Deep-learning models have allowed effective end-to-end SE systems in the Speech Enhancement (SE) field. Most of these methods are trained using a fixed reconstruction loss in a supervised setting. Often these losses do not perfectly represent the desired perceptual quality metrics, resulting in sub-optimal performance. Recently, there have been efforts to learn the behavior of those metrics directly via neural nets for training SE models. However, an accurate estimation of the true metric function introduces statistical complexity for training because it attempts to capture the exact value of the metric. We propose an adversarial training strategy based on statistical correlation that avoids the complexity of estimating the SE metric while learning to mimic its overall behavior. We call this framework CorrGAN and show its significant improvement over standard losses of the SOTA baselines and achieve SOTA performance on the VoiceBank+DEMAND dataset.

Transformers and their attention mechanism have been revolutionary in the field of Machine Learning. While originally proposed for the language data, they quickly found their way to the image, video, graph, etc. data modalities with various signal geometries. Despite this versatility, generalizing the attention mechanism to scenarios where data is presented at different scales from potentially different modalities is not straightforward. The attempts to incorporate hierarchy and multi-modality within transformers are largely based on ad hoc heuristics, which are not seamlessly generalizable to similar problems with potentially different structures. To address this problem, in this paper, we take a fundamentally different approach: we first propose a mathematical construct to represent multi-modal, multi-scale data. We then mathematically derive the neural attention mechanics for the proposed construct from the first principle of entropy minimization. We show that the derived formulation is optimal in the sense of being the closest to the standard Softmax attention while incorporating the inductive biases originating from the hierarchical/geometric information of the problem. We further propose an efficient algorithm based on dynamic programming to compute our derived attention mechanism. By incorporating it within transformers, we show that the proposed hierarchical attention mechanism not only can be employed to train transformer models in hierarchical/multi-modal settings from scratch, but it can also be used to inject hierarchical information into classical, pre-trained transformer models post training, resulting in more efficient models in zero-shot manner.

Conditional sound separation in multi-source audio mixtures without having access to single source sound data during training is a long standing challenge. Existing mix-and-separate based methods suffer from significant performance drop with multi-source training mixtures due to the lack of supervision signal for single source separation cases during training. However, in the case of language-conditional audio separation, we do have access to corresponding text descriptions for each audio mixture in our training data, which can be seen as (rough) representations of the audio samples in the language modality. That raises the curious question of how to generate supervision signal for single-source audio extraction by leveraging the fact that single-source sounding language entities can be easily extracted from the text description. To this end, in this paper, we propose a generic bi-modal separation framework which can enhance the existing unsupervised frameworks to separate single-source signals in a target modality (i.e., audio) using the easily separable corresponding signals in the conditioning modality (i.e., language), without having access to single-source samples in the target modality during training. We empirically show that this is well within reach if we have access to a pretrained joint embedding model between the two modalities (i.e., CLAP). Furthermore, we propose to incorporate our framework into two fundamental scenarios to enhance separation performance. First, we show that our proposed methodology significantly improves the performance of purely unsupervised baselines by reducing the distribution shift between training and test samples. In particular, we show that our framework can achieve 71% boost in terms of Signal-to-Distortion Ratio (SDR) over the baseline, reaching 97.5% of the supervised learning performance. Second, we show that we can further improve the performance of the supervised learning itself by 17% if we augment it by our proposed weakly-supervised framework. Our framework achieves this by making large corpora of unsupervised data available to the supervised learning model as well as utilizing a natural, robust regularization mechanism through weak supervision from the language modality, and hence enabling a powerful semi-supervised framework for audio separation. Code is released at https://github.com/microsoft/BiModalAudioSeparation.

While deep neural networks (DNNs) have shown to be successful in several domains like computer vision, non-DNN models such as linear models and gradient boosting trees are still considered state-of-the-art over tabular data. When using these models, data scientists often author machine learning (ML) pipelines: DAG of ML operators comprising data transforms and ML models, whereby each operator is sequentially trained one-at-a-time. Conversely, when training DNNs, layers composing the neural networks are simultaneously trained using backpropagation. In this paper, we argue that the training scheme of ML pipelines is sub-optimal because it tries to optimize a single operator at a time thus losing the chance of global optimization. We therefore propose WindTunnel: a system that translates a trained ML pipeline into a pipeline of neural network modules and jointly optimizes the modules using backpropagation. We also suggest translation methodologies for several non-differentiable operators such as gradient boosting trees and categorical feature encoders. Our experiments show that fine-tuning of the translated WindTunnel pipelines is a promising technique able to increase the final accuracy.

Cloud providers have recently introduced new offerings whereby spare computing resources are accessible at discounts compared to on-demand computing. Exploiting such opportunity is challenging inasmuch as such resources are accessed with low-priority and therefore can elastically leave (through preemption) and join the computation at any time. In this paper, we design a new technique called coded elastic computing enabling distributed computations over elastic resources. The proposed technique allows machines to leave the computation without sacrificing the algorithm-level performance, and, at the same time, flexibly reduce the workload at existing machines when new ones join the computation. Leveraging coded redundancy, our approach is able to achieve similar computational cost as the original (uncoded) method when all machines are present; the cost gracefully increases when machines are preempted and reduces when machines join. The performance of the proposed technique is evaluated on matrix-vector multiplication and linear regression tasks, and shows improvements over existing techniques.

Machine Learning is transitioning from an art and science into a technology available to every developer. In the near future, every application on every platform will incorporate trained models to encode data-based decisions that would be impossible for developers to author. This presents a significant engineering challenge, since currently data science and modeling are largely decoupled from standard software development processes. This separation makes incorporating machine learning capabilities inside applications unnecessarily costly and difficult, and furthermore discourage developers from embracing ML in first place. In this paper we present ML.NET, a framework developed at Microsoft over the last decade in response to the challenge of making it easy to ship machine learning models in large software applications. We present its architecture, and illuminate the application demands that shaped it. Specifically, we introduce DataView, the core data abstraction of ML.NET which allows it to capture full predictive pipelines efficiently and consistently across training and inference lifecycles. We close the paper with a surprisingly favorable performance study of ML.NET compared to more recent entrants, and a discussion of some lessons learned.

Query optimizers are notorious for inaccurate cost estimates, leading to poor performance. The root of the problem lies in inaccurate cardinality estimates, i.e., the size of intermediate (and final) results in a query plan. These estimates also determine the resources consumed in modern shared cloud infrastructures. In this paper, we present C ARD L EARNER , a machine learning based approach to learn cardinality models from previous job executions and use them to predict the cardinalities in future jobs. The key intuition in our approach is that shared cloud workloads are often recurring and overlapping in nature, and so we could learn cardinality models for overlapping subgraph templates. We discuss various learning approaches and show how learning a large number of smaller models results in high accuracy and explainability. We further present an exploration technique to avoid learning bias by considering alternate join orders and learning cardinality models over them. We describe the feedback loop to apply the learned models back to future job executions. Finally, we show a detailed evaluation of our models (up to 5 orders of magnitude less error), query plans (60% applicability), performance (up to 100% faster, 3x fewer resources), and exploration (optimal in few 10s of executions).

Faced with the problem of characterizing systematic changes in multivariate time series in an unsupervised manner, we derive and test two methods of regularizing hidden Markov models for this task. Regularization on state transitions provides smooth transitioning among states, such that the sequences are split into broad, contiguous segments. Our methods are compared with a recent hierarchical Dirichlet process hidden Markov model (HDP-HMM) and a baseline standard hidden Markov model, of which the former suffers from poor performance on moderate-dimensional data and sensitivity to parameter settings, while the latter suffers from rapid state transitioning, over-segmentation and poor performance on a segmentation task involving human activity accelerometer data from the UCI Repository. The regularized methods developed here are able to perfectly characterize change of behavior in the human activity data for roughly half of the real-data test cases, with accuracy of 94% and low variation of information. In contrast to the HDP-HMM, our methods provide simple, drop-in replacements for standard hidden Markov model update rules, allowing standard expectation maximization (EM) algorithms to be used for learning.

This paper introduces a generic and scalable framework for automated anomaly detection on large scale time-series data. Early detection of anomalies plays a key role in maintaining consistency of person's data and protects corporations against malicious attackers. Current state of the art anomaly detection approaches suffer from scalability, use-case restrictions, difficulty of use and a large number of false positives. Our system at Yahoo, EGADS, uses a collection of anomaly detection and forecasting models with an anomaly filtering layer for accurate and scalable anomaly detection on time-series. We compare our approach against other anomaly detection systems on real and synthetic data with varying time-series characteristics. We found that our framework allows for 50-60% improvement in precision and recall for a variety of use-cases. Both the data and the framework are being open-sourced. The open-sourcing of the data, in particular, represents the first of its kind effort to establish the standard benchmark for anomaly detection.

The focus of this paper is on how to select a small sample of examples for labeling that can help us to evaluate many different classification models unknown at the time of sampling. We are particularly interested in studying the sampling strategies for problems in which the prevalence of the two classes is highly biased toward one of the classes. The evaluation measures of interest we want to estimate as accurately as possible are those obtained from the contingency table. We provide a careful theoretical analysis on sensitivity, specificity, and precision and show how sampling strategies should be adapted to the rate of skewness in data in order to effectively compute the three aforementioned evaluation measures.

Pairwise Markov Networks (PMN) are an important class of Markov networks which, due to their simplicity, are widely used in many applications such as image analysis, bioinformatics, sensor networks, etc. However, learning of Markov networks from data is a challenging task; there are many possible structures one must consider and each of these structures comes with its own parameters making it easy to overfit the model with limited data. To deal with the problem, recent learning methods build upon the L1 regularization to express the bias towards sparse network structures. In this paper, we propose a new and more flexible framework that let us bias the structure, that can, for example, encode the preference to networks with certain local substructures which as a whole exhibit some special global structure. We experiment with and show the benefit of our framework on two types of problems: learning of modular networks and learning of traffic networks models.