Advances in trustworthy machine learning at Alexa AI

The team’s latest research on privacy-preserving machine learning, federated learning, and bias mitigation.

At Amazon, we take the protection of customer data very seriously. We are also committed to eliminating the biases that can exist in off-the-shelf language models — such as GPT-3 and RoBERTa — that are the basis of most modern natural-language processing. Trained on public texts, these language models are known to reflect the biases implicit in those texts.

Related content
Calibrating noise addition to word density in the embedding space improves utility of privacy-protected text.

These two topics — privacy protection and fairness — are at the core of trustworthy machine learning, an important area of research at Alexa AI. In 2021, we made contributions in the following areas:

  • Privacy-preserving machine learningDifferential privacy provides a rigorous way to quantify the privacy of machine learning models. We investigated vulnerabilities presented in the differential-privacy literature and propose computationally efficient mechanisms for protecting against them.
  • Federated learning: Federated learning (FL) is a distributed-training technique that keeps customer data on-device. Devices send only model parameter updates to the cloud, not raw data. We studied several FL challenges arising in an industrial setting.
  • Fairness in machine learning: Machine learning (ML) models should perform equally well regardless of who’s using them. But even knowing how to quantify fairness is a challenge. We introduced measures of fairness and methods to mitigate bias in ML models.
Counterfactuals.png
To reduce binary-gender disparity in a distilled GPT-2 language model, we introduce counterfactual examples, in which binary genders in real-world training examples are swapped.

Below, we summarize our research in these areas, which will be presented at ACL and ICASSP later this year. We also invite readers to participate in workshops and sessions we are organizing at NAACL 2022 and Interspeech 2022.

1. Privacy-preserving ML

The intuition behind differential privacy (DP) is that access to the outputs of a model should not provide any hint about what inputs were used to train the model. DP quantifies that intuition as a difference (in probabilities) between the outputs of a model trained on a given dataset and the outputs of the same model trained on the same dataset after a single input is removed.

One way to meet a DP privacy guarantee is to add some noise to the model parameters during training in order to obfuscate their relationship to training data. But this can compromise accuracy. The so-called privacy/utility tradeoff appears in every DP application.

Another side effect of adding a DP mechanism is increased training time. Given that training natural-language-understanding (NLU) models with large volumes of data can be prohibitively slow and that industry standards require fast training and deployment — e.g., when new features are being released — we developed a training method that meets DP requirements but remains efficient. We describe the method in a paper we’re presenting at this year’s ICASSP, “An efficient DP-SGD mechanism for large scale NLP models”.

In this work, we study the most popular DP mechanism for deep neural networks, DP-SGD, and build a computationally efficient alternative, eDP-SGD, in which we use a batch-processing scheme that leverages the GPU architecture and automates part of the hyperparameter-tuning process. While both DP-SGD and eDP-SGD provide the same privacy guarantees, we show that the training time for our mechanism is very similar to its non-DP counterpart’s. The original DP-SGD extends training time as much as 130-fold.

Related content
ADePT model transforms the texts used to train natural-language-understanding models while preserving semantic coherence.

Since we did our study, researchers have developed methods with stronger theoretical DP guarantees than the ones we impose in our paper, but our approach is consistent with those methods. Overall, this work makes DP more generally accessible and helps us integrate NLU models with DP guarantees into our production systems, where new models are frequently released, and a significant increase in training time is prohibitive.

While DP provides theoretical privacy guarantees, we are also interested in practical guarantees, i.e., measuring the amount of information that could potentially leak from a given model. In addition to the performance and training time of eDP-SGD, we also studied the correlation between theoretical and practical privacy guarantees. We measured practical privacy leakage using the most common method in the field, the success rate of membership inference attacks on a given model. Our experiments provide a general picture of how to optimize the privacy/utility trade-off using DP techniques for NLU models.

We also expanded the set of mechanisms for protecting NLU models against other types of attacks. In “Canary extraction in natural language understanding models”, which we will present at ACL 2022, we study the vulnerability of text classification models to a certain kind of white-box attack called a model inversion attack (ModIvA), where a fictional attack has access to the entire set of model parameters and intends to retrieve examples used during training. Existing model inversion techniques are applied to models with either continuous inputs or continuous outputs. In our work, we adopt a similar approach to text classification tasks where both inputs and outputs are discrete.

As new model architectures are developed that might display new types of vulnerabilities, we will continue innovating efficient ways of protecting our customers’ privacy.

Upcoming activities

2. Federated Learning

The idea behind federated learning (FL) is that, during the training of an ML model, part of the computation is delegated to customers’ devices, leveraging the processing power of those devices while avoiding the centralization of privacy-sensitive datasets. Each device modifies a common, shared model according to locally stored data, then sends an updated model to a central server that aggregates model updates and sends a new shared model to all the devices. At each round, the central server randomly selects a subset of active devices and requests that they perform updates.

Federated Learning Animation.gif
With federated learning, devices send model updates, not data, to a central server.

In the past year, we have made progress toward more-efficient FL and adapted common FL techniques to the industrial setting. For instance, in “Learnings from federated learning in the real world”, which we will present at ICASSP this year, we explore device selection strategies that differ from the standard uniform selection. In particular, we present the first study of device selection based on device “activity” — i.e., the number of available training samples.

These simple selection strategies are lightweight compared to existing methods, which require heavy computation from all the devices. They are thus more suitable to industrial applications, where millions of devices are involved. We study two different settings: the standard “static” setting, where all the data are available at once, and the more realistic “continual” setting, where customers generate new data over time, and past examples might have to be deleted to save storage space. Our experiments on training a language model with FL show that non-uniform sampling outperforms uniform sampling when applied to real-world data, for both the static and continual settings.

Related content
Amazon researchers optimize the distributed-training tool to run efficiently on the Elastic Fabric Adapter network interface.

We also expanded our understanding of FL for natural-language processing (NLP) and, in the process, made FL more accessible to the NLP community. In “FedNLP: A research platform for federated learning in natural language processing”, which will be presented later this year at NAACL, we and our colleagues at the University of Southern California and FedML systematically compare the most popular FL algorithms for four mainstream NLP tasks. We also present different methods to generate dataset partitions that are not independent and identically distributed (IID), as real-world FL methods must be robust against shifts in the distributions of the data used to train ML models.

Our analysis reveals that there is still a large gap between centralized and decentralized training under various settings, and we highlight several directions in which FL for NLP can advance. The paper represents Amazon’s contribution to the open-source framework FedNLP, which is capable of evaluating, analyzing, and developing FL methods for NLP. The codebase contains non-IID partitioning methods, enabling easy experimentation to advance the state of FL research for NLP.

We also designed methods to account for the naturally heterogeneous character of customer-generated data and applied FL to a wide variety of NLP tasks. We are aware that FL still presents many challenges, such as how to do evaluation when access to data is removed, on-device label generation for supervised tasks, and privacy-preserving communication between the server and the different devices. We are actively addressing each of these and plan to leverage our findings to improve FL-based model training and enhance associated capabilities such as analytics and model evaluation.

Upcoming activities

3. Fairness in ML

Natural-language-processing applications’ increased reliance on large language models trained on intrinsically biased web-scale corpora has amplified the importance of accurate fairness metrics and procedures for building more robust models.

In “On the intrinsic and extrinsic fairness evaluation metrics for contextualized language representations”, which we are presenting at ACL 2022, we compare two families of fairness metrics — namely extrinsic and intrinsic — that are widely used for language models. Intrinsic metrics directly probe into the fairness of language models, while extrinsic metrics evaluate the fairness of a whole system through predictions on downstream tasks.

Related content
Method significantly reduces bias while maintaining comparable performance on machine learning tasks.

For example, the contextualized embedding association test (CEAT), an intrinsic metric, measures bias through word embedding distances in semantic vector spaces, and the extrinsic metric HateXPlain measures the bias in a downstream hate speech detection system.

Our experiments show that inconsistencies between intrinsic and extrinsic metrics often reflect inconsistencies between the datasets used to evaluate them, and a clear understanding of bias in ML models requires more careful alignment of evaluation data. The results we report in the paper can help guide the NLP community as to how to best conduct fairness evaluations.

We have also designed new measures of fairness that are adapted to language-processing applications. In “Measuring fairness of text classifiers via prediction sensitivity”, which we will present at ACL 2022, we looked at sensitivity to perturbations of input as a way to measure fairness in ML models. The metric attempts to quantify the extent to which a single prediction depends on an input feature that encodes membership in an underrepresented group.

Accumulated prediction sensitivity.png
Our new bias measure, accumulated prediction sensitivity, combines the outputs of tow models, a task classifier (TC) and a protected status model (PSM).

We provide a theoretical analysis of our formulation and show a statistically significant difference between our metric’s correlation with the human notion of fairness and the existing counterfactual fairness metric’s.

Finally, we proposed a method to mitigate the biases of large language models during knowledge distillation, in which a smaller, more efficient model is trained to match the language model’s output on a particular task. Because large language models are trained on public texts, they can be biased in multiple ways, including the unfounded association of male or female genders with gender-neutral professions.

Distillation examples.png
Examples of texts generated by language models in response to gendered prompts before and after the application of our distillation method.

In another ACL paper, “Mitigating gender bias in distilled language models via counterfactual role reversal”, we introduce two modifications to the standard distillation mechanisms: data augmentation and teacher prediction perturbation.

We use our method to distill a GPT-2 language model for a text-generation task and demonstrate a substantial reduction in gender disparity, with only a minor reduction in utility. Interestingly, we find that reduced disparity in open-ended text generation may not necessarily lead to fairness on other downstream tasks. This finding underscores the importance of evaluating language model fairness along multiple metrics and tasks.

Our work on fairness in ML for NLP applications should help enable models that are more robust against the inherent biases of text datasets. There remain plenty of challenges in this field, but we strive to build models that offer the same experience to any customer, wherever and however they choose to interact with Alexa.

Upcoming activities

Related content

US, CA, San Francisco
About Twitch Launched in 2011, Twitch is a global community that comes together each day to create multiplayer entertainment: unique, live, unpredictable experiences created by the interactions of millions. We bring the joy of co-op to everything, from casual gaming to world-class esports to anime marathons, music, and art streams. Twitch also hosts TwitchCon, where we bring everyone together to celebrate, learn, and grow their personal interests and passions. We’re always live at Twitch. Stay up to date on all things Twitch on Linkedin, Twitter and on our Blog. About the role: Twitch builds data-driven machine learning solutions across several rich problem spaces: Natural Language Processing (NLP), Recommendations, Semantic Search, Classification/Categorization, Anomaly Detection, Forecasting, Safety, and HCI/Social Computing/Computational Social Science. As an Intern, you will work with a dedicated Mentor and Manager on a project in one of these problem areas. You will also be supported by an Advisor and participate in cohort activities such as research teach backs and leadership talks. This position can also be located in San Francisco, CA or virtual. You Will: Solve large-scale data problems. Design solutions for Twitch's problem spaces Explore ML and data research
LU, Luxembourg
Are you a talented and inventive scientist with a strong passion about modern data technologies and interested to improve business processes, extracting value from the data? Would you like to be a part of an organization that is aiming to use self-learning technology to process data in order to support the management of the procurement function? The Global Procurement Technology, as a part of Global Procurement Operations, is seeking a skilled Data Scientist to help build its future data intelligence in business ecosystem, working with large distributed systems of data and providing Machine Learning (ML) and Predictive Modeling expertise. You will be a member of the Data Engineering and ML Team, joining a fast-growing global organization, with a great vision to transform the Procurement field, and become the role model in the market. This team plays a strategic role supporting the core Procurement business domains as well as it is the cornerstone of any transformation and innovation initiative. Our mission is to provide a high-quality data environment to facilitate process optimization and business digitalization, on a global scale. We are supporting business initiatives, including but not limited to, strategic supplier sourcing (e.g. contracting, negotiation, spend analysis, market research, etc.), order management, supplier performance, etc. We are seeking an individual who can thrive in a fast-paced work environment, be collaborative and share knowledge and experience with his colleagues. You are expected to deliver results, but at the same time have fun with your teammates and enjoy working in the company. In Amazon, you will find all the resources required to learn new skills, grow your career, and become a better professional. You will connect with world leaders in your field and you will be tackling Data Science challenges to ensure business continuity, by taking the right decisions for your customers. As a Data Scientist in the team, you will: -be the subject matter expert to support team strategies that will take Global Procurement Operations towards world-class predictive maintenance practices and processes, driving more effective procurement functions, e.g. supplier segmentation, negotiations, shipping supplies volume forecast, spend management, etc. -have strong analytical skills and excel in the design, creation, management, and enterprise use of large data sets, combining raw data from different sources -provide technical expertise to support the development of ML models to facilitate intelligent digital services, such as Contract Lifecycle Management (CLM) and Negotiations platform -cooperate closely with different groups of stakeholders, e.g. data/software engineers, product/program managers, analysts, senior leadership, etc. to evaluate business needs and objectives to set up the best data management environment -create and share with audiences of varying levels technical papers and presentations -deal with ambiguity, prioritizing needs, and delivering results in a dynamic environment Basic qualifications -Master’s Degree in Computer Science/Engineering, Informatics, Mathematics, or a related technical discipline -3+ years of industry experience in data engineering/science, business intelligence or related field -3+ years experience in algorithm design, engineering and implementation for very-large scale applications to solve real problems -Very good knowledge of data modeling and evaluation -Very good understanding of regression modeling, forecasting techniques, time series analysis, machine-learning concepts such as supervised and unsupervised learning, classification, random forest, etc. -SQL and query performance tuning skills Preferred qualifications -2+ years of proficiency in using R, Python, Scala, Java or any modern language for data processing and statistical analysis -Experience with various RDBMS, such as PostgreSQL, MS SQL Server, MySQL, etc. -Experience architecting Big Data and ML solutions with AWS products (Redshift, DynamoDB, Lambda, S3, EMR, SageMaker, Lex, Kendra, Forecast etc.) -Experience articulating business questions and using quantitative techniques to arrive at a solution using available data -Experience with agile/scrum methodologies and its benefits of managing projects efficiently and delivering results iteratively -Excellent written and verbal communication skills including data visualization, especially in regards to quantitative topics discussed with non-technical colleagues
US, CA, San Francisco
About Twitch Launched in 2011, Twitch is a global community that comes together each day to create multiplayer entertainment: unique, live, unpredictable experiences created by the interactions of millions. We bring the joy of co-op to everything, from casual gaming to world-class esports to anime marathons, music, and art streams. Twitch also hosts TwitchCon, where we bring everyone together to celebrate, learn, and grow their personal interests and passions. We’re always live at Twitch. Stay up to date on all things Twitch on Linkedin, Twitter and on our Blog. About the role: Twitch builds data-driven machine learning solutions across several rich problem spaces: Natural Language Processing (NLP), Recommendations, Semantic Search, Classification/Categorization, Anomaly Detection, Forecasting, Safety, and HCI/Social Computing/Computational Social Science. As an Intern, you will work with a dedicated Mentor and Manager on a project in one of these problem areas. You will also be supported by an Advisor and participate in cohort activities such as research teach backs and leadership talks. This position can also be located in San Francisco, CA or virtual. You Will: Solve large-scale data problems. Design solutions for Twitch's problem spaces Explore ML and data research
US, CA, San Francisco
About Twitch Launched in 2011, Twitch is a global community that comes together each day to create multiplayer entertainment: unique, live, unpredictable experiences created by the interactions of millions. We bring the joy of co-op to everything, from casual gaming to world-class esports to anime marathons, music, and art streams. Twitch also hosts TwitchCon, where we bring everyone together to celebrate, learn, and grow their personal interests and passions. We’re always live at Twitch. Stay up to date on all things Twitch on Linkedin, Twitter and on our Blog. About the role: Twitch builds data-driven machine learning solutions across several rich problem spaces: Natural Language Processing (NLP), Recommendations, Semantic Search, Classification/Categorization, Anomaly Detection, Forecasting, Safety, and HCI/Social Computing/Computational Social Science. As an Intern, you will work with a dedicated Mentor and Manager on a project in one of these problem areas. You will also be supported by an Advisor and participate in cohort activities such as research teach backs and leadership talks. This position can also be located in San Francisco, CA or virtual. You Will: Solve large-scale data problems. Design solutions for Twitch's problem spaces Explore ML and data research
US, CA, San Francisco
About Twitch Launched in 2011, Twitch is a global community that comes together each day to create multiplayer entertainment: unique, live, unpredictable experiences created by the interactions of millions. We bring the joy of co-op to everything, from casual gaming to world-class esports to anime marathons, music, and art streams. Twitch also hosts TwitchCon, where we bring everyone together to celebrate, learn, and grow their personal interests and passions. We’re always live at Twitch. Stay up to date on all things Twitch on Linkedin, Twitter and on our Blog. About the role: Twitch builds data-driven machine learning solutions across several rich problem spaces: Natural Language Processing (NLP), Recommendations, Semantic Search, Classification/Categorization, Anomaly Detection, Forecasting, Safety, and HCI/Social Computing/Computational Social Science. As an Intern, you will work with a dedicated Mentor and Manager on a project in one of these problem areas. You will also be supported by an Advisor and participate in cohort activities such as research teach backs and leadership talks. This position can also be located in San Francisco, CA or virtual. You Will: Solve large-scale data problems. Design solutions for Twitch's problem spaces Explore ML and data research
US, CA, San Francisco
About Twitch Launched in 2011, Twitch is a global community that comes together each day to create multiplayer entertainment: unique, live, unpredictable experiences created by the interactions of millions. We bring the joy of co-op to everything, from casual gaming to world-class esports to anime marathons, music, and art streams. Twitch also hosts TwitchCon, where we bring everyone together to celebrate, learn, and grow their personal interests and passions. We’re always live at Twitch. Stay up to date on all things Twitch on Linkedin, Twitter and on our Blog. About the role: Twitch builds data-driven machine learning solutions across several rich problem spaces: Natural Language Processing (NLP), Recommendations, Semantic Search, Classification/Categorization, Anomaly Detection, Forecasting, Safety, and HCI/Social Computing/Computational Social Science. As an Intern, you will work with a dedicated Mentor and Manager on a project in one of these problem areas. You will also be supported by an Advisor and participate in cohort activities such as research teach backs and leadership talks. This position can also be located in San Francisco, CA or virtual. You Will: Solve large-scale data problems. Design solutions for Twitch's problem spaces Explore ML and data research
US, WA, Seattle
Amazon is seeking an experienced, self-directed data scientist to support the research and analytical needs of Amazon Web Services' Sales teams. This is a unique opportunity to invent new ways of leveraging our large, complex data streams to automate sales efforts and to accelerate our customers' journey to the cloud. This is a high-visibility role with significant impact potential. You, as the right candidate, are adept at executing every stage of the machine learning development life cycle in a business setting; from initial requirements gathering to through final model deployment, including adoption measurement and improvement. You will be working with large volumes of structured and unstructured data spread across multiple databases and can design and implement data pipelines to clean and merge these data for research and modeling. Beyond mathematical understanding, you have a deep intuition for machine learning algorithms that allows you to translate business problems into the right machine learning, data science, and/or statistical solutions. You’re able to pick up and grasp new research and identify applications or extensions within the team. You’re talented at communicating your results clearly to business owners in concise, non-technical language. Key job responsibilities • Work with a team of analytics & insights leads, data scientists and engineers to define business problems. • Research, develop, and deliver machine learning & statistical solutions in close partnership with end users, other science and engineering teams, and business stakeholders. • Use AWS services like SageMaker to deploy scalable ML models in the cloud. • Examples of projects include modeling usage of AWS services to optimize sales planning, recommending sales plays based on historical patterns, and building a sales-facing alert system using anomaly detection.
US, WA, Seattle
We are a team of doers working passionately to apply cutting-edge advances in deep learning in the life sciences to solve real-world problems. As a Senior Applied Science Manager you will participate in developing exciting products for customers. Our team rewards curiosity while maintaining a laser-focus in bringing products to market. Competitive candidates are responsive, flexible, and able to succeed within an open, collaborative, entrepreneurial, startup-like environment. At the leading edge of both academic and applied research in this product area, you have the opportunity to work together with a diverse and talented team of scientists, engineers, and product managers and collaborate with others teams. Location is in Seattle, US Embrace Diversity Here at Amazon, we embrace our differences. We are committed to furthering our culture of inclusion. We have ten employee-led affinity groups, reaching 40,000 employees in over 190 chapters globally. We have innovative benefit offerings, and host annual and ongoing learning experiences, including our Conversations on Race and Ethnicity (CORE) and AmazeCon (gender diversity) conferences. Amazon’s culture of inclusion is reinforced within our 14 Leadership Principles, which remind team members to seek diverse perspectives, learn and be curious, and earn trust Balance Work and Life Our team puts a high value on work-life balance. It isn’t about how many hours you spend at home or at work; it’s about the flow you establish that brings energy to both parts of your life. We believe striking the right balance between your personal and professional life is critical to life-long happiness and fulfillment. We offer flexibility in working hours and encourage you to find your own balance between your work and personal lives Mentor & Grow Careers Our team is dedicated to supporting new members. We have a broad mix of experience levels and tenures, and we’re building an environment that celebrates knowledge sharing and mentorship. Our senior members enjoy one-on-one mentoring and thorough, but kind, code reviews. We care about your career growth and strive to assign projects based on what will help each team member develop into a better-rounded engineer and enable them to take on more complex tasks in the future. Key job responsibilities • Manage high performing engineering and science teams • Hire and develop top-performing engineers, scientists, and other managers • Develop and execute on project plans and delivery commitments • Work with business, data science, software engineer, biological, and product leaders to help define product requirements and with managers, scientists, and engineers to execute on them • Build and maintain world-class customer experience and operational excellence for your deliverables
US, Virtual
The Amazon Economics Team is hiring Interns in Economics. We are looking for detail-oriented, organized, and responsible individuals who are eager to learn how to work with large and complicated data sets. Some knowledge of econometrics, as well as basic familiarity with Stata, R, or Python is necessary. Experience with SQL, UNIX, Sawtooth, and Spark would be a plus. These are full-time positions at 40 hours per week, with compensation being awarded on an hourly basis. You will learn how to build data sets and perform applied econometric analysis at Internet speed collaborating with economists, data scientists and MBAʼs. These skills will translate well into writing applied chapters in your dissertation and provide you with work experience that may help you with placement. Roughly 85% of interns from previous cohorts have converted to full time economics employment at Amazon. If you are interested, please send your CV to our mailing list at econ-internship@amazon.com.
US, WA, Seattle
Amazon internships are full-time (40 hours/week) for 12 consecutive weeks with start dates in May - July 2023. Our internship program provides hands-on learning and building experiences for students who are interested in a career in hardware engineering. This role will be based in Seattle, and candidates must be willing to work in-person. Corporate Projects (CPT) is a team that sits within the broader Corporate Development organization at Amazon. We seek to bring net-new, strategic projects to life by working together with customers and evolving projects from ZERO-to-ONE. To do so, we deploy our resources towards proofs-of-concept (POCs) and pilot programs and develop them from high-level ideas (the ZERO) to tangible short-term results that provide validating signal and a path to scale (the ONE). We work with our customers to develop and create net-new opportunities by relentlessly scouring all of Amazon and finding new and innovative ways to strengthen and/or accelerate the Amazon Flywheel. CPT seeks an Applied Science intern to work with a diverse, cross-functional team to build new, innovative customer experiences. Within CPT, you will apply both traditional and novel scientific approaches to solve and scale problems and solutions. We are a team where science meets application. A successful candidate will be a self-starter comfortable with ambiguity, strong attention to detail, and the ability to work in a fast-paced, ever-changing environment. As an Applied Science Intern, you will own the design and development of end-to-end systems. You’ll have the opportunity to create technical roadmaps, and drive production level projects that will support Amazon Science. You will work closely with Amazon scientists, and other science interns to develop solutions and deploy them into production. The ideal scientist must have the ability to work with diverse groups of people and cross-functional teams to solve complex business problems.