In reinforcement learning, slower networks can learn faster

When optimizing for a new solution in deep reinforcement learning, it helps if the optimizer gravitates toward the previous solution.

Reinforcement learning (RL) is an increasingly popular way to model sequential decision-making problems in artificial intelligence. RL agents learn through trial and error, repeatedly interacting with the world to learn a policy that maximizes a reward signal.

Related content
The field motivated him to pursue a PhD, which eventually led him to Amazon.

RL agents have recently achieved remarkable results when used in conjunction with deep neural networks. Chief among these so-called deep-RL results is the 2015 paper that introduced the Deep Q Network (DQN) agent, which surpassed human-level performance on a large set of Atari games.

A core component of DQN is an optimizer that adapts the parameters of the neural network to minimize the DQN objective. We typically use optimization algorithms that are standard in deep learning, but these algorithms were not designed to account for the intricacies that arise when solving deep RL.

At this year's Conference on Neural Information Processing Systems (NeurIPS), my coauthors and I presented a new optimizer that is better equipped to deal with the difficulties of RL. The optimizer makes use of a simple technique, called proximal updates, that enables us to hedge against noisy updates by ensuring that the weights of the neural network change smoothly and slowly. To achieve this, we steer the network toward its previous solution when there is no indication that doing so would harm the agent.

RL hyperparameters.png
The researchers tuned the hyperparameter for their network (c-tilde) by testing several options on six representative games and choosing the option that maximized the reinforcement learning reward function for all six. In this figure, distance from the center of the circle corresponds to reward-function value.

Tuning gravity

We show in the paper that the DQN agent could best be thought of as solving a series of optimization problems. At each iteration, the new optimization problem is based on the previous iterate, or the network weights that resulted from the last iteration. Also known as the target network in the deep-RL literature, the previous iterate is the solution we gravitate toward.

Related content
With a new method, agents can cope better with the differences between simulated training environments and real-world deployment.

While the target network encodes the previous solution, a second network — called the online network in the literature — finds the new solution. This network is updated at each step by moving in the direction that minimizes the DQN objective.

The gradient vector from minimizing the DQN objective needs to be large enough to cancel the default gravity toward the previous solution (the target network). If the online and target networks are close, the proximal update would behave similarly to the standard DQN update. But if the two networks are far apart, the proximal update can be significantly different from the DQN update, in that it would encourage closing the gap between the two networks. In our formulation, we can tune the degree of gravity exerted by the previous solution, with noisier updates requiring higher gravity.

While proximal updates lead to slower shifts in the neural-network parameters, they also lead to faster improvement at obtaining high rewards, the primary quantity of interest in RL. We show in our paper that this improvement applies to both the interim performance of the agent and its asymptotic. It also applies both in the context of planning with noise and in the context of learning on large-scale domains where the presence of noise is all but guaranteed.

Evaluation

To evaluate our approach in the learning setting, we added proximal updates to two standard RL algorithms: the DQN algorithm mentioned above and the more competitive Rainbow algorithm, which combines various existing algorithmic improvements in RL.

Human-normalized performance.png
A comparison of DQN Pro, Rainbow Pro, and their original counterparts. The x-axis indicates the number of environmental interactions (frames), and the y-axis indicates the median human-normalized performance over 55 games. Higher is better, and any point with a y coordinate larger than 1 has surpassed human performance.

We then tested the new algorithms, called DQN with Proximal updates (or DQN Pro) and Rainbow Pro on a standard set of 55 Atari games. We can see from the graph of the results that (1) the Pro agents overperform their counterparts; (2) the basic DQN agent is able to obtain human-level performance after 120 million interactions with the environment (frames); and (3) Rainbow Pro achieves a 40% relative improvement over the original Rainbow agent.

Further, to ensure that proximal updates do in fact result in smoother and slower parameter changes, we measure the norm differences between consecutive DQN solutions. We expect the magnitude of our updates to be smaller when using proximal updates. In the graphs below, we confirm this expectation on the four different Atari games tested.

Performance on representative games.png
Magnitude of updates to target network (distance between two consecutive solutions in DQN iterations). Using proximal updates results in smaller updates, as desired.

Overall, our empirical and theoretical results support the claim that when optimizing for a new solution in deep RL, it is beneficial for the optimizer to gravitate toward the previous solution. More importantly, we see that simple improvements in deep-RL optimization can lead to significant positive gains in the agent’s performance. We take this as evidence that further exploration of optimization algorithms in deep RL would be fruitful.

We have released the source code for our solution on GitHub.

Research areas

Related content

CA, ON, Toronto
Are you motivated to explore research in ambiguous spaces? Are you interested in conducting research that will improve associate, employee and manager experiences at Amazon? Do you want to work on an interdisciplinary team of scientists that collaborate rather than compete? Join us at PXT Central Science! The People eXperience and Technology Central Science Team (PXTCS) uses economics, behavioral science, statistics, and machine learning to proactively identify mechanisms and process improvements which simultaneously improve Amazon and the lives, wellbeing, and the value of work to Amazonians. We are an interdisciplinary team that combines the talents of science and engineering to develop and deliver solutions that measurably achieve this goal. Key job responsibilities As an Applied Scientist for People Experience and Technology (PXT) Central Science, you will be working with our science and engineering teams, specifically on re-imagining Generative AI Applications and Generative AI Infrastructure for HR. Applying Generative AI to HR has unique challenges such as privacy, fairness, and seamlessly integrating Enterprise Knowledge and World Knowledge and knowing which to use when. In addition, the team works on some of Amazon’s most strategic technical investments in the people space and support Amazon’s efforts to be Earth’s Best Employer. In this role you will have a significant impact on 1.5 million Amazonians and the communities Amazon serves and ample scope to demonstrate scientific thought leadership and scientific impact in addition to business impact. You will also play a critical role in the organization's business planning, work closely with senior leaders to develop goals and resource requirements, influence our long-term technical and business strategy, and help hire and develop science and engineering talent. You will also provide support to business partners, helping them use the best scientific methods and science-driven tools to solve current and upcoming challenges and deliver efficiency gains in a changing marke About the team The AI/ML team in PXTCS is working on building Generative AI solutions to reimagine Corp employee and Ops associate experience. Examples of state-of-the-art solutions are Coaching for Amazon employees (available on AZA) and reinventing Employee Recruiting and Employee Listening.
CA, ON, Toronto
Conversational AI ModEling and Learning (CAMEL) team is part of Amazon Devices organization where our mission is to build a best-in-class Conversational AI that is intuitive, intelligent, and responsive, by developing superior Large Language Models (LLM) solutions and services which increase the capabilities built into the model and which enable utilizing thousands of APIs and external knowledge sources to provide the best experience for each request across millions of customers and endpoints. We are looking for a passionate, talented, and resourceful Applied Scientist in the field of LLM, Artificial Intelligence (AI), Natural Language Processing (NLP), Recommender Systems and/or Information Retrieval, to invent and build scalable solutions for a state-of-the-art context-aware conversational AI. A successful candidate will have strong machine learning background and a desire to push the envelope in one or more of the above areas. The ideal candidate would also have hands-on experiences in building Generative AI solutions with LLMs, enjoy operating in dynamic environments, be self-motivated to take on challenging problems to deliver big customer impact, moving fast to ship solutions and then iterating on user feedback and interactions. Key job responsibilities As an Applied Scientist, you will leverage your technical expertise and experience to collaborate with other talented applied scientists and engineers to research and develop novel algorithms and modeling techniques to reduce friction and enable natural and contextual conversations. You will analyze, understand and improve user experiences by leveraging Amazon’s heterogeneous data sources and large-scale computing resources to accelerate advances in artificial intelligence. You will work on core LLM technologies, including Prompt Engineering and Optimization, Supervised Fine-Tuning, Learning from Human Feedback, Evaluation, Self-Learning, etc. Your work will directly impact our customers in the form of novel products and services.
CA, ON, Toronto
Conversational AI ModEling and Learning (CAMEL) team is part of Amazon Devices organization where our mission is to build a best-in-class Conversational AI that is intuitive, intelligent, and responsive, by developing superior Large Language Models (LLM) solutions and services which increase the capabilities built into the model and which enable utilizing thousands of APIs and external knowledge sources to provide the best experience for each request across millions of customers and endpoints. We are looking for a passionate, talented, and resourceful Applied Scientist in the field of LLM, Artificial Intelligence (AI), Natural Language Processing (NLP), Recommender Systems and/or Information Retrieval, to invent and build scalable solutions for a state-of-the-art context-aware conversational AI. A successful candidate will have strong machine learning background and a desire to push the envelope in one or more of the above areas. The ideal candidate would also have hands-on experiences in building Generative AI solutions with LLMs, enjoy operating in dynamic environments, be self-motivated to take on challenging problems to deliver big customer impact, moving fast to ship solutions and then iterating on user feedback and interactions. Key job responsibilities As an Applied Scientist, you will leverage your technical expertise and experience to collaborate with other talented applied scientists and engineers to research and develop novel algorithms and modeling techniques to reduce friction and enable natural and contextual conversations. You will analyze, understand and improve user experiences by leveraging Amazon’s heterogeneous data sources and large-scale computing resources to accelerate advances in artificial intelligence. You will work on core LLM technologies, including Prompt Engineering and Optimization, Supervised Fine-Tuning, Learning from Human Feedback, Evaluation, Self-Learning, etc. Your work will directly impact our customers in the form of novel products and services.
US, CA, San Diego
Do you want to join an innovative team of scientists who use machine learning and statistical techniques to help Amazon provide the best customer experience by preventing eCommerce fraud? Are you excited by the prospect of analyzing and modeling terabytes of data and creating state-of-the-art algorithms to solve real world problems? Do you like to own end-to-end business problems/metrics and directly impact the profitability of the company? Do you enjoy collaborating in a diverse team environment? If yes, then you may be a great fit to join the Amazon Buyer Risk Prevention (BRP) Machine Learning group. We are looking for a talented scientist who is passionate to build advanced algorithmic systems that help manage safety of millions of transactions every day. Key job responsibilities Use machine learning and statistical techniques to create scalable risk management systems Learning and understanding large amounts of Amazon’s historical business data for specific instances of risk or broader risk trends Design, development and evaluation of highly innovative models for risk management Working closely with software engineering teams to drive real-time model implementations and new feature creations Working closely with operations staff to optimize risk management operations, Establishing scalable, efficient, automated processes for large scale data analyses, model development, model validation and model implementation Tracking general business activity and providing clear, compelling management reporting on a regular basis Research and implement novel machine learning and statistical approaches
US, MA, Boston
The Artificial General Intelligence (AGI) team is looking for a passionate, talented, and inventive Applied Scientist with a strong deep learning background, to build industry-leading Generative Artificial Intelligence (GenAI) technology with Large Language Models (LLMs) and multimodal systems. Key job responsibilities As a Applied Scientist with the AGI team, you will work with talented peers to lead the development of novel algorithms and modeling techniques, to advance the state of the art with LLMs. Your work will directly impact our customers in the form of products and services that make use of speech and language technology. You will leverage Amazon’s heterogeneous data sources and large-scale computing resources to accelerate advances in spoken language understanding. About the team The AGI team has a mission to push the envelope in GenAI with LLMs and multimodal systems, in order to provide the best-possible experience for our customers.
US, WA, Seattle
The XCM (Cross Channel Cross-Category Marketing) team seeks an Applied Scientist to revolutionize our marketing strategies. XCM's mission is to build the most measurably effective, creatively impactful, and cross-channel campaigning capabilities possible, with the aim of growing "big-bet" programs, strengthening positive brand perceptions, and increasing long-term free cash flow. As a science team, we're tackling complex challenges in marketing incrementality measurement, optimization and audience segmentation. In this role, you'll collaborate with a diverse team of scientists and economists to build and enhance causal measurement, optimization and prediction models for Amazon's global multi-billion dollar fixed marketing budget. You'll also work closely with various teams to develop scientific roadmaps, drive innovation, and influence key resource allocation decisions. Key job responsibilities 1) Innovating scalable marketing methodologies using causal inference and machine learning. 2) Developing interpretable models that provide actionable business insights. 3) Collaborating with engineers to automate and scale scientific solutions. 4) Engaging with stakeholders to ensure effective adoption of scientific products. 5) Presenting findings to the Amazon Science community to promote excellence and knowledge-sharing.
US, WA, Seattle
Do you want to join an innovative team of scientists who use machine learning and statistical techniques to help Amazon provide the best customer experience by preventing eCommerce fraud? Are you excited by the prospect of analyzing and modeling terabytes of data and creating state-of-the-art algorithms to solve real world problems? Do you like to own end-to-end business problems/metrics and directly impact the profitability of the company? Do you enjoy collaborating in a diverse team environment? If yes, then you may be a great fit to join the Amazon Buyer Risk Prevention (BRP) Machine Learning group. We are looking for a talented scientist who is passionate to build advanced algorithmic systems that help manage safety of millions of transactions every day. Key job responsibilities Use machine learning and statistical techniques to create scalable risk management systems Learning and understanding large amounts of Amazon’s historical business data for specific instances of risk or broader risk trends Design, development and evaluation of highly innovative models for risk management Working closely with software engineering teams to drive real-time model implementations and new feature creations Working closely with operations staff to optimize risk management operations, Establishing scalable, efficient, automated processes for large scale data analyses, model development, model validation and model implementation Tracking general business activity and providing clear, compelling management reporting on a regular basis Research and implement novel machine learning and statistical approaches
US, WA, Seattle
The Global Cross-Channel and Cross- Category Marketing (XCM) org are seeking an experienced Economist to join our team. XCM’s mission is to be the most measurably effective and creatively breakthrough marketing organization in the world in order to strengthen the brand, grow the business, and reduce cost for Amazon overall. We achieve this through scaled campaigning in support of brands, categories, and audiences which aim to create the maximum incremental impact for Amazon as a whole by driving the Amazon flywheel. This is a high impact role with the opportunities to lead the development of state-of-the-art, scalable models to measure the efficacy and effectiveness of a new marketing channel. In this critical role, you will leverage your deep expertise in causal inference to design and implement robust measurement frameworks that provide actionable insights to drive strategic business decisions. Key Responsibilities: Develop advanced econometric and statistical models to rigorously evaluate the causal incremental impact of marketing campaigns on customer perception and customer behaviors. Collaborate cross-functionally with marketing, product, data science and engineering teams to define the measurement strategy and ensure alignment on objectives. Leverage large, complex datasets to uncover hidden patterns and trends, extracting meaningful insights that inform marketing optimization and investment decisions. Work with engineers, applied scientists and product managers to automate the model in production environment. Stay up-to-date with the latest research and methodological advancements in causal inference, causal ML and experiment design to continuously enhance the team's capabilities. Effectively communicate analysis findings, recommendations, and their business implications to key stakeholders, including senior leadership. Mentor and guide junior economists, fostering a culture of analytical excellence and innovation.
US, WA, Seattle
We are open to hiring candidates to work out of one of the following locations: Seattle, WA, USA Do you love using data to solve complex problems? Are you interested in innovating and developing world-class big data solutions? We have the career for you! EPP Analytics team is seeking an exceptional Data Scientist to recommend, design and deliver new advanced analytics and science innovations end-to-end partnering closely with our security/software engineers, and response investigators. Your work enables faster data-driven decision making for Preventive and Response teams by providing them with data management tools, actionable insights, and an easy-to-use reporting experience. The ideal candidate will be passionate about working with big data sets and have the expertise to utilize these data sets to derive insights, drive science roadmap and foster growth. Key job responsibilities - As a Data Scientist (DS) in EPP Analytics, you will do causal data science, build predictive models, conduct simulations, create visualizations, and influence data science practice across the organization. - Provide insights by analyzing historical data - Create experiments and prototype implementations of new learning algorithms and prediction techniques. - Research and build machine learning algorithms that improve Insider Threat risk A day in the life No two days are the same in Insider Risk teams - the nature of the work we do and constantly shifting threat landscape means sometimes you'll be working with an internal service team to find anomalous use of their data, other days you'll be working with IT teams to build improved controls. Some days you'll be busy writing detections, or mentoring or running design review meetings. The EPP Analytics team is made up of SDEs and Security Engineers who partner with Data Scientists to create big data solutions and continue to raise the bar for the EPP organization. As a member of the team you will have the opportunity to work on challenging data modeling solutions, new and innovative Quicksight based reporting, and data pipeline and process improvement projects. About the team Diverse Experiences Amazon Security values diverse experiences. Even if you do not meet all of the qualifications and skills listed in the job description, we encourage candidates to apply. If your career is just starting, hasn’t followed a traditional path, or includes alternative experiences, don’t let it stop you from applying. Why Amazon Security? At Amazon, security is central to maintaining customer trust and delivering delightful customer experiences. Our organization is responsible for creating and maintaining a high bar for security across all of Amazon’s products and services. We offer talented security professionals the chance to accelerate their careers with opportunities to build experience in a wide variety of areas including cloud, devices, retail, entertainment, healthcare, operations, and physical stores Inclusive Team Culture In Amazon Security, it’s in our nature to learn and be curious. Ongoing DEI events and learning experiences inspire us to continue learning and to embrace our uniqueness. Addressing the toughest security challenges requires that we seek out and celebrate a diversity of ideas, perspectives, and voices. Training & Career Growth We’re continuously raising our performance bar as we strive to become Earth’s Best Employer. That’s why you’ll find endless knowledge-sharing, training, and other career-advancing resources here to help you develop into a better-rounded professional. Work/Life Balance We value work-life harmony. Achieving success at work should never come at the expense of sacrifices at home, which is why flexible work hours and arrangements are part of our culture. When we feel supported in the workplace and at home, there’s nothing we can’t achieve.
IN, KA, Bengaluru
Do you want to join an innovative team of scientists who use machine learning and statistical techniques to create state-of-the-art solutions for providing better value to Amazon’s customers? Do you want to build and deploy advanced algorithmic systems that help optimize millions of transactions every day? Are you excited by the prospect of analyzing and modeling terabytes of data to solve real world problems? Do you like to own end-to-end business problems/metrics and directly impact the profitability of the company? Do you like to innovate and simplify? If yes, then you may be a great fit to join the Machine Learning and Data Sciences team for India Consumer Businesses. If you have an entrepreneurial spirit, know how to deliver, love to work with data, are deeply technical, highly innovative and long for the opportunity to build solutions to challenging problems that directly impact the company's bottom-line, we want to talk to you. Major responsibilities - Use machine learning and analytical techniques to create scalable solutions for business problems - Analyze and extract relevant information from large amounts of Amazon’s historical business data to help automate and optimize key processes - Design, development, evaluate and deploy innovative and highly scalable models for predictive learning - Research and implement novel machine learning and statistical approaches - Work closely with software engineering teams to drive real-time model implementations and new feature creations - Work closely with business owners and operations staff to optimize various business operations - Establish scalable, efficient, automated processes for large scale data analyses, model development, model validation and model implementation - Mentor other scientists and engineers in the use of ML techniques