measure of prospect for subgoals by planning in the goal space based on the goal-conditioned value function. Building upon the measure of prospect, we propose a landmark-guided exploration strategy by integrating the measures of prospect and novelty which aims to guide the agent to explore efficiently and improve sample efficiency. To address the non-stationarity arising from the dynamic changes of the low-level policy, we apply a state-specific regularization to the learning of low-level policy, which facilitates stable learning of the hierarchical policy. The experimental results demonstrate that our proposed exploration strategy significantly outperforms the baseline methods across multiple tasks.

Abstract: We present a deep learning-based iterative approach to solve the discrete heterogeneous Helmholtz equation for high wavenumbers. Combining classical iterative multigrid solvers and convolutional neural networks (CNNs) via preconditioning, we obtain a learned neural solver that is faster and scales better than a standard multigrid solver. Our approach offers three main contributions over previous neural methods of this kind. First, we construct a multilevel U-Net-like encoder-solver CNN with an implicit layer on the coarsest grid of the U-Net, where convolution kernels are inverted. This alleviates the field of view problem in CNNs and allows better scalability. Second, we improve upon the previous CNN preconditioner in terms of the number of parameters, computation time, and convergence rates. Third, we propose a multiscale training approach that enables the network to scale to problems of previously unseen dimensions while still maintaining a reasonable training procedure. Our encoder-solver architecture can be used to generalize over different slowness models of various difficulties and is efficient at solving for many right-hand sides per slowness model. We demonstrate the benefits of our novel architecture with numerical experiments on a variety of heterogeneous two-dimensional problems at high wavenumbers.

Abstract: We study the type of solutions to which stochastic gradient descent converges when used to train a single hidden-layer multivariate ReLU network with the quadratic loss. Our results are based on a dynamical stability analysis. In the univariate case, it was shown that linearly stable minima correspond to network functions (predictors), whose second derivative has a bounded weighted $L^1$ norm. Notably, the bound gets smaller as the step size increases, implying that training with a large step size leads to `smoother' predictors. Here we generalize this result to the multivariate case, showing that a similar result applies to the Laplacian of the predictor. We demonstrate the tightness of our bound on the MNIST dataset, and show that it accurately captures the behavior of the solutions as a function of the step size. Additionally, we prove a depth separation result on the approximation power of ReLU networks corresponding to stable minima of the loss. Specifically, although shallow ReLU networks are universal approximators, we prove that stable shallow networks are not. Namely, there is a function that cannot be well-approximated by stable single hidden-layer ReLU networks trained with a non-vanishing step size. This is while the same function can be realized as a stable two hidden-layer ReLU network. Finally, we prove that if a function is sufficiently smooth (in a Sobolev sense) then it can be approximated arbitrarily well using shallow ReLU networks that correspond to stable solutions of gradient descent.

Abstract: We introduce TTSWING, a novel dataset designed for table tennis swing analysis. This dataset comprises comprehensive swing information obtained through 9-axis sensors integrated into custom-made racket grips, accompanied by anonymized demographic data of the players. We detail the data collection and annotation procedures. Furthermore, we conduct pilot studies utilizing diverse machine learning models for swing analysis. TTSWING holds tremendous potential to facilitate innovative research in table tennis analysis and is a valuable resource for the scientific community. We release the dataset and experimental codes at https://github.com/DEPhantom/TTSWING.

Abstract: Class-incremental learning aims to learn new classes in an incremental fashion without forgetting the previously learned ones. Several research works have shown how additional data can be used by incremental models to help mitigate catastrophic forgetting. In this work, following the recent breakthrough in text-to-image generative models and their wide distribution, we propose the use of a pretrained Stable Diffusion model as a source of additional data for class-incremental learning. Compared to competitive methods that rely on external, often unlabeled, datasets of real images, our approach can generate synthetic samples belonging to the same classes as the previously encountered images. This allows us to use those additional data samples not only in the distillation loss but also for replay in the classification loss. Experiments on the competitive benchmarks CIFAR100, ImageNet-Subset, and ImageNet demonstrate how this new approach can be used to further improve the performance of state-of-the-art methods for class-incremental learning on large scale datasets.

Abstract: The design of induction machine is a challenging task due to different electromagnetic and thermal constraints. Quick estimation of machine's dimensions is important in the sales tool to provide quick quotations to customers based on specific requirements. The key part of this process is to select different design parameters like length, diameter, tooth tip height and winding turns to achieve certain torque, current and temperature of the machine. Electrical machine designers, with their experience know how to alter different machine design parameters to achieve a customer specific operation requirements. We propose a reinforcement learning algorithm to design a customised induction motor. The neural network model is trained off-line by simulating different instances of of electrical machine design game with a reward or penalty function when a good or bad design choice is made. The results demonstrate that the suggested method automates electrical machine design without applying any human engineering knowledge.

Abstract: Noise injection and data augmentation strategies have been effective for enhancing the generalisation and robustness of neural networks (NNs). Certain types of noise such as label smoothing and MixUp have also been shown to improve calibration. Since noise can be added in various stages of the NN's training, it motivates the question of when and where the noise is the most effective. We study a variety of noise types to determine how much they improve calibration and generalisation, and under what conditions. More specifically we evaluate various noise-injection strategies in both in-distribution (ID) and out-of-distribution (OOD) scenarios. The findings highlight that activation noise was the most transferable and effective in improving generalisation, while input augmentation noise was prominent in improving calibration on OOD but not necessarily ID data.

Abstract: Visualization is a crucial step in exploratory data analysis. One possible approach is to train an autoencoder with low-dimensional latent space. Large network depth and width can help unfolding the data. However, such expressive networks can achieve low reconstruction error even when the latent representation is distorted. To avoid such misleading visualizations, we propose first a differential geometric perspective on the decoder, leading to insightful diagnostics for an embedding's distortion, and second a new regularizer mitigating such distortion. Our ``Geometric Autoencoder'' avoids stretching the embedding spuriously, so that the visualization captures the data structure more faithfully. It also flags areas where little distortion could not be achieved, thus guarding against misinterpretation.

Abstract: Federated learning (FL) has emerged as a promising approach for training machine learning models on decentralized data without compromising data privacy. In this paper, we propose a FL algorithm for object detection in quality inspection tasks using YOLOv5 as the object detection algorithm and Federated Averaging (FedAvg) as the FL algorithm. We apply this approach to a manufacturing use-case where multiple factories/clients contribute data for training a global object detection model while preserving data privacy on a non-IID dataset. Our experiments demonstrate that our FL approach achieves better generalization performance on the overall clients' test dataset and generates improved bounding boxes around the objects compared to models trained using local clients' datasets. This work showcases the potential of FL for quality inspection tasks in the manufacturing industry and provides valuable insights into the performance and feasibility of utilizing YOLOv5 and FedAvg for federated object detection.

Abstract: Generative flow networks (GFlowNets) are amortized variational inference algorithms that treat sampling from a distribution over compositional objects as a sequential decision-making problem with a learnable action policy. Unlike other algorithms for hierarchical sampling that optimize a variational bound, GFlowNet algorithms can stably run off-policy, which can be advantageous for discovering modes of the target distribution. Despite this flexibility in the choice of behaviour policy, the optimal way of efficiently selecting trajectories for training has not yet been systematically explored. In this paper, we view the choice of trajectories for training as an active learning problem and approach it using Bayesian techniques inspired by methods for multi-armed bandits. The proposed algorithm, Thompson sampling GFlowNets (TS-GFN), maintains an approximate posterior distribution over policies and samples trajectories from this posterior for training. We show in two domains that TS-GFN yields improved exploration and thus faster convergence to the target distribution than the off-policy exploration strategies used in past work.

Abstract: Molecular property prediction (MPP) is a crucial task in the drug discovery pipeline, which has recently gained considerable attention thanks to advances in deep neural networks. However, recent research has revealed that deep models struggle to beat traditional non-deep ones on MPP. In this study, we benchmark 12 representative models (3 non-deep models and 9 deep models) on 14 molecule datasets. Through the most comprehensive study to date, we make the following key observations: \textbf{(\romannumeral 1)} Deep models are generally unable to outperform non-deep ones; \textbf{(\romannumeral 2)} The failure of deep models on MPP cannot be solely attributed to the small size of molecular datasets. What matters is the irregular molecule data pattern; \textbf{(\romannumeral 3)} In particular, tree models using molecular fingerprints as inputs tend to perform better than other competitors. Furthermore, we conduct extensive empirical investigations into the unique patterns of molecule data and inductive biases of various models underlying these phenomena.

Abstract: We study the convergence behavior of the celebrated temporal-difference (TD) learning algorithm. By looking at the algorithm through the lens of optimization, we first argue that TD can be viewed as an iterative optimization algorithm where the function to be minimized changes per iteration. By carefully investigating the divergence displayed by TD on a classical counter example, we identify two forces that determine the convergent or divergent behavior of the algorithm. We next formalize our discovery in the linear TD setting with quadratic loss and prove that convergence of TD hinges on the interplay between these two forces. We extend this optimization perspective to prove convergence of TD in a much broader setting than just linear approximation and squared loss. Our results provide a theoretical explanation for the successful application of TD in reinforcement learning.

Abstract: Personalized cancer treatment requires a thorough understanding of complex interactions between drugs and cancer cell lines in varying genetic and molecular contexts. To address this, high-throughput screening has been used to generate large-scale drug response data, facilitating data-driven computational models. Such models can capture complex drug-cell line interactions across various contexts in a fully data-driven manner. However, accurately prioritizing the most sensitive drugs for each cell line still remains a significant challenge. To address this, we developed neural ranking approaches that leverage large-scale drug response data across multiple cell lines from diverse cancer types. Unlike existing approaches that primarily utilize regression and classification techniques for drug response prediction, we formulated the objective of drug selection and prioritization as a drug ranking problem. In this work, we proposed two neural listwise ranking methods that learn latent representations of drugs and cell lines, and then use those representations to score drugs in each cell line via a learnable scoring function. Specifically, we developed a neural listwise ranking method, List-One, on top of the existing method ListNet. Additionally, we proposed a novel listwise ranking method, List-All, that focuses on all the sensitive drugs instead of the top sensitive drug, unlike List-One. Our results demonstrate that List-All outperforms the best baseline with significant improvements of as much as 8.6% in hit@20 across 50% test cell lines. Furthermore, our analyses suggest that the learned latent spaces from our proposed methods demonstrate informative clustering structures and capture relevant underlying biological features. Moreover, our comprehensive empirical evaluation provides a thorough and objective comparison of the performance of different methods (including our proposed ones).

Abstract: The proliferation of deep learning applications in healthcare calls for data aggregation across various institutions, a practice often associated with significant privacy concerns. This concern intensifies in medical image analysis, where privacy-preserving mechanisms are paramount due to the data being sensitive in nature. Federated learning, which enables cooperative model training without direct data exchange, presents a promising solution. Nevertheless, the inherent vulnerabilities of federated learning necessitate further privacy safeguards. This study addresses this need by integrating differential privacy, a leading privacy-preserving technique, into a federated learning framework for medical image classification. We introduce a novel differentially private federated learning model and meticulously examine its impacts on privacy preservation and model performance. Our research confirms the existence of a trade-off between model accuracy and privacy settings. However, we demonstrate that strategic calibration of the privacy budget in differential privacy can uphold robust image classification performance while providing substantial privacy protection.

Abstract: Black-box zero-th order optimization is a central primitive for applications in fields as diverse as finance, physics, and engineering. In a common formulation of this problem, a designer sequentially attempts candidate solutions, receiving noisy feedback on the value of each attempt from the system. In this paper, we study scenarios in which feedback is also provided on the safety of the attempted solution, and the optimizer is constrained to limit the number of unsafe solutions that are tried throughout the optimization process. Focusing on methods based on Bayesian optimization (BO), prior art has introduced an optimization scheme -- referred to as SAFEOPT -- that is guaranteed not to select any unsafe solution with a controllable probability over feedback noise as long as strict assumptions on the safety constraint function are met. In this paper, a novel BO-based approach is introduced that satisfies safety requirements irrespective of properties of the constraint function. This strong theoretical guarantee is obtained at the cost of allowing for an arbitrary, controllable but non-zero, rate of violation of the safety constraint. The proposed method, referred to as SAFE-BOCP, builds on online conformal prediction (CP) and is specialized to the cases in which feedback on the safety constraint is either noiseless or noisy. Experimental results on synthetic and real-world data validate the advantages and flexibility of the proposed SAFE-BOCP.

Abstract: Knowing the causes of a model's unfairness helps practitioners better understand their data and algorithms. This is an important yet relatively unexplored task. We look into this problem through the lens of the training data - one of the major sources of unfairness. We ask the following questions: how would a model's fairness performance change if, in its training data, some samples (1) were collected from a different (e.g. demographic) group, (2) were labeled differently, or (3) some features were changed? In other words, we quantify the fairness influence of training samples by counterfactually intervening and changing samples based on predefined concepts, i.e. data attributes such as features (X), labels (Y), or sensitive attributes (A). To calculate a training sample's influence on the model's unfairness w.r.t a concept, we first generate counterfactual samples based on the concept, i.e. the counterfactual versions of the sample if the concept were changed. We then calculate the resulting impact on the unfairness, via influence function, if the counterfactual samples were used in training. Our framework not only helps practitioners understand the observed unfairness and repair their training data, but also leads to many other applications, e.g. detecting mislabeling, fixing imbalanced representations, and detecting fairness-targeted poisoning attacks.

Abstract: We focus on the task of approximating the optimal value function in deep reinforcement learning. This iterative process is comprised of approximately solving a sequence of optimization problems where the objective function can change per iteration. The common approach to solving the problem is to employ modern variants of the stochastic gradient descent algorithm such as Adam. These optimizers maintain their own internal parameters such as estimates of the first and the second moment of the gradient, and update these parameters over time. Therefore, information obtained in previous iterations is being used to solve the optimization problem in the current iteration. We hypothesize that this can contaminate the internal parameters of the employed optimizer in situations where the optimization landscape of the previous iterations is quite different from the current iteration. To hedge against this effect, a simple idea is to reset the internal parameters of the optimizer when starting a new iteration. We empirically investigate this resetting strategy by employing various optimizers in conjunction with the Rainbow algorithm. We demonstrate that this simple modification unleashes the true potential of modern optimizers, and significantly improves the performance of deep RL on the Atari benchmark.

Abstract: Do neural networks, trained on well-understood algorithmic tasks, reliably rediscover known algorithms for solving those tasks? Several recent studies, on tasks ranging from group arithmetic to in-context linear regression, have suggested that the answer is yes. Using modular addition as a prototypical problem, we show that algorithm discovery in neural networks is sometimes more complex. Small changes to model hyperparameters and initializations can induce the discovery of qualitatively different algorithms from a fixed training set, and even parallel implementations of multiple such algorithms. Some networks trained to perform modular addition implement a familiar Clock algorithm; others implement a previously undescribed, less intuitive, but comprehensible procedure which we term the Pizza algorithm, or a variety of even more complex procedures. Our results show that even simple learning problems can admit a surprising diversity of solutions, motivating the development of new tools for characterizing the behavior of neural networks across their algorithmic phase space.

Abstract: Due to effective pattern mining and feature representation, neural forecasting models based on deep learning have achieved great progress. The premise of effective learning is to collect sufficient data. However, in time series forecasting, it is difficult to obtain enough data, which limits the performance of neural forecasting models. To alleviate the data scarcity limitation, we design Sequence Decomposition Adaptation Network (SeDAN) which is a novel transfer architecture to improve forecasting performance on the target domain by aligning transferable knowledge from cross-domain datasets. Rethinking the transferability of features in time series data, we propose Implicit Contrastive Decomposition to decompose the original features into components including seasonal and trend features, which are easier to transfer. Then we design the corresponding adaptation methods for decomposed features in different domains. Specifically, for seasonal features, we perform joint distribution adaptation and for trend features, we design an Optimal Local Adaptation. We conduct extensive experiments on five benchmark datasets for multivariate time series forecasting. The results demonstrate the effectiveness of our SeDAN. It can provide more efficient and stable knowledge transfer.

Abstract: Machine learning from training data with a skewed distribution of examples per class can lead to models that favor performance on common classes at the expense of performance on rare ones. AudioSet has a very wide range of priors over its 527 sound event classes. Classification performance on AudioSet is usually evaluated by a simple average over per-class metrics, meaning that performance on rare classes is equal in importance to the performance on common ones. Several recent papers have used dataset balancing techniques to improve performance on AudioSet. We find, however, that while balancing improves performance on the public AudioSet evaluation data it simultaneously hurts performance on an unpublished evaluation set collected under the same conditions. By varying the degree of balancing, we show that its benefits are fragile and depend on the evaluation set. We also do not find evidence indicating that balancing improves rare class performance relative to common classes. We therefore caution against blind application of balancing, as well as against paying too much attention to small improvements on a public evaluation set.

Abstract: Regardless of the domain, forecasting the future behaviour of a running process instance is a question of interest for decision makers, especially when multiple instances interact. Fostered by the recent advances in machine learning research, several methods have been proposed to predict the next activity, outcome or remaining time of a process automatically. Still, building a model with high predictive power requires both - intrinsic knowledge of how to extract meaningful features from the event log data and a model that captures complex patterns in data. This work builds upon the recent progress in inter-case Predictive Process Monitoring (PPM) and comprehensively benchmarks the impact of inter-case features on prediction accuracy. Moreover, it includes quantum machine learning models, which are expected to provide an advantage over classical models with a scaling amount of feature dimensions. The evaluation on real-world training data from the BPI challenge shows that the inter-case features provide a significant boost by more than four percent in accuracy and quantum algorithms are indeed competitive in a handful of feature configurations. Yet, as quantum hardware is still in its early stages of development, this paper critically discusses these findings in the light of runtime, noise and the risk to overfit on the training data. Finally, the implementation of an open-source plugin demonstrates the technical feasibility to connect a state-of-the-art workflow engine such as Camunda to an IBM quantum computing cloud service.

Abstract: With the explosion of applications of Data Science, the field is has come loose from its foundations. This article argues for a new program of applied research in areas familiar to researchers in Bayesian methods in AI that are needed to ground the practice of Data Science by borrowing from AI techniques for model formulation that we term ``Decision Modelling.'' This article briefly reviews the formulation process as building a causal graphical model, then discusses the process in terms of six principles that comprise \emph{Decision Quality}, a framework from the popular business literature. We claim that any successful applied ML modelling effort must include these six principles. We explain how Decision Modelling combines a conventional machine learning model with an explicit value model. To give a specific example we show how this is done by integrating a model's ROC curve with a utility model.

Abstract: Data normalization is an essential task when modeling a classification system. When dealing with data streams, data normalization becomes especially challenging since we may not know in advance the properties of the features, such as their minimum/maximum values, and these properties may change over time. We compare the accuracies generated by eight well-known distance functions in data streams without normalization, normalized considering the statistics of the first batch of data received, and considering the previous batch received. We argue that experimental protocols for streams that consider the full stream as normalized are unrealistic and can lead to biased and poor results. Our results indicate that using the original data stream without applying normalization, and the Canberra distance, can be a good combination when no information about the data stream is known beforehand.

Abstract: Graph Neural Networks (GNNs) have emerged as a powerful tool for data-driven learning on various graph domains. They are usually based on a message-passing mechanism and have gained increasing popularity for their intuitive formulation, which is closely linked to the Weisfeiler-Lehman (WL) test for graph isomorphism to which they have been proven equivalent in terms of expressive power. In this work, we establish new generalization properties and fundamental limits of GNNs in the context of learning so-called identity effects, i.e., the task of determining whether an object is composed of two identical components or not. Our study is motivated by the need to understand the capabilities of GNNs when performing simple cognitive tasks, with potential applications in computational linguistics and chemistry. We analyze two case studies: (i) two-letters words, for which we show that GNNs trained via stochastic gradient descent are unable to generalize to unseen letters when utilizing orthogonal encodings like one-hot representations; (ii) dicyclic graphs, i.e., graphs composed of two cycles, for which we present positive existence results leveraging the connection between GNNs and the WL test. Our theoretical analysis is supported by an extensive numerical study.

Abstract: Traditional reinforcement learning methods optimize agents without considering safety, potentially resulting in unintended consequences. In this paper, we propose an optimal actor-free policy that optimizes a risk-sensitive criterion based on the conditional value at risk. The risk-sensitive objective function is modeled using an input-convex neural network ensuring convexity with respect to the actions and enabling the identification of globally optimal actions through simple gradient-following methods. Experimental results demonstrate the efficacy of our approach in maintaining effective risk control.

Abstract: Short-term forecasting of residential and commercial building energy consumption is widely used in power systems and continues to grow in importance. Data-driven short-term load forecasting (STLF), although promising, has suffered from a lack of open, large-scale datasets with high building diversity. This has hindered exploring the pretrain-then-finetune paradigm for STLF. To help address this, we present BuildingsBench, which consists of 1) Buildings-900K, a large-scale dataset of 900K simulated buildings representing the U.S. building stock, and 2) an evaluation platform with over 1,900 real residential and commercial buildings from 7 open datasets. BuildingsBench benchmarks two under-explored tasks: zero-shot STLF, where a pretrained model is evaluated on unseen buildings without fine-tuning, and transfer learning, where a pretrained model is fine-tuned on a target building. The main finding of our benchmark analysis is that synthetically pretrained models generalize surprisingly well to real commercial buildings. An exploration of the effect of increasing dataset size and diversity on zero-shot commercial building performance reveals a power-law with diminishing returns. We also show that fine-tuning pretrained models on real commercial and residential buildings improves performance for a majority of target buildings. We hope that BuildingsBench encourages and facilitates future research on generalizable STLF. All datasets and code can be accessed from \url{https://github.com/NREL/BuildingsBench}.

Abstract: Understanding the geometric properties of gradient descent dynamics is a key ingredient in deciphering the recent success of very large machine learning models. A striking observation is that trained over-parameterized models retain some properties of the optimization initialization. This "implicit bias" is believed to be responsible for some favorable properties of the trained models and could explain their good generalization properties. The purpose of this article is threefold. First, we rigorously expose the definition and basic properties of "conservation laws", which are maximal sets of independent quantities conserved during gradient flows of a given model (e.g. of a ReLU network with a given architecture) with any training data and any loss. Then we explain how to find the exact number of these quantities by performing finite-dimensional algebraic manipulations on the Lie algebra generated by the Jacobian of the model. Finally, we provide algorithms (implemented in SageMath) to: a) compute a family of polynomial laws; b) compute the number of (not necessarily polynomial) conservation laws. We provide showcase examples that we fully work out theoretically. Besides, applying the two algorithms confirms for a number of ReLU network architectures that all known laws are recovered by the algorithm, and that there are no other laws. Such computational tools pave the way to understanding desirable properties of optimization initialization in large machine learning models.

Abstract: Imbalanced data poses a significant challenge in classification as model performance is affected by insufficient learning from minority classes. Balancing methods are often used to address this problem. However, such techniques can lead to problems such as overfitting or loss of information. This study addresses a more challenging aspect of balancing methods - their impact on model behavior. To capture these changes, Explainable Artificial Intelligence tools are used to compare models trained on datasets before and after balancing. In addition to the variable importance method, this study uses the partial dependence profile and accumulated local effects techniques. Real and simulated datasets are tested, and an open-source Python package edgaro is developed to facilitate this analysis. The results obtained show significant changes in model behavior due to balancing methods, which can lead to biased models toward a balanced distribution. These findings confirm that balancing analysis should go beyond model performance comparisons to achieve higher reliability of machine learning models. Therefore, we propose a new method performance gain plot for informed data balancing strategy to make an optimal selection of balancing method by analyzing the measure of change in model behavior versus performance gain.

Abstract: Generative modeling has been used frequently in synthetic data generation. Fairness and privacy are two big concerns for synthetic data. Although Recent GAN [\cite{goodfellow2014generative}] based methods show good results in preserving privacy, the generated data may be more biased. At the same time, these methods require high computation resources. In this work, we design a fast, fair, flexible and private data generation method. We show the effectiveness of our method theoretically and empirically. We show that models trained on data generated by the proposed method can perform well (in inference stage) on real application scenarios.

Abstract: We consider the problem of evaluating forecasts of binary events whose predictions are consumed by rational agents who take an action in response to a prediction, but whose utility is unknown to the forecaster. We show that optimizing forecasts for a single scoring rule (e.g., the Brier score) cannot guarantee low regret for all possible agents. In contrast, forecasts that are well-calibrated guarantee that all agents incur sublinear regret. However, calibration is not a necessary criterion here (it is possible for miscalibrated forecasts to provide good regret guarantees for all possible agents), and calibrated forecasting procedures have provably worse convergence rates than forecasting procedures targeting a single scoring rule. Motivated by this, we present a new metric for evaluating forecasts that we call U-calibration, equal to the maximal regret of the sequence of forecasts when evaluated under any bounded scoring rule. We show that sublinear U-calibration error is a necessary and sufficient condition for all agents to achieve sublinear regret guarantees. We additionally demonstrate how to compute the U-calibration error efficiently and provide an online algorithm that achieves $O(\sqrt{T})$ U-calibration error (on par with optimal rates for optimizing for a single scoring rule, and bypassing lower bounds for the traditionally calibrated learning procedures). Finally, we discuss generalizations to the multiclass prediction setting.