Rapid advancements in genome sequencing have led to the collection of vast amounts of genomics data. Researchers may be interested in using machine learning models on such data to predict the pathogenicity or clinical significance of a genetic mutation. However, many genetic datasets contain imbalanced target variables that pose challenges to machine learning models: observations are skewed/imbalanced in regression tasks or class-imbalanced in classification tasks. Genetic datasets are also often high-cardinal and contain skewed predictor variables, which poses further challenges. We aimed to investigate the effects of data preprocessing, feature selection techniques, and model selection on the performance of models trained on these datasets. We measured performance with 5-fold cross-validation and compared averaged r-squared and accuracy metrics across different combinations of techniques. We found that outliers/skew in predictor or target variables did not pose a challenge to regression models. We also found that class-imbalanced target variables and skewed predictors had little to no impact on classification performance. Random forest was the best model to use for imbalanced regression tasks. While our study uses a genetic dataset as an example of a real-world application, our findings can be generalized to any similar datasets.
Prediction of symptomatic crises in chronic diseases allows to take decisions before the symptoms occur, such as the intake of drugs to avoid the symptoms or the activation of medical alarms. The prediction horizon is in this case an important parameter in order to fulfill the pharmacokinetics of medications, or the time response of medical services. This paper presents a study about the prediction limits of a chronic disease with symptomatic crises: the migraine. For that purpose, this work develops a methodology to build predictive migraine models and to improve these predictions beyond the limits of the initial models. The maximum prediction horizon is analyzed, and its dependency on the selected features is studied. A strategy for model selection is proposed to tackle the trade off between conservative but robust predictive models, with respect to less accurate predictions with higher horizons. The obtained results show a prediction horizon close to 40 minutes, which is in the time range of the drug pharmacokinetics. Experiments have been performed in a realistic scenario where input data have been acquired in an ambulatory clinical study by the deployment of a non-intrusive Wireless Body Sensor Network. Our results provide an effective methodology for the selection of the future horizon in the development of prediction algorithms for diseases experiencing symptomatic crises.
We introduce a novel Information Criterion (IC), termed Learning under Singularity (LS), designed to enhance the functionality of the Widely Applicable Bayes Information Criterion (WBIC) and the Singular Bayesian Information Criterion (sBIC). LS is effective without regularity constraints and demonstrates stability. Watanabe defined a statistical model or a learning machine as regular if the mapping from a parameter to a probability distribution is one-to-one and its Fisher information matrix is positive definite. In contrast, models not meeting these conditions are termed singular. Over the past decade, several information criteria for singular cases have been proposed, including WBIC and sBIC. WBIC is applicable in non-regular scenarios but faces challenges with large sample sizes and redundant estimation of known learning coefficients. Conversely, sBIC is limited in its broader application due to its dependence on maximum likelihood estimates. LS addresses these limitations by enhancing the utility of both WBIC and sBIC. It incorporates the empirical loss from the Widely Applicable Information Criterion (WAIC) to represent the goodness of fit to the statistical model, along with a penalty term similar to that of sBIC. This approach offers a flexible and robust method for model selection, free from regularity constraints.
This paper presents a benchmark self-evolving framework to dynamically evaluate rapidly advancing Large Language Models (LLMs), aiming for a more accurate assessment of their capabilities and limitations. We utilize a multi-agent system to manipulate the context or question of original instances, reframing new evolving instances with high confidence that dynamically extend existing benchmarks. Towards a more scalable, robust and fine-grained evaluation, we implement six reframing operations to construct evolving instances testing LLMs against diverse queries, data noise and probing their problem-solving sub-abilities. With this framework, we extend benchmark datasets of four tasks. Experimental results show a general performance decline in most LLMs against their original results. This decline under our scalable and robust evaluations, alongside our fine-grained evaluation, more accurately reflect models' capabilities. Besides, our framework widens performance discrepancies both between different models and within the same model across various tasks, facilitating more informed model selection for specific tasks (Code and data are available at https://github.com/NanshineLoong/Self-Evolving-Benchmark).
Graph-structured data are the commonly used and have wide application scenarios in the real world. For these diverse applications, the vast variety of learning tasks, graph domains, and complex graph learning procedures present challenges for human experts when designing versatile graph learning approaches. Facing these challenges, large language models (LLMs) offer a potential solution due to the extensive knowledge and the human-like intelligence. This paper proposes a novel conceptual prototype for designing versatile graph learning methods with LLMs, with a particular focus on the "where" and "how" perspectives. From the "where" perspective, we summarize four key graph learning procedures, including task definition, graph data feature engineering, model selection and optimization, deployment and serving. We then explore the application scenarios of LLMs in these procedures across a wider spectrum. In the "how" perspective, we align the abilities of LLMs with the requirements of each procedure. Finally, we point out the promising directions that could better leverage the strength of LLMs towards versatile graph learning methods.
This study assesses deep learning models for audio classification in a clinical setting with the constraint of small datasets reflecting real-world prospective data collection. We analyze CNNs, including DenseNet and ConvNeXt, alongside transformer models like ViT, SWIN, and AST, and compare them against pre-trained audio models such as YAMNet and VGGish. Our method highlights the benefits of pre-training on large datasets before fine-tuning on specific clinical data. We prospectively collected two first-of-their-kind patient audio datasets from stroke patients. We investigated various preprocessing techniques, finding that RGB and grayscale spectrogram transformations affect model performance differently based on the priors they learn from pre-training. Our findings indicate CNNs can match or exceed transformer models in small dataset contexts, with DenseNet-Contrastive and AST models showing notable performance. This study highlights the significance of incremental marginal gains through model selection, pre-training, and preprocessing in sound classification; this offers valuable insights for clinical diagnostics that rely on audio classification.
In Sequential Recommendation Systems (SRecsys), traditional training approaches that rely on Cross-Entropy (CE) loss often prioritize accuracy but fail to align well with user satisfaction metrics. CE loss focuses on maximizing the confidence of the ground truth item, which is challenging to achieve universally across all users and sessions. It also overlooks the practical acceptability of ranking the ground truth item within the top-$K$ positions, a common metric in SRecsys. To address this limitation, we propose \textbf{CPFT}, a novel fine-tuning framework that integrates Conformal Prediction (CP)-based losses with CE loss to optimize accuracy alongside confidence that better aligns with widely used top-$K$ metrics. CPFT embeds CP principles into the training loop using differentiable proxy losses and computationally efficient calibration strategies, enabling the generation of high-confidence prediction sets. These sets focus on items with high relevance while maintaining robust coverage guarantees. Extensive experiments on five real-world datasets and four distinct sequential models demonstrate that CPFT improves precision metrics and confidence calibration. Our results highlight the importance of confidence-aware fine-tuning in delivering accurate, trustworthy recommendations that enhance user satisfaction.
We investigate model assessment and selection in a changing environment, by synthesizing datasets from both the current time period and historical epochs. To tackle unknown and potentially arbitrary temporal distribution shift, we develop an adaptive rolling window approach to estimate the generalization error of a given model. This strategy also facilitates the comparison between any two candidate models by estimating the difference of their generalization errors. We further integrate pairwise comparisons into a single-elimination tournament, achieving near-optimal model selection from a collection of candidates. Theoretical analyses and numerical experiments demonstrate the adaptivity of our proposed methods to the non-stationarity in data.
Foundation models have recently expanded into robotics after excelling in computer vision and natural language processing. The models are accessible in two ways: open-source or paid, closed-source options. Users with access to both face a problem when deciding between effective yet costly closed-source models and free but less powerful open-source alternatives. We call it the model selection problem. Existing supervised-learning methods are impractical due to the high cost of collecting extensive training data from closed-source models. Hence, we focus on the online learning setting where algorithms learn while collecting data, eliminating the need for large pre-collected datasets. We thus formulate a user-centric online model selection problem and propose a novel solution that combines an open-source encoder to output context and an online learning algorithm that processes this context. The encoder distills vast data distributions into low-dimensional features, i.e., the context, without additional training. The online learning algorithm aims to maximize a composite reward that includes model performance, execution time, and costs based on the context extracted from the data. It results in an improved trade-off between selecting open-source and closed-source models compared to non-contextual methods, as validated by our theoretical analysis. Experiments across language-based robotic tasks such as Waymo Open Dataset, ALFRED, and Open X-Embodiment demonstrate real-world applications of the solution. The results show that the solution significantly improves the task success rate by up to 14%.
Large language models (LLMs) have recently received considerable attention as alternative solutions for task planning. However, comparing the performance of language-oriented task planners becomes difficult, and there exists a dearth of detailed exploration regarding the effects of various factors such as pre-trained model selection and prompt construction. To address this, we propose a benchmark system for automatically quantifying performance of task planning for home-service embodied agents. Task planners are tested on two pairs of datasets and simulators: 1) ALFRED and AI2-THOR, 2) an extension of Watch-And-Help and VirtualHome. Using the proposed benchmark system, we perform extensive experiments with LLMs and prompts, and explore several enhancements of the baseline planner. We expect that the proposed benchmark tool would accelerate the development of language-oriented task planners.
This paper presents a comprehensive local projections (LP) framework for estimating future responses to current shocks, robust to high-dimensional controls without relying on sparsity assumptions. The approach is applicable to various settings, including impulse response analysis and difference-in-differences (DiD) estimation. While methods like LASSO exist, they often assume most parameters are exactly zero, limiting their effectiveness in dense data generation processes. I propose a novel technique incorporating high-dimensional covariates in local projections using the Orthogonal Greedy Algorithm with a high-dimensional AIC (OGA+HDAIC) model selection method. This approach offers robustness in both sparse and dense scenarios, improved interpretability, and more reliable causal inference in local projections. Simulation studies show superior performance in dense and persistent scenarios compared to conventional LP and LASSO-based approaches. In an empirical application to Acemoglu, Naidu, Restrepo, and Robinson (2019), I demonstrate efficiency gains and robustness to a large set of controls. Additionally, I examine the effect of subjective beliefs on economic aggregates, demonstrating robustness to various model specifications. A novel state-dependent analysis reveals that inflation behaves more in line with rational expectations in good states, but exhibits more subjective, pessimistic dynamics in bad states.
The deployment of ever-larger machine learning models reflects a growing consensus that the more expressive the model class one optimizes over$\unicode{x2013}$and the more data one has access to$\unicode{x2013}$the more one can improve performance. As models get deployed in a variety of real-world scenarios, they inevitably face strategic environments. In this work, we consider the natural question of how the interplay of models and strategic interactions affects the relationship between performance at equilibrium and the expressivity of model classes. We find that strategic interactions can break the conventional view$\unicode{x2013}$meaning that performance does not necessarily monotonically improve as model classes get larger or more expressive (even with infinite data). We show the implications of this result in several contexts including strategic regression, strategic classification, and multi-agent reinforcement learning. In particular, we show that each of these settings admits a Braess' paradox-like phenomenon in which optimizing over less expressive model classes allows one to achieve strictly better equilibrium outcomes. Motivated by these examples, we then propose a new paradigm for model selection in games wherein an agent seeks to choose amongst different model classes to use as their action set in a game.
Model selection for a given target task can be costly, as it may entail extensive annotation of the quality of outputs of different models. We introduce DiffUse, an efficient method to make an informed decision between candidate text generation models based on preference annotations. DiffUse reduces the required amount of annotations, thus saving valuable time and resources in performing evaluation. DiffUse intelligently selects instances by clustering embeddings that represent the semantic differences between model outputs. Thus, it is able to identify a subset of examples that are more informative for preference decisions. Our method is model-agnostic, and can be applied to any text generation model for selecting between models, prompts and configurations. Moreover, we propose a practical iterative approach for dynamically determining how many instances to annotate. In a series of experiments over hundreds of model pairs, we demonstrate that DiffUse can dramatically reduce the required number of annotations -- by up to 75% -- while maintaining high evaluation reliability.
Recovery of signals with elements defined on the nodes of a graph, from compressive measurements is an important problem, which can arise in various domains such as sensor networks, image reconstruction and group testing. In some scenarios, the graph may not be accurately known, and there may exist a few edge additions or deletions relative to a ground truth graph. Such perturbations, even if small in number, significantly affect the Graph Fourier Transform (GFT). This impedes recovery of signals which may have sparse representations in the GFT bases of the ground truth graph. We present an algorithm which simultaneously recovers the signal from the compressive measurements and also corrects the graph perturbations. We analyze some important theoretical properties of the algorithm. Our approach to correction for graph perturbations is based on model selection techniques such as cross-validation in compressed sensing. We validate our algorithm on signals which have a sparse representation in the GFT bases of many commonly used graphs in the network science literature. An application to compressive image reconstruction is also presented, where graph perturbations are modeled as undesirable graph edges linking pixels with significant intensity difference. In all experiments, our algorithm clearly outperforms baseline techniques which either ignore the perturbations or use first order approximations to the perturbations in the GFT bases.
Graph Neural Networks (GNNs) can be trained to detect communities within a graph by learning from the duality of feature and connectivity information. Currently, the common approach for optimisation of GNNs is to use comparisons to ground-truth for hyperparameter tuning and model selection. In this work, we show that nodes can be clustered into communities with GNNs by solely optimising for modularity, without any comparison to ground-truth. Although modularity is a graph partitioning quality metric, we show that this can be used to optimise GNNs that also encode features without a drop in performance. We take it a step further and also study whether the unsupervised metric performance can predict ground-truth performance. To investigate why modularity can be used to optimise GNNs, we design synthetic experiments that show the limitations of this approach. The synthetic graphs are created to highlight current capabilities in distinct, random and zero information space partitions in attributed graphs. We conclude that modularity can be used for hyperparameter optimisation and model selection on real-world datasets as well as being a suitable proxy for predicting ground-truth performance, however, GNNs fail to balance the information duality when the spaces contain conflicting signals.
Change-point processes are one flexible approach to model long time series. We propose a method to uncover which model parameter truly vary when a change-point is detected. Given a set of breakpoints, we use a penalized likelihood approach to select the best set of parameters that changes over time and we prove that the penalty function leads to a consistent selection of the true model. Estimation is carried out via the deterministic annealing expectation-maximization algorithm. Our method accounts for model selection uncertainty and associates a probability to all the possible time-varying parameter specifications. Monte Carlo simulations highlight that the method works well for many time series models including heteroskedastic processes. For a sample of 14 Hedge funds (HF) strategies, using an asset based style pricing model, we shed light on the promising ability of our method to detect the time-varying dynamics of risk exposures as well as to forecast HF returns.
Large Language Models (LLMs) have ushered in a new era of text annotation, as their ease-of-use, high accuracy, and relatively low costs have meant that their use has exploded in recent months. However, the rapid growth of the field has meant that LLM-based annotation has become something of an academic Wild West: the lack of established practices and standards has led to concerns about the quality and validity of research. Researchers have warned that the ostensible simplicity of LLMs can be misleading, as they are prone to bias, misunderstandings, and unreliable results. Recognizing the transformative potential of LLMs, this paper proposes a comprehensive set of standards and best practices for their reliable, reproducible, and ethical use. These guidelines span critical areas such as model selection, prompt engineering, structured prompting, prompt stability analysis, rigorous model validation, and the consideration of ethical and legal implications. The paper emphasizes the need for a structured, directed, and formalized approach to using LLMs, aiming to ensure the integrity and robustness of text annotation practices, and advocates for a nuanced and critical engagement with LLMs in social scientific research.
We propose that small pretrained foundational generative language models with millions of parameters can be utilized as a general learning framework for sequence-based tasks. Our proposal overcomes the computational resource, skill set, and timeline challenges associated with training neural networks and language models from scratch. Further, our approach focuses on creating small and highly specialized models that can accurately execute a challenging task of which the base model is incapable of performing. We demonstrate that 125M, 350M, and 1.3B parameter pretrained foundational language models can be instruction fine-tuned with 10,000-to-1,000,000 instruction examples to achieve near state-of-the-art results on challenging cheminformatics tasks. We also demonstrate the role of successive language model fine-tuning epochs on improved outcomes, as well as the importance of both data formatting and pretrained foundational language model selection for instruction fine-tuning success.
This work formulates model selection as an infinite-armed bandit problem, namely, a problem in which a decision maker iteratively selects one of an infinite number of fixed choices (i.e., arms) when the properties of each choice are only partially known at the time of allocation and may become better understood over time, via the attainment of rewards.Here, the arms are machine learning models to train and selecting an arm corresponds to a partial training of the model (resource allocation).The reward is the accuracy of the selected model after its partial training.We aim to identify the best model at the end of a finite number of resource allocations and thus consider the best arm identification setup. We propose the algorithm Mutant-UCB that incorporates operators from evolutionary algorithms into the UCB-E (Upper Confidence Bound Exploration) bandit algorithm introduced by Audiber et al.Tests carried out on three open source image classification data sets attest to the relevance of this novel combining approach, which outperforms the state-of-the-art for a fixed budget.
Recent studies have highlighted the benefits of generating multiple synthetic datasets for supervised learning, from increased accuracy to more effective model selection and uncertainty estimation. These benefits have clear empirical support, but the theoretical understanding of them is currently very light. We seek to increase the theoretical understanding by deriving bias-variance decompositions for several settings of using multiple synthetic datasets, including differentially private synthetic data. Our theory yields a simple rule of thumb to select the appropriate number of synthetic datasets in the case of mean-squared error and Brier score. We investigate how our theory works in practice with several real datasets, downstream predictors and error metrics. As our theory predicts, multiple synthetic datasets often improve accuracy, while a single large synthetic dataset gives at best minimal improvement, showing that our insights are practically relevant.