Johannes Gehrke

Sébastien Bubeck, Varun Chandrasekaran, Ronen Eldan et al. · microsoft-research, uw

Artificial intelligence (AI) researchers have been developing and refining large language models (LLMs) that exhibit remarkable capabilities across a variety of domains and tasks, challenging our understanding of learning and cognition. The latest model developed by OpenAI, GPT-4, was trained using an unprecedented scale of compute and data. In this paper, we report on our investigation of an early version of GPT-4, when it was still in active development by OpenAI. We contend that (this early version of) GPT-4 is part of a new cohort of LLMs (along with ChatGPT and Google's PaLM for example) that exhibit more general intelligence than previous AI models. We discuss the rising capabilities and implications of these models. We demonstrate that, beyond its mastery of language, GPT-4 can solve novel and difficult tasks that span mathematics, coding, vision, medicine, law, psychology and more, without needing any special prompting. Moreover, in all of these tasks, GPT-4's performance is strikingly close to human-level performance, and often vastly surpasses prior models such as ChatGPT. Given the breadth and depth of GPT-4's capabilities, we believe that it could reasonably be viewed as an early (yet still incomplete) version of an artificial general intelligence (AGI) system. In our exploration of GPT-4, we put special emphasis on discovering its limitations, and we discuss the challenges ahead for advancing towards deeper and more comprehensive versions of AGI, including the possible need for pursuing a new paradigm that moves beyond next-word prediction. We conclude with reflections on societal influences of the recent technological leap and future research directions.

Jayant Gupchup, Ashkan Aazami, Yaran Fan et al.

Large software systems tune hundreds of 'constants' to optimize their runtime performance. These values are commonly derived through intuition, lab tests, or A/B tests. A 'one-size-fits-all' approach is often sub-optimal as the best value depends on runtime context. In this paper, we provide an experimental approach to replace constants with learned contextual functions for Skype - a widely used real-time communication (RTC) application. We present Resonance, a system based on contextual bandits (CB). We describe experiences from three real-world experiments: applying it to the audio, video, and transport components in Skype. We surface a unique and practical challenge of performing machine learning (ML) inference in large software systems written using encapsulation principles. Finally, we open-source FeatureBroker, a library to reduce the friction in adopting ML models in such development environments

Jamie Pool, Ebrahim Beyrami, Vishak Gopal et al.

Web-scale applications can ship code on a daily to weekly cadence. These applications rely on online metrics to monitor the health of new releases. Regressions in metric values need to be detected and diagnosed as early as possible to reduce the disruption to users and product owners. Regressions in metrics can surface due to a variety of reasons: genuine product regressions, changes in user population, and bias due to telemetry loss (or processing) are among the common causes. Diagnosing the cause of these metric regressions is costly for engineering teams as they need to invest time in finding the root cause of the issue as soon as possible. We present Lumos, a Python library built using the principles of AB testing to systematically diagnose metric regressions to automate such analysis. Lumos has been deployed across the component teams in Microsoft's Real-Time Communication applications Skype and Microsoft Teams. It has enabled engineering teams to detect 100s of real changes in metrics and reject 1000s of false alarms detected by anomaly detectors. The application of Lumos has resulted in freeing up as much as 95% of the time allocated to metric-based investigations. In this work, we open source Lumos and present our results from applying it to two different components within the RTC group over millions of sessions. This general library can be coupled with any production system to manage the volume of alerting efficiently.

Chandan K. A. Reddy, Vishak Gopal, Ross Cutler et al.

The INTERSPEECH 2020 Deep Noise Suppression (DNS) Challenge is intended to promote collaborative research in real-time single-channel Speech Enhancement aimed to maximize the subjective (perceptual) quality of the enhanced speech. A typical approach to evaluate the noise suppression methods is to use objective metrics on the test set obtained by splitting the original dataset. While the performance is good on the synthetic test set, often the model performance degrades significantly on real recordings. Also, most of the conventional objective metrics do not correlate well with subjective tests and lab subjective tests are not scalable for a large test set. In this challenge, we open-sourced a large clean speech and noise corpus for training the noise suppression models and a representative test set to real-world scenarios consisting of both synthetic and real recordings. We also open-sourced an online subjective test framework based on ITU-T P.808 for researchers to reliably test their developments. We evaluated the results using P.808 on a blind test set. The results and the key learnings from the challenge are discussed. The datasets and scripts can be found here for quick access https://github.com/microsoft/DNS-Challenge.

Chandan K. A. Reddy, Ebrahim Beyrami, Harishchandra Dubey et al.

The INTERSPEECH 2020 Deep Noise Suppression Challenge is intended to promote collaborative research in real-time single-channel Speech Enhancement aimed to maximize the subjective (perceptual) quality of the enhanced speech. A typical approach to evaluate the noise suppression methods is to use objective metrics on the test set obtained by splitting the original dataset. Many publications report reasonable performance on the synthetic test set drawn from the same distribution as that of the training set. However, often the model performance degrades significantly on real recordings. Also, most of the conventional objective metrics do not correlate well with subjective tests and lab subjective tests are not scalable for a large test set. In this challenge, we open-source a large clean speech and noise corpus for training the noise suppression models and a representative test set to real-world scenarios consisting of both synthetic and real recordings. We also open source an online subjective test framework based on ITU-T P.808 for researchers to quickly test their developments. The winners of this challenge will be selected based on subjective evaluation on a representative test set using P.808 framework.

Chandan K. A. Reddy, Ebrahim Beyrami, Jamie Pool et al.

Background noise is a major source of quality impairments in Voice over Internet Protocol (VoIP) and Public Switched Telephone Network (PSTN) calls. Recent work shows the efficacy of deep learning for noise suppression, but the datasets have been relatively small compared to those used in other domains (e.g., ImageNet) and the associated evaluations have been more focused. In order to better facilitate deep learning research in Speech Enhancement, we present a noisy speech dataset (MS-SNSD) that can scale to arbitrary sizes depending on the number of speakers, noise types, and Speech to Noise Ratio (SNR) levels desired. We show that increasing dataset sizes increases noise suppression performance as expected. In addition, we provide an open-source evaluation methodology to evaluate the results subjectively at scale using crowdsourcing, with a reference algorithm to normalize the results. To demonstrate the dataset and evaluation framework we apply it to several noise suppressors and compare the subjective Mean Opinion Score (MOS) with objective quality measures such as SNR, PESQ, POLQA, and VISQOL and show why MOS is still required. Our subjective MOS evaluation is the first large scale evaluation of Speech Enhancement algorithms that we are aware of.

Ross Cutler, Yasaman Hosseinkashi, Jamie Pool et al.

A primary goal of remote collaboration tools is to provide effective and inclusive meetings for all participants. To study meeting effectiveness and meeting inclusiveness, we first conducted a large-scale email survey (N=4,425; after filtering N=3,290) at a large technology company (pre-COVID-19); using this data we derived a multivariate model of meeting effectiveness and show how it correlates with meeting inclusiveness, participation, and feeling comfortable to contribute. We believe this is the first such model of meeting effectiveness and inclusiveness. The large size of the data provided the opportunity to analyze correlations that are specific to sub-populations such as the impact of video. The model shows the following factors are correlated with inclusiveness, effectiveness, participation, and feeling comfortable to contribute in meetings: sending a pre-meeting communication, sending a post-meeting summary, including a meeting agenda, attendee location, remote-only meeting, audio/video quality and reliability, video usage, and meeting size. The model and survey results give a quantitative understanding of how and where to improve meeting effectiveness and inclusiveness and what the potential returns are. Motivated by the email survey results, we implemented a post-meeting survey into a leading computer-mediated communication (CMC) system to directly measure meeting effectiveness and inclusiveness (during COVID-19). Using initial results based on internal flighting we created a similar model of effectiveness and inclusiveness, with many of the same findings as the email survey. This shows a method of measuring and understanding these metrics which are both practical and useful in a commercial CMC system.

Nagarajan Natarajan, Ajaykrishna Karthikeyan, Prateek Jain et al.

We formalize and study ``programming by rewards'' (PBR), a new approach for specifying and synthesizing subroutines for optimizing some quantitative metric such as performance, resource utilization, or correctness over a benchmark. A PBR specification consists of (1) input features $x$, and (2) a reward function $r$, modeled as a black-box component (which we can only run), that assigns a reward for each execution. The goal of the synthesizer is to synthesize a "decision function" $f$ which transforms the features to a decision value for the black-box component so as to maximize the expected reward $E[r \circ f (x)]$ for executing decisions $f(x)$ for various values of $x$. We consider a space of decision functions in a DSL of loop-free if-then-else programs, which can branch on linear functions of the input features in a tree-structure and compute a linear function of the inputs in the leaves of the tree. We find that this DSL captures decision functions that are manually written in practice by programmers. Our technical contribution is the use of continuous-optimization techniques to perform synthesis of such decision functions as if-then-else programs. We also show that the framework is theoretically-founded ---in cases when the rewards satisfy nice properties, the synthesized code is optimal in a precise sense. We have leveraged PBR to synthesize non-trivial decision functions related to search and ranking heuristics in the PROSE codebase (an industrial strength program synthesis framework) and achieve competitive results to manually written procedures over multiple man years of tuning. We present empirical evaluation against other baseline techniques over real-world case studies (including PROSE) as well on simple synthetic benchmarks.

Zongheng Yang, Badrish Chandramouli, Chi Wang et al.

Corporations today collect data at an unprecedented and accelerating scale, making the need to run queries on large datasets increasingly important. Technologies such as columnar block-based data organization and compression have become standard practice in most commercial database systems. However, the problem of best assigning records to data blocks on storage is still open. For example, today's systems usually partition data by arrival time into row groups, or range/hash partition the data based on selected fields. For a given workload, however, such techniques are unable to optimize for the important metric of the number of blocks accessed by a query. This metric directly relates to the I/O cost, and therefore performance, of most analytical queries. Further, they are unable to exploit additional available storage to drive this metric down further. In this paper, we propose a new framework called a query-data routing tree, or qd-tree, to address this problem, and propose two algorithms for their construction based on greedy and deep reinforcement learning techniques. Experiments over benchmark and real workloads show that a qd-tree can provide physical speedups of more than an order of magnitude compared to current blocking schemes, and can reach within 2X of the lower bound for data skipping based on selectivity, while providing complete semantic descriptions of created blocks.

Joyce Fang, Martin Ellis, Bin Li et al.

Bandwidth estimation and congestion control for real-time communications (i.e., audio and video conferencing) remains a difficult problem, despite many years of research. Achieving high quality of experience (QoE) for end users requires continual updates due to changing network architectures and technologies. In this paper, we apply reinforcement learning for the first time to the problem of real-time communications (RTC), where we seek to optimize user-perceived quality. We present initial proof-of-concept results, where we learn an agent to control sending rate in an RTC system, evaluating using both network simulation and real Internet video calls. We discuss the challenges we observed, particularly in designing realistic reward functions that reflect QoE, and in bridging the gap between the training environment and real-world networks.

Chandan K A Reddy, Ross Cutler, Johannes Gehrke

Voice-over-Internet-Protocol (VoIP) calls are prone to various speech impairments due to environmental and network conditions resulting in bad user experience. A reliable audio impairment classifier helps to identify the cause for bad audio quality. The user feedback after the call can act as the ground truth labels for training a supervised classifier on a large audio dataset. However, the labels are noisy as most of the users lack the expertise to precisely articulate the impairment in the perceived speech. In this paper, we analyze the effects of massive noise in labels in training dense networks and Convolutional Neural Networks (CNN) using engineered features, spectrograms and raw audio samples as inputs. We demonstrate that CNN can generalize better on the training data with a large number of noisy labels and gives remarkably higher test performance. The classifiers were trained both on randomly generated label noise and the label noise introduced by human errors. We also show that training with noisy labels requires a significant increase in the training dataset size, which is in proportion to the amount of noise in the labels.

Jialin Ding, Umar Farooq Minhas, Jia Yu et al.

Recent work on "learned indexes" has changed the way we look at the decades-old field of DBMS indexing. The key idea is that indexes can be thought of as "models" that predict the position of a key in a dataset. Indexes can, thus, be learned. The original work by Kraska et al. shows that a learned index beats a B+Tree by a factor of up to three in search time and by an order of magnitude in memory footprint. However, it is limited to static, read-only workloads. In this paper, we present a new learned index called ALEX which addresses practical issues that arise when implementing learned indexes for workloads that contain a mix of point lookups, short range queries, inserts, updates, and deletes. ALEX effectively combines the core insights from learned indexes with proven storage and indexing techniques to achieve high performance and low memory footprint. On read-only workloads, ALEX beats the learned index from Kraska et al. by up to 2.2X on performance with up to 15X smaller index size. Across the spectrum of read-write workloads, ALEX beats B+Trees by up to 4.1X while never performing worse, with up to 2000X smaller index size. We believe ALEX presents a key step towards making learned indexes practical for a broader class of database workloads with dynamic updates.

Anderson R. Avila, Hannes Gamper, Chandan Reddy et al.

Estimating the perceived quality of an audio signal is critical for many multimedia and audio processing systems. Providers strive to offer optimal and reliable services in order to increase the user quality of experience (QoE). In this work, we present an investigation of the applicability of neural networks for non-intrusive audio quality assessment. We propose three neural network-based approaches for mean opinion score (MOS) estimation. We compare our results to three instrumental measures: the perceptual evaluation of speech quality (PESQ), the ITU-T Recommendation P.563, and the speech-to-reverberation energy ratio. Our evaluation uses a speech dataset contaminated with convolutive and additive noise, labeled using a crowd-based QoE evaluation, evaluated with Pearson correlation with MOS labels, and mean-squared-error of the estimated MOS. Our proposed approaches outperform the aforementioned instrumental measures, with a fully connected deep neural network using Mel-frequency features providing the best correlation (0.87) and the lowest mean squared error (0.15)

Jennifer Ortiz, Magdalena Balazinska, Johannes Gehrke et al.

Deep reinforcement learning is quickly changing the field of artificial intelligence. These models are able to capture a high level understanding of their environment, enabling them to learn difficult dynamic tasks in a variety of domains. In the database field, query optimization remains a difficult problem. Our goal in this work is to explore the capabilities of deep reinforcement learning in the context of query optimization. At each state, we build queries incrementally and encode properties of subqueries through a learned representation. The challenge here lies in the formation of the state transition function, which defines how the current subquery state combines with the next query operation (action) to yield the next state. As a first step in this direction, we focus the state representation problem and the formation of the state transition function. We describe our approach and show preliminary results. We further discuss how we can use the state representation to improve query optimization using reinforcement learning.

Florian Schröder, Raphael M. Reischuk, Johannes Gehrke

In the landscape of application ecosystems, today's cloud users wish to personalize not only their browsers with various extensions or their smartphones with various applications, but also the various extensions and applications themselves. The resulting personalization significantly raises the attractiveness for typical Web 2.0 users, but gives rise to various security risks and privacy concerns, such as unforeseen access to certain critical components, undesired information flow of personal information to untrusted applications, or emerging attack surfaces that were not possible before a personalization has taken place. In this paper, we propose a novel extensibility mechanism which is used for implementing personalization of existing cloud applications towards (possibly untrusted) components in a secure and privacy-friendly manner. Our model provides a clean component abstraction, thereby in particular ruling out undesired component accesses and ensuring that no undesired information flow takes place between application components -- either trusted from the base application or untrusted from various extensions. We then instantiate our model in the SAFE web application framework (WWW 2012), resulting in a novel methodology that is inspired by traditional access control and specifically designed for the newly emerging needs of extensibility in application ecosystems. We illustrate the convenient usage of our techniques by showing how to securely extend an existing social network application.

Yin Lou, Jacob Bien, Rich Caruana et al.

The generalized partially linear additive model (GPLAM) is a flexible and interpretable approach to building predictive models. It combines features in an additive manner, allowing each to have either a linear or nonlinear effect on the response. However, the choice of which features to treat as linear or nonlinear is typically assumed known. Thus, to make a GPLAM a viable approach in situations in which little is known $a~priori$ about the features, one must overcome two primary model selection challenges: deciding which features to include in the model and determining which of these features to treat nonlinearly. We introduce the sparse partially linear additive model (SPLAM), which combines model fitting and $both$ of these model selection challenges into a single convex optimization problem. SPLAM provides a bridge between the lasso and sparse additive models. Through a statistical oracle inequality and thorough simulation, we demonstrate that SPLAM can outperform other methods across a broad spectrum of statistical regimes, including the high-dimensional ($p\gg N$) setting. We develop efficient algorithms that are applied to real data sets with half a million samples and over 45,000 features with excellent predictive performance.

Vinod Nair, Rahul Kidambi, Sundararajan Sellamanickam et al.

We consider the problem of quantitatively evaluating missing value imputation algorithms. Given a dataset with missing values and a choice of several imputation algorithms to fill them in, there is currently no principled way to rank the algorithms using a quantitative metric. We develop a framework based on treating imputation evaluation as a problem of comparing two distributions and show how it can be used to compute quantitative metrics. We present an efficient procedure for applying this framework to practical datasets, demonstrate several metrics derived from the existing literature on comparing distributions, and propose a new metric called Neighborhood-based Dissimilarity Score which is fast to compute and provides similar results. Results are shown on several datasets, metrics, and imputations algorithms.