When does transfer learning work?

New transferability metric is more accurate and more generally applicable than predecessors.

Transfer learning is a widely used technique for improving the performance of neural networks when labeled training data is scarce. Before a network is trained on a target task with limited data, it might be pretrained on a source task with more abundant data. The knowledge it gains from pretraining would then be transferred to the target task.

When is transfer learning effective, and when is it not? And if you’re going to do transfer learning, what task should you use for pretraining? Answering these and related questions is the aim of transferability metrics, which measure the applicability of a pretrained network to new tasks.

At this year’s International Conference on Machine Learning, my colleagues and I are presenting a new transferability metric, which in our experiments proved to be a better predictor of transferability than existing metrics.

It’s also more general than existing metrics: it’s theoretically grounded and does not make strong assumptions about the relatedness of source and target tasks, as previous metrics often do.

In the paper, we offer theoretical analyses indicating that our metric — which we call LEEP, for log expected empirical prediction — should offer a good estimate of transferability.

We also conducted a battery of empirical tests, comparing our metric to its two leading predecessors in 23 different transfer settings, which combined three types of transfer learning, two image recognition source tasks, and target tasks that involved different subsets of the training data for two further image recognition tasks.

On 15 of those 23 comparisons, LEEP’s predictions correlated better than those of the other two metrics with the ultimate accuracy of the transferred model. In some cases, the difference was dramatic — an improvement of as much as 30% over the second-best metric.

As can be seen from the graphs below, which show our results for two different source data sets, LEEP correlates very well with the ultimate accuracy of the transferred model.

Leep correlations high-res.png
Correlations between our new transferability metric (LEEP) and the accuracy of transferred models, using two different source data sets and two transfer methods. In one transfer method (fine-tune), the whole network is fine-tuned on data from a new task; in the other (retrain head), only part of the network (the head, or classifier) is retrained.

One of the settings we considered was that of meta-transfer learning, which is a combination of transfer learning and meta-learning. With meta-transfer learning, a single deep-learning model is adapted to many new tasks during training, so that it learns to adapt to unseen tasks with very little data.

On that task, LEEP’s correlation with ultimate model accuracy was .591, while the other methods’ correlations were .310 and .025. We believe that LEEP is the first transferability metric that applies to meta-transfer learning.

How do we compute LEEP?

LEEP measures the transferability between a trained machine learning model and the labeled data set for a new task. Suppose, for instance, that the model has been trained to recognize images of land animals, and the target task is to recognize sea animals.

The first step in the LEEP process is to use the trained model to classify the data in the training set for the target task. The resulting classifications are dummy labels: the land animal classifier might, for instance, recognize seals as dogs and seahorses as giraffes. But that doesn’t matter for purposes of transferability prediction.

Conditional probability.png
LEEP first applies a classifier trained on a source task (e.g., identifying land animals) to the data set for a target task (e.g., identifying sea animals). From the resulting classifications, it calculates the probability of a particular target task classification (e.g., seal) given a particular source task classification (e.g., dog, horse, giraffe).
credit: Stacy Reilly

For every example in the training set, we have the correct label, so the next step is to compute the conditional probabilities of target-task labels given the dummy classifications of the trained model — the probability that an example classified as a dog is actually a seal, a seahorse, a shark, and so on.

Finally, we imagine a classifier that will, according to that conditional distribution, randomly pick a dummy label for each input and, on the basis of that dummy label, randomly pick a target-task label.

LEEP is the average log-likelihood of this hypothetical classifier on the data in the target-task training set. Log-likelihood is a standard measure of how well a statistical model fits a particular collection of data samples, so LEEP tells us how well the hypothetical classifier fits the target-task training set. That makes it an easy-to-interpret as well as accurate metric.

Potential applications of LEEP

We believe that LEEP has several potential applications.

First, it can be used to select source models for transfer learning. It could also help select groups of highly transferable tasks for multitask learning, in which a single machine learning model is trained on several tasks at once.

LEEP could also be useful for the transfer and optimization of hyperparameters, which are structural features of machine learning models or parameters of learning algorithms that help maximize the efficacy of training on particular data.

Research areas

Related content

US, WA, Redmond
Amazon Leo is Amazon’s low Earth orbit satellite network. Our mission is to deliver fast, reliable internet connectivity to customers beyond the reach of existing networks. From individual households to schools, hospitals, businesses, and government agencies, Amazon Leo will serve people and organizations operating in locations without reliable connectivity. Export Control Requirement: Due to applicable export control laws and regulations, candidates must be a U.S. citizen or national, U.S. permanent resident (i.e., current Green Card holder), or lawfully admitted into the U.S. as a refugee or granted asylum. This position is part of the Satellite Attitude Determination and Control team. You will design and analyze the control system and algorithms, support development of our flight hardware and software, help integrate the satellite in our labs, participate in flight operations, and see a constellation of satellites flow through the production line in the building next door. Key job responsibilities - Design and analyze algorithms for estimation, flight control, and precise pointing using linear methods and simulation. - Develop and apply models and simulations, with various levels of fidelity, of the satellite and our constellation. - Component level environmental testing, functional and performance checkout, subsystem integration, satellite integration, and in space operations. - Manage the spacecraft constellation as it grows and evolves. - Continuously improve our ability to serve customers by maximizing payload operations time. - Develop autonomy for Fault Detection and Isolation on board the spacecraft. A day in the life This is an opportunity to play a significant role in the design of an entirely new satellite system with challenging performance requirements. The large, integrated constellation brings opportunities for advanced capabilities that need investigation and development. The constellation size also puts emphasis on engineering excellence so our tools and methods, from conceptualization through manufacturing and all phases of test, will be state of the art as will the satellite and supporting infrastructure on the ground. You will find that Amazon Leo's mission is compelling, so our program is staffed with some of the top engineers in the industry. Our daily collaboration with other teams on the program brings constant opportunity for discovery, learning, and growth. About the team Our team has lots of experience with various satellite systems and many other flight vehicles. We have bench strength in both our mission and core GNC disciplines. We design, prototype, test, iterate and learn together. Because GNC is central to safe flight, we tend to drive Concepts of Operation and many system level analyses.
US, CA, San Francisco
If you are interested in this position, please apply on Twitch's Career site https://www.twitch.tv/jobs/en/ About Us: Twitch is the world’s biggest live streaming service, with global communities built around gaming, entertainment, music, sports, cooking, and more. It is where thousands of communities come together for whatever, every day. We’re about community, inside and out. You’ll find coworkers who are eager to team up, collaborate, and smash (or elegantly solve) problems together. We’re on a quest to empower live communities, so if this sounds good to you, see what we’re up to on LinkedIn and X, and discover the projects we’re solving on our Blog. Be sure to explore our Interviewing Guide to learn how to ace our interview process. About the Role We are looking for applied scientists to solve challenging and open-ended problems in the domain of user and content safety. As an applied scientist on Twitch's Community team, you will use machine learning to develop data products tackling problems such as harassment, spam, and illegal content. You will use a wide toolbox of ML tools to handle multiple types of data, including user behavior, metadata, and user generated content such as text and video. You will collaborate with a team of passionate scientists and engineers to develop these models and put them into production, where they can help Twitch's creators and viewers succeed and build communities. You will report to our Senior Applied Science Manager in San Francisco, CA. You can work from San Francisco, CA or Seattle, WA. You Will - Build machine learning products to protect Twitch and its users from abusive behavior such as harassment, spam, and violent or illegal content. - Work backwards from customer problems to develop the right solution for the job, whether a classical ML model or a state-of-the-art one. - Collaborate with Community Health's engineering and product management team to productionize your models into flexible data pipelines and ML-based services. - Continue to learn and experiment with new techniques in ML, software engineering, or safety so that we can better help communities on Twitch grow and stay safe. Perks * Medical, Dental, Vision & Disability Insurance * 401(k) * Maternity & Parental Leave * Flexible PTO * Amazon Employee Discount
US, WA, Redmond
As a Guidance, Navigation & Control Hardware Engineer, you will directly contribute to the planning, selection, development, and acceptance of Guidance, Navigation & Control hardware for Amazon Leo's constellation of satellites. Specializing in critical satellite hardware components including reaction wheels, star trackers, magnetometers, sun sensors, and other spacecraft sensors and actuators, you will play a crucial role in the integration and support of these precision systems. You will work closely with internal Amazon Leo hardware teams who develop these components, as well as Guidance, Navigation & Control engineers, software teams, systems engineering, configuration & data management, and Assembly, Integration & Test teams. A key aspect of your role will be actively resolving hardware issues discovered during both factory testing phases and operational space missions, working hand-in-hand with internal Amazon Leo hardware development teams to implement solutions and ensure optimal satellite performance. Export Control Requirement: Due to applicable export control laws and regulations, candidates must be a U.S. citizen or national, U.S. permanent resident (i.e., current Green Card holder), or lawfully admitted into the U.S. as a refugee or granted asylum. Key job responsibilities * Planning and coordination of resources necessary to successfully accept and integrate satellite Guidance, Navigation & Control components including reaction wheels, star trackers, magnetometers, and sun sensors provided by internal Amazon Leo teams * Partner with internal Amazon Leo hardware teams to develop and refine spacecraft actuator and sensor solutions, ensuring they meet requirements and providing technical guidance for future satellite designs * Collaborate with internal Amazon Leo hardware development teams to resolve issues discovered during both factory test phases and operational space missions, implementing corrective actions and design improvements * Work with internal Amazon Leo teams to ensure state-of-the-art satellite hardware technologies including precision pointing systems, attitude determination sensors, and spacecraft actuators meet mission requirements * Lead verification and testing activities, ensuring satellite Guidance, Navigation & Control hardware components meet stringent space-qualified requirements * Drive implementation of hardware-in-the-loop testing for satellite systems, coordinating with internal Amazon Leo hardware engineers to validate component performance in simulated space environments * Troubleshoot and resolve complex hardware integration issues working directly with internal Amazon Leo hardware development teams
US, CA, San Francisco
Are you interested in a unique opportunity to advance the accuracy and efficiency of Artificial General Intelligence (AGI) systems? If so, you're at the right place! We are the AGI Autonomy organization, and we are looking for a driven and talented Member of Technical Staff to join us to build state-of-the art agents. As an MTS on our team, you will design, build, and maintain a Spark-based infrastructure to process and manage large datasets critical for machine learning research. You’ll work closely with our researchers to develop data workflows and tools that streamline the preparation and analysis of massive multimodal datasets, ensuring efficiency and scalability. We operate at Amazon's large scale with the energy of a nimble start-up. If you have a learner's mindset, enjoy solving challenging problems and value an inclusive and collaborative team culture, you will thrive in this role, and we hope to hear from you. Key job responsibilities * Develop and maintain reliable infrastructure to enable large-scale data extraction and transformation. * Work closely with researchers to create tooling for emerging data-related needs. * Manage project prioritization, deliverables, timelines, and stakeholder communication. * Illuminate trade-offs, educate the team on best practices, and influence technical strategy. * Operate in a dynamic environment to deliver high quality software.
IN, KA, Bangalore
Have you ever ordered a product on Amazon and when that box with the smile arrived you wondered how it got to you so fast? Have you wondered where it came from and how much it cost Amazon to deliver it to you? If so, the WW Amazon Logistics, Business Analytics team is for you. We manage the delivery of tens of millions of products every week to Amazon’s customers, achieving on-time delivery in a cost-effective manner. We are looking for an enthusiastic, customer obsessed, Applied Scientist with good analytical skills to help manage projects and operations, implement scheduling solutions, improve metrics, and develop scalable processes and tools. The primary role of an Operations Research Scientist within Amazon is to address business challenges through building a compelling case, and using data to influence change across the organization. This individual will be given responsibility on their first day to own those business challenges and the autonomy to think strategically and make data driven decisions. Decisions and tools made in this role will have significant impact to the customer experience, as it will have a major impact on how the final phase of delivery is done at Amazon. Candidates will be a high potential, strategic and analytic graduate with a PhD in (Operations Research, Statistics, Engineering, and Supply Chain) ready for challenging opportunities in the core of our world class operations space. Great candidates have a history of operations research, and the ability to use data and research to make changes. This role requires robust program management skills and research science skills in order to act on research outcomes. This individual will need to be able to work with a team, but also be comfortable making decisions independently, in what is often times an ambiguous environment. Responsibilities may include: - Develop input and assumptions based preexisting models to estimate the costs and savings opportunities associated with varying levels of network growth and operations - Creating metrics to measure business performance, identify root causes and trends, and prescribe action plans - Managing multiple projects simultaneously - Working with technology teams and product managers to develop new tools and systems to support the growth of the business - Communicating with and supporting various internal stakeholders and external audiences
US, NY, New York
Amazon is investing heavily in building a world class advertising business and we are responsible for defining and delivering a collection of self-service performance advertising products that drive discovery and sales. Our products are strategically important to our Retail and Marketplace businesses driving long term growth. We deliver billions of ad impressions and millions of clicks daily and are breaking fresh ground to create world-class products. We are highly motivated, collaborative and fun-loving with an entrepreneurial spirit and bias for action. With a broad mandate to experiment and innovate, we are growing at an unprecedented rate with a seemingly endless range of new opportunities. The Ad Response Prediction team in the Sponsored Products organization builds GenAI-based shopper understanding and audience targeting systems, along with advanced deep-learning models for Click-through Rate (CTR) and Conversion Rate (CVR) predictions. We develop large-scale machine-learning (ML) pipelines and real-time serving infrastructure to match shoppers' intent with relevant ads across all devices, contexts, and marketplaces. Through precise estimation of shoppers' interactions with ads and their long-term value, we aim to drive optimal ad allocation and pricing, helping to deliver a relevant, engaging, and delightful advertising experience to Amazon shoppers. As our business grows and we undertake increasingly complex initiatives, we are looking for entrepreneurial, and self-driven science leaders to join our team. Key job responsibilities As a Principal Applied Scientist in the team, you will: * Seek to understand in depth the Sponsored Products offering at Amazon and identify areas of opportunities to grow our business via principled ML solutions. * Mentor and guide the applied scientists in our organization and hold us to a high standard of technical rigor and excellence in ML. * Design and lead organization wide ML roadmaps to help our Amazon shoppers have a delightful shopping experience while creating long term value for our sellers. * Work with our engineering partners and draw upon your experience to meet latency and other system constraints. * Identify untapped, high-risk technical and scientific directions, and simulate new research directions that you will drive to completion and deliver. * Be responsible for communicating our ML innovations to the broader internal & external scientific community.
US, WA, Seattle
PXTCS is looking for an economist who can apply economic methods to address business problems. The ideal candidate will work with engineers and computer scientists to estimate models and algorithms on large scale data, design pilots and measure impact, and transform successful prototypes into improved policies and programs at scale. PXTCS is looking for creative thinkers who can combine a strong technical economic toolbox with a desire to learn from other disciplines, and who know how to execute and deliver on big ideas as part of an interdisciplinary technical team. Ideal candidates will work in a team setting with individuals from diverse disciplines and backgrounds. They will work with teammates to develop scientific models and conduct the data analysis, modeling, and experimentation that is necessary for estimating and validating models. They will work closely with engineering teams to develop scalable data resources to support rapid insights, and take successful models and findings into production as new products and services. They will be customer-centric and will communicate scientific approaches and findings to business leaders, listening to and incorporate their feedback, and delivering successful scientific solutions. A day in the life The Economist will work with teammates to apply economic methods to business problems. This might include identifying the appropriate research questions, writing code to implement a DID analysis or estimate a structural model, or writing and presenting a document with findings to business leaders. Our economists also collaborate with partner teams throughout the process, from understanding their challenges, to developing a research agenda that will address those challenges, to help them implement solutions. About the team The People eXperience and Technology Central Science (PXTCS) team 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. PXTCS is an interdisciplinary team that combines the talents of science and engineering to develop and deliver solutions that measurably achieve this goal.
US, CA, San Francisco
The Amazon General Intelligence “AGI” organization is looking for an Executive Assistant to support leaders of our Autonomy Team in our growing AI Lab space located in San Francisco. This role is ideal for exceptionally talented, dependable, customer-obsessed, and self-motivated individuals eager to work in a fast paced, exciting and growing team. This role serves as a strategic business partner, managing complex executive operations across the AGI organization. The position requires superior attention to detail, ability to meet tight deadlines, excellent organizational skills, and juggling multiple critical requests while proactively anticipating needs and driving improvements. High integrity, discretion with confidential information, and professionalism are essential. The successful candidate will complete complex tasks and projects quickly with minimal guidance, react with appropriate urgency, and take effective action while navigating ambiguity. Flexibility to change direction at a moment's notice is critical for success in this role. Key job responsibilities - Serve as strategic partner to senior leadership, identifying opportunities to improve organizational effectiveness and drive operational excellence - Manage complex calendars and scheduling for multiple executives - Drive continuous improvement through process optimization and new mechanisms - Coordinate team activities including staff meetings, offsites, and events - Schedule and manage cost-effective travel - Attend key meetings, track deliverables, and ensure timely follow-up - Create expense reports and manage budget tracking - Serve as liaison between executives and internal/external stakeholders - Build collaborative relationships with Executive Assistants across the company and with critical external partners - Help us build a great team culture in the SF Lab!
US, CA, San Francisco
Join the next revolution in robotics at Amazon's Frontier AI & Robotics team, where you'll work alongside world-renowned AI pioneers to push the boundaries of what's possible in robotic intelligence. As an Applied Scientist, you'll be at the forefront of developing breakthrough foundation models that enable robots to perceive, understand, and interact with the world in unprecedented ways. You'll drive independent research initiatives in areas such as perception, manipulation, science understanding, locomotion, manipulation, sim2real transfer, multi-modal foundation models and multi-task robot learning, designing novel frameworks that bridge the gap between state-of-the-art research and real-world deployment at Amazon scale. In this role, you'll balance innovative technical exploration with practical implementation, collaborating with platform teams to ensure your models and algorithms perform robustly in dynamic real-world environments. You'll have access to Amazon's vast computational resources, enabling you to tackle ambitious problems in areas like very large multi-modal robotic foundation models and efficient, promptable model architectures that can scale across diverse robotic applications. Key job responsibilities - Drive independent research initiatives across the robotics stack, including robotics foundation models, focusing on breakthrough approaches in perception, and manipulation, for example open-vocabulary panoptic scene understanding, scaling up multi-modal LLMs, sim2real/real2sim techniques, end-to-end vision-language-action models, efficient model inference, video tokenization - Design and implement novel deep learning architectures that push the boundaries of what robots can understand and accomplish - Lead full-stack robotics projects from conceptualization through deployment, taking a system-level approach that integrates hardware considerations with algorithmic development, ensuring robust performance in production environments - Collaborate with platform and hardware teams to ensure seamless integration across the entire robotics stack, optimizing and scaling models for real-world applications - Contribute to the team's technical strategy and help shape our approach to next-generation robotics challenges A day in the life - Design and implement novel foundation model architectures and innovative systems and algorithms, leveraging our extensive infrastructure to prototype and evaluate at scale - Collaborate with our world-class research team to solve complex technical challenges - Lead technical initiatives from conception to deployment, working closely with robotics engineers to integrate your solutions into production systems - Participate in technical discussions and brainstorming sessions with team leaders and fellow scientists - Leverage our massive compute cluster and extensive robotics infrastructure to rapidly prototype and validate new ideas - Transform theoretical insights into practical solutions that can handle the complexities of real-world robotics applications About the team At Frontier AI & Robotics, we're not just advancing robotics – we're reimagining it from the ground up. Our team is building the future of intelligent robotics through innovative foundation models and end-to-end learned systems. We tackle some of the most challenging problems in AI and robotics, from developing sophisticated perception systems to creating adaptive manipulation strategies that work in complex, real-world scenarios. What sets us apart is our unique combination of ambitious research vision and practical impact. We leverage Amazon's massive computational infrastructure and rich real-world datasets to train and deploy state-of-the-art foundation models. Our work spans the full spectrum of robotics intelligence – from multimodal perception using images, videos, and sensor data, to sophisticated manipulation strategies that can handle diverse real-world scenarios. We're building systems that don't just work in the lab, but scale to meet the demands of Amazon's global operations. Join us if you're excited about pushing the boundaries of what's possible in robotics, working with world-class researchers, and seeing your innovations deployed at unprecedented scale.
US, CA, San Francisco
Join the next revolution in robotics at Amazon's Frontier AI & Robotics team, where you'll work alongside world-renowned AI pioneers to push the boundaries of what's possible in robotic intelligence. As a Senior Applied Scientist, you'll spearhead the development of breakthrough foundation models and full-stack robotics systems that enable robots to perceive, understand, and interact with the world in unprecedented ways. You'll drive technical excellence in areas such as perception, manipulation, science understanding, locomotion, manipulation, sim2real transfer, multi-modal foundation models and multi-task robot learning, designing novel frameworks that bridge the gap between state-of-the-art research and real-world deployment at Amazon scale. In this role, you'll combine hands-on technical work with scientific leadership, ensuring your team delivers robust solutions for dynamic real-world environments. You'll leverage Amazon's vast computational resources to tackle ambitious problems in areas like very large multi-modal robotic foundation models and efficient, promptable model architectures that can scale across diverse robotic applications. Key job responsibilities - Lead technical initiatives across the robotics stack, driving breakthrough approaches through hands-on research and development in areas including robotics foundation models, focusing on breakthrough approaches in perception, and manipulation, for example open-vocabulary panoptic scene understanding, scaling up multi-modal LLMs, sim2real/real2sim techniques, end-to-end vision-language-action models, efficient model inference, video tokenization - Guide technical direction for full-stack robotics projects from conceptualization through deployment, taking a system-level approach that integrates hardware considerations with algorithmic development, ensuring robust performance in production environments - Mentor fellow scientists while maintaining strong individual technical contributions - Collaborate with platform and hardware teams to ensure seamless integration across the entire robotics stack - Influence technical decisions and implementation strategies within your area of focus A day in the life - Design and implement novel foundation model architectures and innovative systems and algorithms, leveraging our extensive infrastructure to prototype and evaluate at scale - Guide fellow scientists in solving complex technical challenges across the full robotics stack - Lead focused technical initiatives from conception through deployment, ensuring successful integration with production systems - Drive technical discussions within your team and with key stakeholders - Conduct experiments and prototype new ideas using our massive compute cluster and extensive robotics infrastructure - Mentor team members while maintaining significant hands-on contribution to technical solutions About the team At Frontier AI & Robotics, we're not just advancing robotics – we're reimagining it from the ground up. Our team is building the future of intelligent robotics through innovative foundation models and end-to-end learned systems. We tackle some of the most challenging problems in AI and robotics, from developing sophisticated perception systems to creating adaptive manipulation strategies that work in complex, real-world scenarios. What sets us apart is our unique combination of ambitious research vision and practical impact. We leverage Amazon's massive computational infrastructure and rich real-world datasets to train and deploy state-of-the-art foundation models. Our work spans the full spectrum of robotics intelligence – from multimodal perception using images, videos, and sensor data, to sophisticated manipulation strategies that can handle diverse real-world scenarios. We're building systems that don't just work in the lab, but scale to meet the demands of Amazon's global operations. Join us if you're excited about pushing the boundaries of what's possible in robotics, working with world-class researchers, and seeing your innovations deployed at unprecedented scale.