Massively multilingual language models have displayed strong performance in zero-shot (ZS-XLT) and few-shot (FS-XLT) cross-lingual transfer setups, where models fine-tuned on task data in a source language are transferred without any or with only a few annotated instances to the target language(s). However, current work typically overestimates model performance as fine-tuned models are frequently evaluated at model checkpoints that generalize best to validation instances in the target languages. This effectively violates the main assumptions of "true" ZS-XLT and FS-XLT. Such XLT setups require robust methods that do not depend on labeled target language data for validation and model selection. In this work, aiming to improve the robustness of "true" ZS-XLT and FS-XLT, we propose a simple and effective method that averages different checkpoints (i.e., model snapshots) during task fine-tuning. We conduct exhaustive ZS-XLT and FS-XLT experiments across higher-level semantic tasks (NLI, extractive QA) and lower-level token classification tasks (NER, POS). The results indicate that averaging model checkpoints yields systematic and consistent performance gains across diverse target languages in all tasks. Importantly, it simultaneously substantially desensitizes XLT to varying hyperparameter choices in the absence of target language validation. We also show that checkpoint averaging benefits performance when further combined with run averaging (i.e., averaging the parameters of models fine-tuned over independent runs).
Generative Adversarial Networks (GANs) have shown immense potential in fields such as text and image generation. Only very recently attempts to exploit GANs to statistical-mechanics models have been reported. Here we quantitatively test this approach by applying it to a prototypical stochastic process on a lattice. By suitably adding noise to the original data we succeed in bringing both the Generator and the Discriminator loss functions close to their ideal value. Importantly, the discreteness of the model is retained despite the noise. As typical for adversarial approaches, oscillations around the convergence limit persist also at large epochs. This undermines model selection and the quality of the generated trajectories. We demonstrate that a simple multi-model procedure where stochastic trajectories are advanced at each step upon randomly selecting a Generator leads to a remarkable increase in accuracy. This is illustrated by quantitative analysis of both the predicted equilibrium probability distribution and of the escape-time distribution. Based on the reported findings, we believe that GANs are a promising tool to tackle complex statistical dynamics by machine learning techniques
Domain generalization aims to solve the challenge of Out-of-Distribution (OOD) generalization by leveraging common knowledge learned from multiple training domains to generalize to unseen test domains. To accurately evaluate the OOD generalization ability, it is required that test data information is unavailable. However, the current domain generalization protocol may still have potential test data information leakage. This paper examines the risks of test data information leakage from two aspects of the current evaluation protocol: supervised pretraining on ImageNet and oracle model selection. We propose modifications to the current protocol that we should employ self-supervised pretraining or train from scratch instead of employing the current supervised pretraining, and we should use multiple test domains. These would result in a more precise evaluation of OOD generalization ability. We also rerun the algorithms with the modified protocol and introduce new leaderboards to encourage future research in domain generalization with a fairer comparison.
Contextual Bayesian Optimization (CBO) efficiently optimizes black-box functions with respect to design variables, while simultaneously integrating contextual information regarding the environment, such as experimental conditions. However, the relevance of contextual variables is not necessarily known beforehand. Moreover, contextual variables can sometimes be optimized themselves at an additional cost, a setting overlooked by current CBO algorithms. Cost-sensitive CBO would simply include optimizable contextual variables as part of the design variables based on their cost. Instead, we adaptively select a subset of contextual variables to include in the optimization, based on the trade-off between their relevance and the additional cost incurred by optimizing them compared to leaving them to be determined by the environment. We learn the relevance of contextual variables by sensitivity analysis of the posterior surrogate model while minimizing the cost of optimization by leveraging recent developments on early stopping for BO. We empirically evaluate our proposed Sensitivity-Analysis-Driven Contextual BO (SADCBO) method against alternatives on both synthetic and real-world experiments, together with extensive ablation studies, and demonstrate a consistent improvement across examples.
Chain-of-Thought (CoT) and Program-Aided Language Models (PAL) represent two distinct reasoning methods, each with its own strengths. CoT employs natural language, offering flexibility and interpretability, while PAL utilizes programming language, yielding more structured and rigorous logic. We introduce a model selection method to combine the best of both worlds by employing a large language model (LLM) to dynamically select between them. Our theoretical analysis underscores the feasibility of this method, which is further corroborated by empirical results. Our proposed method demonstrates significant performance improvements across eight reasoning datasets with Codex, ChatGPT, and GPT-4. Additionally, our method is complementary to self-consistency; when integrated, it can further enhance performance while significantly reducing computation costs. Moreover, we achieve new state-of-the-art results on GSM8K and SVAMP, with respective accuracies of 96.8% and 93.7%. Our code, data and prompts are available at https://github.com/XuZhao0/Model-Selection-Reasoning
Classification model selection is a process of identifying a suitable model class for a given classification task on a dataset. Traditionally, model selection is based on cross-validation, meta-learning, and user preferences, which are often time-consuming and resource-intensive. The performance of any machine learning classification task depends on the choice of the model class, the learning algorithm, and the dataset's characteristics. Our work proposes a novel method for automatic classification model selection from a set of candidate model classes by determining the empirical model-fitness for a dataset based only on its clustering indices. Clustering Indices measure the ability of a clustering algorithm to induce good quality neighborhoods with similar data characteristics. We propose a regression task for a given model class, where the clustering indices of a given dataset form the features and the dependent variable represents the expected classification performance. We compute the dataset clustering indices and directly predict the expected classification performance using the learned regressor for each candidate model class to recommend a suitable model class for dataset classification. We evaluate our model selection method through cross-validation with 60 publicly available binary class datasets and show that our top3 model recommendation is accurate for over 45 of 60 datasets. We also propose an end-to-end Automated ML system for data classification based on our model selection method. We evaluate our end-to-end system against popular commercial and noncommercial Automated ML systems using a different collection of 25 public domain binary class datasets. We show that the proposed system outperforms other methods with an excellent average rank of 1.68.
Searching for potential active compounds in large databases is a necessary step to reduce time and costs in modern drug discovery pipelines. Such virtual screening methods seek to provide predictions that allow the search space to be narrowed down. Although cheminformatics has made great progress in exploiting the potential of available big data, caution is needed to avoid introducing bias and provide useful predictions with new compounds. In this work, we propose the decision-support tool ALMERIA (Advanced Ligand Multiconformational Exploration with Robust Interpretable Artificial Intelligence) for estimating compound similarities and activity prediction based on pairwise molecular contrasts while considering their conformation variability. The methodology covers the entire pipeline from data preparation to model selection and hyperparameter optimization. It has been implemented using scalable software and methods to exploit large volumes of data -- in the order of several terabytes -- , offering a very quick response even for a large batch of queries. The implementation and experiments have been performed in a distributed computer cluster using a benchmark, the public access DUD-E database. In addition to cross-validation, detailed data split criteria have been used to evaluate the models on different data partitions to assess their true generalization ability with new compounds. Experiments show state-of-the-art performance for molecular activity prediction (ROC AUC: $0.99$, $0.96$, $0.87$), proving that the chosen data representation and modeling have good properties to generalize. Molecular conformations -- prediction performance and sensitivity analysis -- have also been evaluated. Finally, an interpretability analysis has been performed using the SHAP method.
Safety-critical applications require transparency in artificial intelligence (AI) components, but widely used convolutional neural networks (CNNs) widely used for perception tasks lack inherent interpretability. Hence, insights into what CNNs have learned are primarily based on performance metrics, because these allow, e.g., for cross-architecture CNN comparison. However, these neglect how knowledge is stored inside. To tackle this yet unsolved problem, our work proposes two methods for estimating the layer-wise similarity between semantic information inside CNN latent spaces. These allow insights into both the flow and likeness of semantic information within CNN layers, and into the degree of their similarity between different network architectures. As a basis, we use two renowned explainable artificial intelligence (XAI) techniques, which are used to obtain concept activation vectors, i.e., global vector representations in the latent space. These are compared with respect to their activation on test inputs. When applied to three diverse object detectors and two datasets, our methods reveal that (1) similar semantic concepts are learned regardless of the CNN architecture, and (2) similar concepts emerge in similar relative layer depth, independent of the total number of layers. Finally, our approach poses a promising step towards semantic model comparability and comprehension of how different CNNs process semantic information.
The cold start problem is a common challenge in various domains, including media use cases such as predicting viewership for newly launched shows on Over-The-Top (OTT) platforms. In this study, we propose a generic approach to tackle cold start problems by leveraging metadata and employing multi-model ensemble techniques. Our methodology includes feature engineering, model selection, and an ensemble approach based on a weighted average of predictions. The performance of our proposed method is evaluated using various performance metrics. Our results indicate that the multi-model ensemble approach significantly improves prediction accuracy compared to individual models.
In this study, we demonstrate a sequential experimental design for spectral measurements by active learning using parametric models as predictors. In spectral measurements, it is necessary to reduce the measurement time because of sample fragility and high energy costs. To improve the efficiency of experiments, sequential experimental designs are proposed, in which the subsequent measurement is designed by active learning using the data obtained before the measurement. Conventionally, parametric models are employed in data analysis; when employed for active learning, they are expected to afford a sequential experimental design that improves the accuracy of data analysis. However, due to the complexity of the formulas, a sequential experimental design using general parametric models has not been realized. Therefore, we applied Bayesian inference-based data analysis using the exchange Monte Carlo method to realize a sequential experimental design with general parametric models. In this study, we evaluated the effectiveness of the proposed method by applying it to Bayesian spectral deconvolution and Bayesian Hamiltonian selection in X-ray photoelectron spectroscopy. Using numerical experiments with artificial data, we demonstrated that the proposed method improves the accuracy of model selection and parameter estimation while reducing the measurement time compared with the results achieved without active learning or with active learning using the Gaussian process regression.
Recent work has shown that standard training via empirical risk minimization (ERM) can produce models that achieve high accuracy on average but low accuracy on underrepresented groups due to the prevalence of spurious features. A predominant approach to tackle this group robustness problem minimizes the worst group error (akin to a minimax strategy) on the training data, hoping it will generalize well on the testing data. However, this is often suboptimal, especially when the out-of-distribution (OOD) test data contains previously unseen groups. Inspired by ideas from the information retrieval and learning-to-rank literature, this paper first proposes to use Discounted Cumulative Gain (DCG) as a metric of model quality for facilitating better hyperparameter tuning and model selection. Being a ranking-based metric, DCG weights multiple poorly-performing groups (instead of considering just the group with the worst performance). As a natural next step, we build on our results to propose a ranking-based training method called Discounted Rank Upweighting (DRU), which differentially reweights a ranked list of poorly-performing groups in the training data to learn models that exhibit strong OOD performance on the test data. Results on several synthetic and real-world datasets highlight the superior generalization ability of our group-ranking-based (akin to soft-minimax) approach in selecting and learning models that are robust to group distributional shifts.
Accurately predicting the prices of financial time series is essential and challenging for the financial sector. Owing to recent advancements in deep learning techniques, deep learning models are gradually replacing traditional statistical and machine learning models as the first choice for price forecasting tasks. This shift in model selection has led to a notable rise in research related to applying deep learning models to price forecasting, resulting in a rapid accumulation of new knowledge. Therefore, we conducted a literature review of relevant studies over the past three years with a view to aiding researchers and practitioners in the field. This review delves deeply into deep learning-based forecasting models, presenting information on model architectures, practical applications, and their respective advantages and disadvantages. In particular, detailed information is provided on advanced models for price forecasting, such as Transformers, generative adversarial networks (GANs), graph neural networks (GNNs), and deep quantum neural networks (DQNNs). The present contribution also includes potential directions for future research, such as examining the effectiveness of deep learning models with complex structures for price forecasting, extending from point prediction to interval prediction using deep learning models, scrutinising the reliability and validity of decomposition ensembles, and exploring the influence of data volume on model performance.
Probability predictions are essential to inform decision making across many fields. Ideally, probability predictions are (i) well calibrated, (ii) accurate, and (iii) bold, i.e., spread out enough to be informative for decision making. However, there is a fundamental tension between calibration and boldness, since calibration metrics can be high when predictions are overly cautious, i.e., non-bold. The purpose of this work is to develop a Bayesian model selection-based approach to assess calibration, and a strategy for boldness-recalibration that enables practitioners to responsibly embolden predictions subject to their required level of calibration. Specifically, we allow the user to pre-specify their desired posterior probability of calibration, then maximally embolden predictions subject to this constraint. We demonstrate the method with a case study on hockey home team win probabilities and then verify the performance of our procedures via simulation. We find that very slight relaxation of calibration probability (e.g., from 0.99 to 0.95) can often substantially embolden predictions when they are well calibrated and accurate (e.g., widening hockey predictions range from .26-.78 to .10-.91).
The adoption of machine learning in applications where it is crucial to ensure fairness and accountability has led to a large number of model proposals in the literature, largely formulated as optimisation problems with constraints reducing or eliminating the effect of sensitive attributes on the response. While this approach is very flexible from a theoretical perspective, the resulting models are somewhat black-box in nature: very little can be said about their statistical properties, what are the best practices in their applied use, and how they can be extended to problems other than those they were originally designed for. Furthermore, the estimation of each model requires a bespoke implementation involving an appropriate solver which is less than desirable from a software engineering perspective. In this paper, we describe the fairml R package which implements our previous work (Scutari, Panero, and Proissl 2022) and related models in the literature. fairml is designed around classical statistical models (generalised linear models) and penalised regression results (ridge regression) to produce fair models that are interpretable and whose properties are well-known. The constraint used to enforce fairness is orthogonal to model estimation, making it possible to mix-and-match the desired model family and fairness definition for each application. Furthermore, fairml provides facilities for model estimation, model selection and validation including diagnostic plots.
We present the analysis of the topological graph descriptor Local Degree Profile (LDP), which forms a widely used structural baseline for graph classification. Our study focuses on model evaluation in the context of the recently developed fair evaluation framework, which defines rigorous routines for model selection and evaluation for graph classification, ensuring reproducibility and comparability of the results. Based on the obtained insights, we propose a new baseline algorithm called Local Topological Profile (LTP), which extends LDP by using additional centrality measures and local vertex descriptors. The new approach provides the results outperforming or very close to the latest GNNs for all datasets used. Specifically, state-of-the-art results were obtained for 4 out of 9 benchmark datasets. We also consider computational aspects of LDP-based feature extraction and model construction to propose practical improvements affecting execution speed and scalability. This allows for handling modern, large datasets and extends the portfolio of benchmarks used in graph representation learning. As the outcome of our work, we obtained LTP as a simple to understand, fast and scalable, still robust baseline, capable of outcompeting modern graph classification models such as Graph Isomorphism Network (GIN). We provide open-source implementation at \href{https://github.com/j-adamczyk/LTP}{GitHub}.
Theoretical studies on transfer learning or domain adaptation have so far focused on situations with a known hypothesis class or model; however in practice, some amount of model selection is usually involved, often appearing under the umbrella term of hyperparameter-tuning: for example, one may think of the problem of tuning for the right neural network architecture towards a target task, while leveraging data from a related source task. Now, in addition to the usual tradeoffs on approximation vs estimation errors involved in model selection, this problem brings in a new complexity term, namely, the transfer distance between source and target distributions, which is known to vary with the choice of hypothesis class. We present a first study of this problem, focusing on classification; in particular, the analysis reveals some remarkable phenomena: adaptive rates, i.e., those achievable with no distributional information, can be arbitrarily slower than oracle rates, i.e., when given knowledge on distances.
In continuum-armed bandit problems where the underlying function resides in a reproducing kernel Hilbert space (RKHS), namely, the kernelised bandit problems, an important open problem remains of how well learning algorithms can adapt if the regularity of the associated kernel function is unknown. In this work, we study adaptivity to the regularity of translation-invariant kernels, which is characterized by the decay rate of the Fourier transformation of the kernel, in the bandit setting. We derive an adaptivity lower bound, proving that it is impossible to simultaneously achieve optimal cumulative regret in a pair of RKHSs with different regularities. To verify the tightness of this lower bound, we show that an existing bandit model selection algorithm applied with minimax non-adaptive kernelised bandit algorithms matches the lower bound in dependence of $T$, the total number of steps, except for log factors. By filling in the regret bounds for adaptivity between RKHSs, we connect the statistical difficulty for adaptivity in continuum-armed bandits in three fundamental types of function spaces: RKHS, Sobolev space, and H\"older space.
LASSO regularized logistic regression is particularly useful for its built-in feature selection, allowing coefficients to be removed from deployment and producing sparse solutions. Differentially private versions of LASSO logistic regression have been developed, but generally produce dense solutions, reducing the intrinsic utility of the LASSO penalty. In this paper, we present a differentially private method for sparse logistic regression that maintains hard zeros. Our key insight is to first train a non-private LASSO logistic regression model to determine an appropriate privatized number of non-zero coefficients to use in final model selection. To demonstrate our method's performance, we run experiments on synthetic and real-world datasets.
Recently deep learning based quantitative structure-activity relationship (QSAR) models has shown surpassing performance than traditional methods for property prediction tasks in drug discovery. However, most DL based QSAR models are restricted to limited labeled data to achieve better performance, and also are sensitive to model scale and hyper-parameters. In this paper, we propose Uni-QSAR, a powerful Auto-ML tool for molecule property prediction tasks. Uni-QSAR combines molecular representation learning (MRL) of 1D sequential tokens, 2D topology graphs, and 3D conformers with pretraining models to leverage rich representation from large-scale unlabeled data. Without any manual fine-tuning or model selection, Uni-QSAR outperforms SOTA in 21/22 tasks of the Therapeutic Data Commons (TDC) benchmark under designed parallel workflow, with an average performance improvement of 6.09\%. Furthermore, we demonstrate the practical usefulness of Uni-QSAR in drug discovery domains.
Recommender Systems today are still mostly evaluated in terms of accuracy, with other aspects beyond the immediate relevance of recommendations, such as diversity, long-term user retention and fairness, often taking a back seat. Moreover, reconciling multiple performance perspectives is by definition indeterminate, presenting a stumbling block to those in the pursuit of rounded evaluation of Recommender Systems. EvalRS 2022 -- a data challenge designed around Multi-Objective Evaluation -- was a first practical endeavour, providing many insights into the requirements and challenges of balancing multiple objectives in evaluation. In this work, we reflect on EvalRS 2022 and expound upon crucial learnings to formulate a first-principles approach toward Multi-Objective model selection, and outline a set of guidelines for carrying out a Multi-Objective Evaluation challenge, with potential applicability to the problem of rounded evaluation of competing models in real-world deployments.