ICLR: Why does deep learning work, and what are its limits?

Two recent trends in the theory of deep learning are examinations of the double-descent phenomenon and more-realistic approaches to neural kernel methods.

At this year’s International Conference on Learning Representations (ICLR), René Vidal, a professor of radiology and electrical engineering at the University of Pennsylvania and an Amazon Scholar, was a senior area chair, overseeing a team of reviewers charged with evaluating paper submissions to the conference. And the paper topic that his team focused on, Vidal says, was the theory of deep learning.

Vidal at ICLR.AS.16x9.png
René Vidal, the Rachleff University Professor at the University of Pennsylvania, with joint appointments in the School of Medicine's Department of Radiology and the Department of Electrical and Systems Engineering, a Penn Integrates Knowledge University Professor, and an Amazon Scholar.

“While representation learning and deep learning have been incredibly successful and have produced spectacular results for many application domains, deep networks remain black boxes,” Vidal explains. “How you design deep networks remains an art; there is a lot of trial and error on each and every dataset. So by and large, the area of mathematics of deep learning aims to have theorems, mathematical proofs, that guarantee the performance of deep networks.

“You can ask questions such as ‘Why is it the case that deep networks generalize from one data set to another?’ ‘Can you have a theorem that tells you the classification error on a new dataset versus the classification error on the training data set?’ ‘Can you derive a bound on that error as, say, a function of the number of training examples?’

“There are questions that pertain to optimization. These days, you are minimizing a loss function over, sometimes, billions of parameters. And because the optimization problems are so large, and you have so many training examples, for computational reasons, you are limited to very simple optimization methods. Can you prove convergence for these nonconvex problems? Can you understand what you converge to? Why is it the case that these very simple optimization methods are so successful for these very complex problems?’”

Double descent

In particular, Vidal says, two topics in the theory of deep learning have been drawing increased attention recently. The first is the so-called double-descent phenomenon. The conventional wisdom in AI used to hold that the size of a neural network had to be carefully tailored to both the problem it addressed and the amount of training data available. If the network was too small, it couldn’t learn complex patterns in the data; but if it got too large, it could simply memorize the correct answers for all the data in its training set — a particularly egregious case of overfitting — and it wouldn’t generalize to new inputs.

Related content
The surprising dynamics related to learning that are common to artificial and biological systems.

As a consequence, for a given problem and a given set of training data, as the size of a neural network grows, its error rate on the previously unseen data of the test set goes down. At some point, however, the error rate starts to go up again, as the network begins to overfit the data.

In the last few years, however, a number of papers have reported the surprising result that as the network continues to grow, the error rate goes back down again. This the double-descent phenomenon — and no one is sure why it happens.

“The error goes down as the size of the model grows, then back up as it overfits,” Vidal explains. “And it gets to a peak at the so-called interpolation limit, which is exactly when, during training, you can achieve zero error, because the network is big enough that it can memorize. But from then on, the testing error goes down again. There have been a lot of papers trying to explain why this happens.”

The neural tangent kernel

Another interesting recent trend in the theory of deep networks, Vidal says, involves new forms of analysis based on the neural tangent kernel.

Related content
Machine learning systems often act on “features” extracted from input data. In a natural-language-understanding system, for instance, the features might include words’ parts of speech, as assessed by an automatic syntactic parser, or whether a sentence is in the active or passive voice.

“In the past — say, the year 2000 — the way we did learning was by using so-called kernel methods,” Vidal explains. “Kernel methods are based on taking your data and embedding it with a fixed embedding into a very-high-dimensional space, where everything looks linear. We can use classical linear learning techniques in that embedding space, but the embedding space was fixed.

“You can think of deep learning as learning that embedding — mapping the input data to some high-dimensional space. In fact, that’s exactly representation learning. The neural-tangent-kernel regime — a type of initialization, a type of neural network, a type of training — is a regime under which you can approximate the learning dynamics of a deep network using kernels. And therefore you can use classical techniques to understand why they generalize and why not.

“That regime is very unrealistic — networks with infinite width or initializations that don't change the weights too much during training. In this very contrived and specialized setting, things are easier and we can understand them better. The current trend is how we go away from these unrealistic assumptions and acknowledge that the problem is hard: you do want weights to change during training, because if they don't, you're not learning much.”

Related content
Technique that mixes public and private training data can meet differential-privacy criteria while cutting error increase by 60%-70%.

Indeed, Vidal has engaged this topic himself, in a paper accepted to this year’s Conference on Artificial Intelligence and Statistics (AISTATS), whose coauthors are his old research team from Johns Hopkins University.

“The three assumptions we are trying to get rid of are, one, can we get theorems for networks with finite width as opposed to infinite width?” Vidal says. “Number two is, can we get theorems for gradient-descent-like methods that have a finite step size? Because many earlier theorems assumed a really teeny tiny step size — like, infinitesimally small. And the third assumption we are relaxing is this assumption on the initialization, which becomes much more general.”

The limits of representation learning

When ICLR was founded, in 2013, it was a venue for researchers to explore alternatives to machine learning methods, such as kernel methods, that represented data in fixed, prespecified ways. Now, however, deep learning — which uses learned representations — has taken over the field of machine learning, and the difference between ICLR and the other major machine learning conferences has shrunk.

As someone who spent 20 years as a professor of biomedical engineering at Hopkins, however, Vidal has a keen awareness of the limitations of representation learning. For some applications, he says, domain knowledge is still essential.

Related content
The first step in training a neural network to solve a problem is usually the selection of an architecture: a specification of the number of computational nodes in the network and the connections between them. Architectural decisions are generally based on historical precedent, intuition, and plenty of trial and error.

“It happens in domains where data or labels may not be abundant,” he explains. “This is the case, for example, in the medical domain, where maybe there are 100 patients in a study, or maybe you can't put the data on a website where everyone can annotate it.

“Just to give you one concrete example, I had a project where we needed to produce a blood test, and we needed to classify white blood cells into different kinds. No one is ever going to take videos of millions of cells, and you're not going to have a pathologist annotate each and every cell to do object detection the way we do in computer vision.

“So all we could get were the actual results of the blood test: what are the concentrations? And you might have a million cells of class one, class two, and class three, and you just have these very weak labels. But the domain experts said, we can do cell purification by adding these chemicals here and there, and we do centrifugation and I don't know what, and then we get cells of only one type in this specimen. Therefore you can now pretend that you have labels, because we know that cells that had different labels didn't survive this chemistry. And we said, ‘Wow, that’s great!’

“If you do things with 100% people who are all data scientists and machine learning people, they tend to think that all you need is a bigger network and more data. But I think, as at Amazon, where you need to think backwards from the customer, you need to solve real problems, and the solution isn't always more data and more annotations.”

Research areas

Related content

US, CA, Palo Alto
Amazon is the 4th most popular site in the US. Our product search engine, one of the most heavily used services in the world, indexes billions of products and serves hundreds of millions of customers world-wide. We are working on a new initiative to transform our search engine into a shopping engine that assists customers with their shopping missions. We look at all aspects of search CX, query understanding, Ranking, Indexing and ask how we can make big step improvements by applying advanced Machine Learning (ML) and Deep Learning (DL) techniques. We’re seeking a thought leader to direct science initiatives for the Search Relevance and Ranking at Amazon. This person will also be a deep learning practitioner/thinker and guide the research in these three areas. They’ll also have the ability to drive cutting edge, product oriented research and should have a notable publication record. This intellectual thought leader will help enhance the science in addition to developing the thinking of our team. This leader will direct and shape the science philosophy, planning and strategy for the team, as we explore multi-modal, multi lingual search through the use of deep learning . We’re seeking an individual that can enhance the science thinking of our team: The org is made of 60+ applied scientists, (2 Principal scientists and 5 Senior ASMs). This person will lead and shape the science philosophy, planning and strategy for the team, as we push into Deep Learning to solve problems like cold start, discovery and personalization in the Search domain. Joining this team, you’ll experience the benefits of working in a dynamic, entrepreneurial environment, while leveraging the resources of Amazon [Earth's most customer-centric internet company]. We provide a highly customer-centric, team-oriented environment in our offices located in Palo Alto, California.
JP, 13, Tokyo
Our mission is to help every vendor drive the most significant impact selling on Amazon. Our team invent, test and launch some of the most innovative services, technology, processes for our global vendors. Our new AVS Professional Services (ProServe) team will go deep with our largest and most sophisticated vendor customers, combining elite client-service skills with cutting edge applied science techniques, backed up by Amazon’s 20+ years of experience in Japan. We start from the customer’s problem and work backwards to apply distinctive results that “only Amazon” can deliver. Amazon is looking for a talented and passionate Applied Science Manager to manage our growing team of Applied Scientists and Business Intelligence Engineers to build world class statistical and machine learning models to be delivered directly to our largest vendors, and working closely with the vendors' senior leaders. The Applied Science Manager will set the strategy for the services to invent, collaborating with the AVS business consultants team to determine customer needs and translating them to a science and development roadmap, and finally coordinating its execution through the team. In this position, you will be part of a larger team touching all areas of science-based development in the vendor domain, not limited to Japan-only products, but collaborating with worldwide science and business leaders. Our current projects touch on the areas of causal inference, reinforcement learning, representation learning, anomaly detection, NLP and forecasting. As the AVS ProServe Applied Science Manager, you will be empowered to expand your scope of influence, and use ProServe as an incubator for products that can be expanded to all Amazon vendors, worldwide. We place strong emphasis on talent growth. As the Applied Science Manager, you will be expected in actively growing future Amazon science leaders, and providing mentoring inside and outside of your team. Key job responsibilities The Applied Science Manager is accountable for: (1) Creating a vision, a strategy, and a roadmap tackling the most challenging business questions from our leading vendors, assess quantitatively their feasibility and entitlement, and scale their scope beyond the ProServe team. (2) Coordinate execution of the roadmap, through direct reports, consisting of scientists and business intelligence engineers. (3) Grow and manage a technical team, actively mentoring, developing, and promoting team members. (4) Work closely with other science managers, program/product managers, and business leadership worldwide to scope new areas of growth, creating new partnerships, and proposing new business initiatives. (5) Act as a technical supervisor, able to assess scientific direction, technical design documents, and steer development efforts to maximize project delivery.
US, NY, New York
Amazon Advertising is one of Amazon's fastest growing and most profitable businesses. As a core product offering within our advertising portfolio, Sponsored Products (SP) helps merchants, retail vendors, and brand owners succeed via native advertising, which grows incremental sales of their products sold through Amazon. The SP team's primary goals are to help shoppers discover new products they love, be the most efficient way for advertisers to meet their business objectives, and build a sustainable business that continuously innovates on behalf of customers. Our products and solutions are strategically important to enable our Retail and Marketplace businesses to drive long-term growth. We deliver billions of ad impressions and millions of clicks and break fresh ground in product and technical innovations every day! The Supply team (within Sponsored Products) is looking for an Applied Scientist to join a fast-growing team with the mandate of creating new ad experiences that elevate the shopping experience for hundreds of millions customers worldwide. The Applied Scientist will take end-to-end ownership of driving new product/feature innovation by applying advanced statistical and machine learning models. The role will handle petabytes of unstructured data (images, text, videos) to extract insights into what metadata can be useful for us to highlight to simplify purchase decisions, and propose new experiences that increase shopper engagement. Why you love this opportunity Amazon is investing heavily in building a world-class advertising business. This team is responsible for defining and delivering a collection of advertising products that drive discovery and sales. Our solutions generate billions in revenue and drive long-term growth for Amazon’s Retail and Marketplace businesses. We deliver billions of ad impressions, millions of clicks daily, and break fresh ground to create world-class products. We are highly motivated, collaborative, and fun-loving team with an entrepreneurial spirit - with a broad mandate to experiment and innovate. Key job responsibilities As an Applied Scientist on this team you will: Build machine learning models and perform data analysis to deliver scalable solutions to business problems. Perform hands-on analysis and modeling with very large data sets to develop insights that increase traffic monetization and merchandise sales without compromising shopper experience. Work closely with software engineers on detailed requirements, technical designs and implementation of end-to-end solutions in production. Design and run A/B experiments that affect hundreds of millions of customers, evaluate the impact of your optimizations and communicate your results to various business stakeholders. Work with scientists and economists to model the interaction between organic sales and sponsored content and to further evolve Amazon's marketplace. Establish scalable, efficient, automated processes for large-scale data analysis, machine-learning model development, model validation and serving. Research new predictive learning approaches for the sponsored products business. Write production code to bring models into production. A day in the life You will invent new experiences and influence customer-facing shopping experiences to help suppliers grow their retail business and the auction dynamics that leverage native advertising; this is your opportunity to work within the fastest-growing businesses across all of Amazon! Define a long-term science vision for our advertising business, driven fundamentally from our customers' needs, translating that direction into specific plans for research and applied scientists, as well as engineering and product teams. This role combines science leadership, organizational ability, technical strength, product focus, and business understanding.
US, CA, Palo Alto
The Amazon Search team creates powerful, customer-focused search solutions and technologies. Whenever a customer visits an Amazon site worldwide and types in a query or browses through product categories, Amazon Search services go to work. We design, develop, and deploy high performance, fault-tolerant distributed search systems used by millions of Amazon customers every day. We are seeking a Principal Scientist with deep expertise in Search. Your responsibility will be to advance the state-of-the-art for search science that leads to world-class products that impact Amazon's customers. Key job responsibilities You will be responsible for defining key research directions, adopting or inventing new machine learning techniques, conducting rigorous experiments, publishing results, and ensuring that research is translated into practice. You will develop long-term strategies, persuade teams to adopt those strategies, propose goals and deliver on them. You will also participate in organizational planning, hiring, mentorship and leadership development. You will be technically fearless and with a passion for building scalable science and engineering solutions. You will serve as a key scientific resource in full-cycle development (conception, design, implementation, testing to documentation, delivery, and maintenance). About the team This is a position on Core Ranking and Experimentation team in Palo Alto, CA. The team works on a variety of topics in search ranking and relevance, such as multi-objective optimization, personalization, and fast online experimentation. We work closely with teams in various parts of the stack to ensure that our science is translated to customer facing products.
US, WA, Bellevue
Amazon is looking for a passionate, talented, and inventive Applied Scientists with a strong machine learning background to help build industry-leading Speech and Language technology. Our mission is to provide a delightful experience to Amazon’s customers by pushing the envelope in Automatic Speech Recognition (ASR), Machine Translation (MT), Natural Language Understanding (NLU), Machine Learning (ML) and Computer Vision (CV). As part of our AI team in Amazon AWS, you will work alongside internationally recognized experts to develop novel algorithms and modeling techniques to advance the state-of-the-art in human language technology. Your work will directly impact millions of our customers in the form of products and services that make use of speech and language technology. You will gain hands on experience with Amazon’s heterogeneous speech, text, and structured data sources, and large-scale computing resources to accelerate advances in spoken language understanding. We are hiring in all areas of human language technology: ASR, MT, NLU, text-to-speech (TTS), and Dialog Management, in addition to Computer Vision.
IN, KA, Bangalore
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. The ATT team, based in Bangalore, is responsible for ensuring that ads are compliant to world-wide advertising policies and are of high quality, leading to higher conversion for the advertisers and providing a great experience for the shoppers. Machine learning, particularly multi-modal data understanding, is fundamental to the way we drive our business, meet our goals and satisfy our customers. ATT team invests in researching and developing state of art models that analyze various type of ad assets – text, audio, images and videos - to ensure compliance to advertising policies. We also help advertisers create more successful ads by creating ML models to assist ad generation as well as to provide data-driven interpretable insights. Key job responsibilities Major responsibilities · Deliver key goals to enhance advertiser experience and protect shopper trust by innovative use of computer vision, NLP and statistical techniques · Drive core business analytics and data science explorations to inform key business decisions and algorithm roadmap · Establish scalable, efficient, automated processes for large scale data analyses, model development, model validation and model implementation · Hire and develop top talent in machine learning and data science and accelerate the pace of innovation in the group · Work proactively with engineering teams and product managers to evangelize new algorithms and drive the implementation of large-scale complex ML models in production
US, WA, Seattle
We are seeking a talented applied researcher to join the Search team responsible for developing reinforcement learning systems for Amazon's shopping experience and delivering it to millions of customers. We believe that shopping on Amazon should be simple, delightful, and full of "wow" moments for everyone.
US, NY, New York
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 Python is necessary, and experience with SQL and UNIX 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, scientists, and product managers. 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. About the team Amazon's Weblab team enables experimentation at massive scale to help Amazon build better products for customers. A/B testing is in Amazon's DNA and we're at the core of how Amazon innovates on behalf of customers.
US, WA, Seattle
Amazon Advertising is one of Amazon's fastest growing and most profitable businesses, responsible for defining and delivering a collection of advertising products that drive discovery and sales. As a core product offering within our advertising portfolio, Sponsored Products (SP) helps merchants, retail vendors, and brand owners succeed via native advertising, which grows incremental sales of their products sold through Amazon. The SP team's primary goals are to help shoppers discover new products they love, be the most efficient way for advertisers to meet their business objectives, and build a sustainable business that continuously innovates on behalf of customers. Our products and solutions are strategically important to enable our Retail and Marketplace businesses to drive long-term growth. We deliver billions of ad impressions and millions of clicks and break fresh ground in product and technical innovations every day! As an Applied Science Manager in Machine Learning, you will: Directly manage and lead a cross-functional team of Applied Scientists, Data Scientists, Economists, and Business Intelligence Engineers. Develop and manage a research agenda that balances short term deliverables with measurable business impact as well as long term investments. Lead marketplace design and development based on economic theory and data analysis. Provide technical and scientific guidance to team members. Rapidly design, prototype and test many possible hypotheses in a high-ambiguity environment, making use of both quantitative and business judgment Advance the team's engineering craftsmanship and drive continued scientific innovation as a thought leader and practitioner. Develop science and engineering roadmaps, run annual planning, and foster cross-team collaboration to execute complex projects. Perform hands-on data analysis, build machine-learning models, run regular A/B tests, and communicate the impact to senior management. Collaborate with business and software teams across Amazon Ads. Stay up to date with recent scientific publications relevant to the team. Hire and develop top talent, provide technical and career development guidance to scientists and engineers within and across the organization. Why you will love this opportunity: Amazon is investing heavily in building a world-class advertising business. This team defines and delivers a collection of advertising products that drive discovery and sales. Our solutions generate billions in revenue and drive long-term growth for Amazon’s Retail and Marketplace businesses. We deliver billions of ad impressions, millions of clicks daily, and break fresh ground to create world-class products. We are a highly motivated, collaborative, and fun-loving team with an entrepreneurial spirit - with a broad mandate to experiment and innovate. Impact and Career Growth: You will invent new experiences and influence customer-facing shopping experiences to help suppliers grow their retail business and the auction dynamics that leverage native advertising; this is your opportunity to work within the fastest-growing businesses across all of Amazon! Define a long-term science vision for our advertising business, driven from our customers' needs, translating that direction into specific plans for research and applied scientists, as well as engineering and product teams. This role combines science leadership, organizational ability, technical strength, product focus, and business understanding. Team video ~ https://youtu.be/zD_6Lzw8raE
US, CA, Palo Alto
The Amazon Search team creates powerful, customer-focused search and advertising solutions and technologies. Whenever a customer visits an Amazon site worldwide and types in a query or browses through product categories, Amazon Search services go to work. We design, develop, and deploy high performance, fault-tolerant distributed search systems used by millions of Amazon customers every day. Our Search Relevance team works to maximize the quality and effectiveness of the search experience for visitors to Amazon websites worldwide. Amazon’s large scale brings with it unique problems to solve in designing, testing, and deploying relevance models. We are seeking a strong applied Scientist to join the Experimentation Infrastructure and Methods team. This team’s charter is to innovate and evaluate ranking at Amazon Search. In practice, we aim to create infrastructure and metrics, enable new experimental methods, and do proof-of-concept experiments, that enable Search Relevance teams to introduce new features faster, reduce the cost of experimentation, and deliver faster against Search goals. Key job responsibilities You will build search ranking systems and evaluation framework that extend to Amazon scale -- thousands of product types, billions of queries, and hundreds of millions of customers spread around the world. As a Senior Applied Scientist you will find the next set of big improvements to ranking evaluation, get your hands dirty by building models to help understand complexities of customer behavior, and mentor junior engineers and scientists. In addition to typical topics in ranking, we are particularly interested in evaluation, feature selection, explainability. A day in the life Our primary focus is improving search ranking systems. On a day-to-day this means building ML models, analyzing data from your recent A/B tests, and guiding teams on best practices. You will also find yourself in meetings with business and tech leaders at Amazon communicating your next big initiative. About the team We are a team consisting of software engineers and applied scientists. Our interests and activities span machine learning for better ranking, experimentation, statistics for better decision making, and infrastructure to make it all happen efficiently at scale.