Auto encoding models have been extensively studied in recent years. They provide an efficient framework for sample generation, as well as for analysing feature learning. Furthermore, they are efficient in performing interpolations between data-points in semantically meaningful ways. In this paper, we build further on a previously introduced method for generating canonical, dimension independent, stochastic interpolations. Here, the distribution of interpolation paths is represented as the distribution of a bridge process constructed from an artificial random data generating process in the latent space, having the prior distribution as its invariant distribution. As a result the stochastic interpolation paths tend to reside in regions of the latent space where the prior has high mass. This is a desirable feature since, generally, such areas produce semantically meaningful samples. In this paper, we extend the bridge process method by introducing a discriminator network that accurately identifies areas of high latent representation density. The discriminator network is incorporated as a change of measure of the underlying bridge process and sampling of interpolation paths is implemented using sequential Monte Carlo. The resulting sampling procedure allows for greater variability in interpolation paths and stronger drift towards areas of high data density.

Linear discriminant analysis (LDA) based classifiers tend to falter in many practical settings where the training data size is smaller than, or comparable to, the number of features. As a remedy, different regularized LDA (RLDA) methods have been proposed. These methods may still perform poorly depending on the size and quality of the available training data. In particular, the test data deviation from the training data model, for example, due to noise contamination, can cause severe performance degradation. Moreover, these methods commit further to the Gaussian assumption (upon which LDA is established) to tune their regularization parameters, which may compromise accuracy when dealing with real data. To address these issues, we propose a doubly regularized LDA classifier that we denote as R2LDA. In the proposed R2LDA approach, the RLDA score function is converted into an inner product of two vectors. By substituting the expressions of the regularized estimators of these vectors, we obtain the R2LDA score function that involves two regularization parameters. To set the values of these parameters, we adopt three existing regularization techniques; the constrained perturbation regularization approach (COPRA), the bounded perturbation regularization (BPR) algorithm, and the generalized cross-validation (GCV) method. These methods are used to tune the regularization parameters based on linear estimation models, with the sample covariance matrix's square root being the linear operator. Results obtained from both synthetic and real data demonstrate the consistency and effectiveness of the proposed R2LDA approach, especially in scenarios involving test data contaminated with noise that is not observed during the training phase.

The right to be forgotten states that a data subject has the right to erase their data from an entity storing it. In the context of machine learning (ML), it requires the ML model provider to remove the data subject's data from the training set used to build the ML model, a process known as \textit{machine unlearning}. While straightforward and legitimate, retraining the ML model from scratch upon receiving unlearning requests incurs high computational overhead when the training set is large. To address this issue, a number of approximate algorithms have been proposed in the domain of image and text data, among which SISA is the state-of-the-art solution. It randomly partitions the training set into multiple shards and trains a constituent model for each shard. However, directly applying SISA to the graph data can severely damage the graph structural information, and thereby the resulting ML model utility. In this paper, we propose GraphEraser, a novel machine unlearning method tailored to graph data. Its contributions include two novel graph partition algorithms, and a learning-based aggregation method. We conduct extensive experiments on five real-world datasets to illustrate the unlearning efficiency and model utility of GraphEraser. We observe that GraphEraser achieves 2.06$\times$ (small dataset) to 35.94$\times$ (large dataset) unlearning time improvement compared to retraining from scratch. On the other hand, GraphEraser achieves up to $62.5\%$ higher F1 score than that of random partitioning. In addition, our proposed learning-based aggregation method achieves up to $112\%$ higher F1 score than that of the majority vote aggregation.

I became a data scientist because I like finding solutions for complex problems, the creative part of the job and the insights I gain from the data is what I enjoy the most. The boring stuff like cleaning data, preprocessing, and tuning hyperparameters brings me little joy, and that's why I try to automate these tasks as much as possible. If you also like automating the boring stuff you will love the library I am about to introduce in this article. As I mentioned in a previous article, the current state of the art in machine learning is dominated by deep learning in the case of perceptual problems and boosting methods for regression problems. Nobody is using the linear regression from Scikit-Learn to predict house prices in a Kaggle competition these days because the XGboost method is just more accurate.

The ability to develop robust machine-learning (ML) models is considered imperative to the adoption of ML techniques in biology and medicine fields. This challenge is particularly acute when data available for training is not independent and identically distributed (iid), in which case trained models are vulnerable to out-of-distribution generalization problems. Of particular interest are problems where data correspond to observations made on phylogenetically related samples (e.g. We introduce DendroNet, a new approach to train neural networks in the context of evolutionary data. DendroNet explicitly accounts for the relatedness of the training/testing data, while allowing the model to evolve along the branches of the phylogenetic tree, hence accommodating potential changes in the rules that relate genotypes to phenotypes. Using simulated data, we demonstrate that DendroNet produces models that can be significantly better than non-phylogenetically aware approaches. DendroNet also outperforms other approaches at two biological tasks of significant practical importance: antiobiotic resistance prediction in bacteria and trophic level prediction in fungi. In supervised machine learning, most work operates under the assumption that the available data points are independent and identically distributed. Yet in many bioinformatics applications, this is not the case. This assumption is particularly strongly violated when examples are phylogenetically or genealogically related.

Clustering data into meaningful subsets is a major task in scientific data analysis. To date, various strategies ranging from model-based approaches to data-driven schemes, have been devised for efficient and accurate clustering. One important class of clustering methods that is of a particular interest is the class of exemplar-based approaches. This interest primarily stems from the amount of compressed information encoded in these exemplars that effectively reflect the major characteristics of the respective clusters. Affinity propagation (AP) has proven to be a powerful exemplar-based approach that refines the set of optimal exemplars by iterative pairwise message updates. However, a critical limitation is its inability to capitalize on known networked relations between data points often available for various scientific datasets. To mitigate this shortcoming, we propose geometric-AP, a novel clustering algorithm that effectively extends AP to take advantage of the network topology. Geometric-AP obeys network constraints and uses max-sum belief propagation to leverage the available network topology for generating smooth clusters over the network. Extensive performance assessment reveals a significant enhancement in the quality of the clustering results when compared to benchmark clustering schemes. Especially, we demonstrate that geometric-AP performs extremely well even in cases where the original AP fails drastically.

Stochastic gradient descent (SGD) is the main approach for training deep networks: it moves towards the optimum of the cost function by iteratively updating the parameters of a model in the direction of the gradient of the loss evaluated on a minibatch. Several variants of SGD have been proposed to make adaptive step sizes for each parameter (adaptive gradient) and take into account the previous updates (momentum). Among several alternative of SGD the most popular are AdaGrad, AdaDelta, RMSProp and Adam which scale coordinates of the gradient by square roots of some form of averaging of the squared coordinates in the past gradients and automatically adjust the learning rate on a parameter basis. In this work, we compare Adam based variants based on the difference between the present and the past gradients, the step size is adjusted for each parameter. We run several tests benchmarking proposed methods using medical image data. The experiments are performed using ResNet50 architecture neural network. Moreover, we have tested ensemble of networks and the fusion with ResNet50 trained with stochastic gradient descent. To combine the set of ResNet50 the simple sum rule has been applied. Proposed ensemble obtains very high performance, it obtains accuracy comparable or better than actual state of the art. To improve reproducibility and research efficiency the MATLAB source code used for this research is available at GitHub: https://github.com/LorisNanni.

The machine learning (ML) life cycle involves a series of iterative steps, from the effective gathering and preparation of the data, including complex feature engineering processes, to the presentation and improvement of results, with various algorithms to choose from in every step. Feature engineering in particular can be very beneficial for ML, leading to numerous improvements such as boosting the predictive results, decreasing computational times, reducing excessive noise, and increasing the transparency behind the decisions taken during the training. Despite that, while several visual analytics tools exist to monitor and control the different stages of the ML life cycle (especially those related to data and algorithms), feature engineering support remains inadequate. In this paper, we present FeatureEnVi, a visual analytics system specifically designed to assist with the feature engineering process. Our proposed system helps users to choose the most important feature, to transform the original features into powerful alternatives, and to experiment with different feature generation combinations. Additionally, data space slicing allows users to explore the impact of features on both local and global scales. FeatureEnVi utilizes multiple automatic feature selection techniques; furthermore, it visually guides users with statistical evidence about the influence of each feature (or subsets of features). The final outcome is the extraction of heavily engineered features, evaluated by multiple validation metrics. The usefulness and applicability of FeatureEnVi are demonstrated with two use cases, using a popular red wine quality data set and publicly available data related to vehicle recognition from their silhouettes. We also report feedback from interviews with ML experts and a visualization researcher who assessed the effectiveness of our system.

The amount of data collected from patients involved in clinical trials is continuously growing. All patient characteristics are potential covariates that could be used to improve clinical trial analysis and power. However, the restricted number of patients in phases I and II studies limits the possible number of covariates included in the analyses. In this paper, we investigate the cost/benefit ratio of including covariates in the analysis of clinical trials. Within this context, we address the long-running question "What is the optimum number of covariates to include in a clinical trial?" To further improve the cost/benefit ratio of covariates, historical data can be leveraged to pre-specify the covariate weights, which can be viewed as the definition of a new composite covariate. We analyze the use of a composite covariate while estimating the treatment effect in small clinical trials. A composite covariate limits the loss of degrees of freedom and the risk of overfitting.

The Stochastic Gradient Descent (SGD) or other algorithms derived from it are used extensively in Deep Learning, a branch of Machine Learning; but the proof of convergence is not always easy to find. The goal of this paper is to adapt various proofs from the literature in a simple format. In particular no claim of originality is made (see [1-4] for some of my recent research papers in this area); on the contrary please cite this work if you find it useful (arxiv or DOI: 10.5281/zenodo.4638695). This proof can be used in any domain where a self-contained presentation is needed.