David Cortes

David Cortes, Carlos Juiz, Belen Bermejo

Training large-scale deep learning models has become a key challenge for the scientific community and industry. While the massive use of GPUs can significantly speed up training times, this approach has a negative impact on efficiency. In this article, we present a detailed analysis of the times reported by MLPerf Training v4.1 on four workloads: BERT, Llama2 LoRA, RetinaNet, and Stable Diffusion, showing that there are configurations that optimise the relationship between performance, GPU usage, and efficiency. The results point to a break-even point that allows training times to be reduced while maximising efficiency.

David Cortes

The isolation forest algorithm for outlier detection exploits a simple yet effective observation: if taking some multivariate data and making uniformly random cuts across the feature space recursively, it will take fewer such random cuts for an outlier to be left alone in a given subspace as compared to regular observations. The original idea proposed an outlier score based on the tree depth (number of random cuts) required for isolation, but experiments here show that using information about the size of the feature space taken and the number of points assigned to it can result in improved results in many situations without any modification to the tree structure, especially in the presence of categorical features.

David Cortes

Isolation forest or "iForest" is an intuitive and widely used algorithm for anomaly detection that follows a simple yet effective idea: in a given data distribution, if a threshold (split point) is selected uniformly at random within the range of some variable and data points are divided according to whether they are greater or smaller than this threshold, outlier points are more likely to end up alone or in the smaller partition. The original procedure suggested the choice of variable to split and split point within a variable to be done uniformly at random at each step, but this paper shows that "clustered" diverse outliers - oftentimes a more interesting class of outliers than others - can be more easily identified by applying a non-uniformly-random choice of variables and/or thresholds. Different split guiding criteria are compared and some are found to result in significantly better outlier discrimination for certain classes of outliers.

David Cortes

This work describes an outlier detection procedure (named "OutlierTree") loosely based on the GritBot software developed by RuleQuest research, which works by evaluating and following supervised decision tree splits on variables, in whose branches 1-d confidence intervals are constructed for the target variable and potential outliers flagged according to these confidence intervals. Under this logic, it's possible to produce human-readable explanations for why a given value of a variable in an observation can be considered as outlier, by considering the decision tree branch conditions along with general distribution statistics among the non-outlier observations that fell into the same branch, which can then be contrasted against the value which lies outside the CI. The supervised splits help to ensure that the generated conditions are not spurious, but rather related to the target variable and having logical breakpoints.

David Cortes

This work proposes a non-iterative strategy for missing value imputations which is guided by similarity between observations, but instead of explicitly determining distances or nearest neighbors, it assigns observations to overlapping buckets through recursive semi-random hyperplane cuts, in which weighted averages are determined as imputations for each variable. The quality of these imputations is oftentimes not as good as that of chained equations, but the proposed technique is much faster, non-iterative, can make imputations on new data without re-calculating anything, and scales easily to large and high-dimensional datasets, providing a significant boost over simple mean/median imputation in regression and classification metrics with imputed values when other methods are not feasible.

David Cortes

This work briefly explores the possibility of approximating spatial distance (alternatively, similarity) between data points using the Isolation Forest method envisioned for outlier detection. The logic is similar to that of isolation: the more similar or closer two points are, the more random splits it will take to separate them. The separation depth between two points can be standardized in the same way as the isolation depth, transforming it into a distance metric that is limited in range, centered, and in compliance with the axioms of distance. This metric presents some desirable properties such as being invariant to the scales of variables or being able to account for non-linear relationships between variables, which other metrics such as Euclidean or Mahalanobis distance do not. Extensions to the Isolation Forest method are also proposed for handling categorical variables and missing values, resulting in a more generalizable and robust metric.

David Cortes

This work explores adaptations of successful multi-armed bandits policies to the online contextual bandits scenario with binary rewards using binary classification algorithms such as logistic regression as black-box oracles. Some of these adaptations are achieved through bootstrapping or approximate bootstrapping, while others rely on other forms of randomness, resulting in more scalable approaches than previous works, and the ability to work with any type of classification algorithm. In particular, the Adaptive-Greedy algorithm shows a lot of promise, in many cases achieving better performance than upper confidence bound and Thompson sampling strategies, at the expense of more hyperparameters to tune.

David Cortes

This work explores non-negative low-rank matrix factorization based on regularized Poisson models (PF or "Poisson factorization" for short) for recommender systems with implicit-feedback data. The properties of Poisson likelihood allow a shortcut for very fast computations over zero-valued inputs, and oftentimes results in very sparse factors for both users and items. Compared to HPF (a popular Bayesian formulation of the problem with hierarchical priors), the frequentist optimization-based approach presented here tends to produce better top-N recommendations with significantly shorter fitting times, on top of having sparse solutions.

David Cortes

This work explores the ability of collective matrix factorization models in recommender systems to make predictions about users and items for which there is side information available but no feedback or interactions data, and proposes a new formulation with a faster cold-start prediction formula that can be used in real-time systems. While these cold-start recommendations are not as good as warm-start ones, they were found to be of better quality than non-personalized recommendations, and predictions about new users were found to be more reliable than those about new items. The formulation proposed here resulted in improved cold-start recommendations in many scenarios, at the expense of worse warm-start ones.