Tuning hyperparameters for machine learning algorithms is a tedious task, one that is typically done manually. To enable automated hyperparameter tuning, recent works have started to use techniques based on Bayesian optimization. However, to practically enable automated tuning for large scale machine learning training pipelines, significant gaps remain in existing libraries, including lack of abstractions, fault tolerance, and flexibility to support scheduling on any distributed computing framework. To address these challenges, we present Mango, a Python library for parallel hyperparameter tuning. Mango enables the use of any distributed scheduling framework, implements intelligent parallel search strategies, and provides rich abstractions for defining complex hyperparameter search spaces that are compatible with scikit-learn. Mango is comparable in performance to Hyperopt, another widely used library. Mango is available open-source and is currently used in production at Arm Research to provide state-of-art hyperparameter tuning capabilities.

Hierarchical modeling and learning has proven very powerful in the field of Gaussian process regression and kernel methods, especially for machine learning applications and, increasingly, within the field of inverse problems more generally. The classical approach to learning hierarchical information is through Bayesian formulations of the problem, implying a posterior distribution on the hierarchical parameters or, in the case of empirical Bayes, providing an optimization criterion for them. Recent developments in the machine learning literature have suggested new criteria for hierarchical learning, based on approximation theoretic considerations that can be interpreted as variants of cross-validation, and exploiting approximation consistency in data splitting. The purpose of this paper is to compare the empirical Bayesian and approximation theoretic approaches to hierarchical learning, in terms of large data consistency, variance of estimators, robustness of the estimators to model misspecification, and computational cost. Our analysis is rooted in the setting of Mat\'ern-like Gaussian random field priors, with smoothness, amplitude and inverse lengthscale as hierarchical parameters, in the regression setting. Numerical experiments validate the theory and extend the scope of the paper beyond the Mat\'ern setting.

Deep speaker embedding models have been commonly used as a building block for speaker diarization systems; however, the speaker embedding model is usually trained according to a global loss defined on the training data, which could be sub-optimal for distinguishing speakers locally in a specific meeting session. In this work we present the first use of graph neural networks (GNNs) for the speaker diarization problem, utilizing a GNN to refine speaker embeddings locally using the structural information between speech segments inside each session. The speaker embeddings extracted by a pre-trained model are remapped into a new embedding space, in which the different speakers within a single session are better separated. The model is trained for linkage prediction in a supervised manner by minimizing the difference between the affinity matrix constructed by the refined embeddings and the ground-truth adjacency matrix. Spectral clustering is then applied on top of the refined embeddings. We show that the clustering performance of the refined speaker embeddings outperforms the original embeddings significantly on both simulated and real meeting data, and our system achieves the state-of-the-art result on the NIST SRE 2000 CALLHOME database.

A Tree Architecture of LSTM Networks for Sequential Regression with Missing Data S. Onur Sahin and Suleyman S. Kozat, Senior Member, IEEE Abstract --We investigate regression for variable length sequential data containing missing samples and introduce a novel tree architecture based on the Long Short-T erm Memory (LSTM) networks. In our architecture, we employ a variable number of LSTM networks, which use only the existing inputs in the sequence, in a treelike architecture without any statistical assumptions or imputations on the missing data, unlike all the previous approaches. In particular, we incorporate the missingness information by selecting a subset of these LSTM networks based on "presence-pattern" of a certain number of previous inputs. From the mixture of experts perspective, we train different LSTM networks as our experts for various missingness patterns and then combine their outputs to generate the final prediction. We also provide the computational complexity analysis of the proposed architecture, which is in the same order of the complexity of the conventional LSTM architectures for the sequence length. Our method can be readily extended to similar structures such as GRUs, RNNs as remarked in the paper . In the experiments, we achieve significant performance improvements with respect to the state-of-the-art methods for the well-known financial and real life datasets. I NTRODUCTION A. Preliminaries We study regression of variable length sequential data containing missing samples. Here, we sequentially receive a data sequence suffering from missing input values and estimate an unknown desired signal related to this data sequence. In most regression tasks involving sequential data, one usually assumes that we have the complete data sequence [1]. However, nearly in every real life application, the data sequences usually contain missing input values due to various reasons such as inconvenience, anomalies and cost savings [2], [3]. Furthermore, in many real life problems such as medical imaging applications [4] and finance [5], we encounter nonuniformly sampled data, which can be modelled as a missing data case [6]. To mitigate these issues, the widely used approaches make certain statistical assumptions on the missing data [7], [8], however, these assumptions usually do not hold and the This works is in part supported by Turkish Academy of Sciences Outstanding Researcher Programme and TUBIT AK Project No: 117E153.

Several public security systems depend directly on human action in numerous stages of its operation. The monitoring of public areas, for instance, is usually done with the use of cameras spread over the busiest places in large urban centers. In general, these systems depend on an operator to pay attention to the images so that the agencies responsible for security can be activated when events such as thefts, vandalism, and traffic accidents are observed. Considering the amount of information to which the operator is exposed, there is a high probability that surveillance failures will occur, even if the patrol center has a large team [1]. Although the operators are attentive at all times, this type of monitoring has some disadvantages: the images are limited to the direction in which the camera points and have low visibility at dusk and in cases of rain or bright light. Besides, events such as gunshots, alarms, distress calls, among others, are much more noticeable in the auditory field than in the visual [2, 3]. In this sense, the monitoring of risk areas could be done through the use of audio processing techniques, reducing the need for human participation in the surveillance process, and making public security systems more efficient [4]. To support this argument, it is worth recalling two very favorable characteristics concerning these signals: initially, the sound consumes less bandwidth in the transmission of information, reducing the need for high transmission rates, as in the case of high definition images; in addition, sound processing techniques require, in general, less computational power than techniques for video processing and analysis, which would enable the implementation of simpler and therefore less costly embedded systems [3, 5].

Inverse probability weighted estimators are the oldest and potentially most commonly used class of procedures for the estimation of causal effects. By adjusting for selection biases via a weighting mechanism, these procedures estimate an effect of interest by constructing a pseudo-population in which selection biases are eliminated. Despite their ease of use, these estimators require the correct specification of a model for the weighting mechanism, are known to be inefficient, and suffer from the curse of dimensionality. We propose a class of nonparametric inverse probability weighted estimators in which the weighting mechanism is estimated via undersmoothing of the highly adaptive lasso, a nonparametric regression function proven to converge at $n^{-1/3}$-rate to the true weighting mechanism. We demonstrate that our estimators are asymptotically linear with variance converging to the nonparametric efficiency bound. Unlike doubly robust estimators, our procedures require neither derivation of the efficient influence function nor specification of the conditional outcome model. Our theoretical developments have broad implications for the construction of efficient inverse probability weighted estimators in large statistical models and a variety of problem settings. We assess the practical performance of our estimators in simulation studies and demonstrate use of our proposed methodology with data from a large-scale epidemiologic study.

High dimensional integration is essential to many areas of science, ranging from particle physics to Bayesian inference. Approximating these integrals is hard, due in part to the difficulty of locating and sampling from regions of the integration domain that make significant contributions to the overall integral. Here, we present a new algorithm called Tree Quadrature (TQ) that separates this sampling problem from the problem of using those samples to produce an approximation of the integral. TQ places no qualifications on how the samples provided to it are obtained, allowing it to use state-of-the-art sampling algorithms that are largely ignored by existing integration algorithms. Given a set of samples, TQ constructs a surrogate model of the integrand in the form of a regression tree, with a structure optimised to maximise integral precision. The tree divides the integration domain into smaller containers, which are individually integrated and aggregated to estimate the overall integral. Any method can be used to integrate each individual container, so existing integration methods, like Bayesian Monte Carlo, can be combined with TQ to boost their performance. On a set of benchmark problems, we show that TQ provides accurate approximations to integrals in up to 15 dimensions; and in dimensions 4 and above, it outperforms simple Monte Carlo and the popular Vegas method (Lepage, 1978).

Building rich machine learning datasets in a scalable manner often necessitates a crowd-sourced data collection pipeline. In this work, we use human studies to investigate the consequences of employing such a pipeline, focusing on the popular ImageNet dataset. We study how specific design choices in the ImageNet creation process impact the fidelity of the resulting dataset---including the introduction of biases that state-of-the-art models exploit. Our analysis pinpoints how a noisy data collection pipeline can lead to a systematic misalignment between the resulting benchmark and the real-world task it serves as a proxy for. Finally, our findings emphasize the need to augment our current model training and evaluation toolkit to take such misalignments into account. To facilitate further research, we release our refined ImageNet annotations at https://github.com/MadryLab/ImageNetMultiLabel.

This paper studies a high-dimensional inference problem involving the matrix tensor product of random matrices. This problem generalizes a number of contemporary data science problems including the spiked matrix models used in sparse principal component analysis and covariance estimation. It is shown that the information-theoretic limits can be described succinctly by formulas involving low-dimensional quantities. On the technical side, this paper introduces some new techniques for the analysis of high-dimensional matrix-valued signals. Specific contributions include a novel extension of the adaptive interpolation method that uses order-preserving positive semidefinite interpolation paths and a variance inequality based on continuous-time I-MMSE relations.

De novo therapeutic design is challenged by a vast chemical repertoire and multiple constraints such as high broad-spectrum potency and low toxicity. We propose CLaSS (Controlled Latent attribute Space Sampling) - a novel and efficient computational method for attribute-controlled generation of molecules, which leverages guidance from classifiers trained on an informative latent space of molecules modeled using a deep generative autoencoder. We further screen the generated molecules by using a set of deep learning classifiers in conjunction with novel physicochemical features derived from high-throughput molecular simulations. The proposed approach is employed for designing non-toxic antimicrobial peptides (AMPs) with strong broad-spectrum potency, which are emerging drug candidates for tackling antibiotic resistance. Synthesis and wet lab testing of only twenty designed sequences identified two novel and minimalist AMPs with high potency against diverse Gram-positive and Gram-negative pathogens, including the hard-to-treat multidrug-resistant K. pneumoniae, as well as low in vitro and in vivo toxicity. The proposed approach thus presents a viable path for faster discovery of potent and selective broad-spectrum antimicrobials with a higher success rate than state-of-the-art methods.