Hackensack Meridian Health (HMH) executive Mark Eimer explains how an ambitiously-timed rollout of a comprehensive suite of Google products helped the entire organization—from doctors to IT staff—achieve better security, cultivate a more equitable work environment, and ultimately, improve patient outcomes.

How does a recently merged, 17-hospital healthcare system fast-track a platform migration and hardware rollout securely and in a way that improves work for everyone, regardless of location or role? These are the questions that kept me up at night in early 2020, when the pandemic demanded a “big bang”—something our legacy laptops and operating systems couldn’t handle.

We began our work with Google in 2020 with the adoption of Chrome as our default browser. As we migrated platforms, keeping patient data safe was of the utmost importance to us, along with providing every staff member with the tools they needed to work virtually. Our staff often experienced issues accessing our web-based applications using Internet Explorer or Edge Browser, a problem that went away when we switched to Chrome. Chrome’s versatile compatibility also made it easier for my team to migrate all of our web-based operations, and Chrome’s security and manageability were key components to making this switch a huge win for the organization.

The success of this migration led us to extend our Google partnership to patient care applications—where Google’s expertise in AI and ML helps scale the use of diagnostics tools and improve other aspects of the patient journey.

Achieving security at every step 

Like so many other healthcare organizations, we’ve been concerned about ransomware attacks. This is part of why we moved to Google Workspace and distributed over 3,000 Chrome OS devices in kiosk mode in March of 2020, when many of us went remote due to the pandemic. We were very concerned about team members accessing corporate applications through home devices that were running EOL operating systems (WIN7), as well as a general lack of antivirus and encryption measures.

We were protected by the fact that Google’s software and hardware both had built-in security features that we needed to stave off sophisticated attackers. For example, Chrome OS automatically updates to the latest security update and encrypts data living outside the cloud on the hardware. These features protected us from security-related disruptions, letting us securely move a huge library of file shares and emails across thousands of accounts to Google Chrome OS in just four months.

A year later, in March 2021, we migrated the enterprise over to Google Workspace and saw an immediate reduction in spam by 30% from the inherent built-in AI/ML. This meant staff were less likely to receive (and click through) phishing attempts. My team could connect, create, and collaborate easily and securely—even as more of us were working from home and needed to access sensitive data remotely. 

Leveling the playing field

As an organization, we were surprised by how many team members didn’t have personal computers at home. We quickly decided that if we needed team members to work from home, the health network would have to supply hardware. Chromebooks’ lower price tag compared to PCs—on top of their built-in security controls—allowed us to purchase, deploy, and support that initial distribution of 3,000 Chromebooks to team members in less than three weeks, providing devices to every eligible remote employee instead of just a select few. This was vital to reaching our equitable technology goal as part of our diversity and inclusion initiative: everybody has the same tools to do good work.

When all employees have what they need to do their jobs well, we get better patient outcomes. Before we began this cloud adoption journey, patient and staff experiences were different within the hospitals and outside of them. 

Now it’s the same wherever our staff is, and we’ve seen efficiency and accessibility benefits extend to the patient side. For example, we built a web-based contact center that supports 80 locations that use Workspace and Chrome OS devices. Since customer service, admin, and providers are all on the same system, it has become a one-stop shop for patients.

Furthermore, through the Grow with Google program, we were able to provide another benefit to employees that drove our equity goals. Google trained 50 non-IT staff members—from environmental services, food and nutrition, and other non-tech areas who were interested in making a career change to IT—on the Google products we were using. They may not have thought about switching to a career in IT before the Grow with Google program came to our organization, but through this partnership, they now have that opportunity.

A strategic, long-term partner

With any large-scale rollout, the work doesn’t end once laptops are in employee hands. Google has shown their commitment to long-term collaboration as they continuously optimize their products for the unique needs of healthcare providers and go the extra mile in tailoring tools to our staff’s workflows. 

For example, on the Chrome OS side, the Google team has helped our registration desks and document centers with device integration for hardware like credit card readers and e-signature pads. They’ve also helped us meet security and privacy requirements mandated by state and federal governments around HIPAA, Medicaid, and Medicare reimbursements. Over this next year, we’ll look at a feature roadmap with Google Cloud to deliver further enhancements, iterating on the product itself to meet our needs for the present and the future.

Delivering the future of healthcare

The benefits we’ve seen around security and usability—and the ability to provide all staff with equal access to Google’s technology—are why we’re expanding our partnership with Google to both the administrative and clinical sides of HMH. In addition to further Google rollouts with corporate, next year we’re distributing Chromebooks to all 350 of our ambulatory clinics.

We’re also working with the Google professional services team to create a custom AI model that analyzes 3D mammogram images. This AI model will enable two providers to read mammograms—which adheres to international best practices but is currently rare in the US—without requiring additional time. Conducting double readings of mammograms will yield better health outcomes for our patients, such as a lower patient recall rate and an increased accuracy in detecting breast cancer.

We’re currently building the model using a variety of Google Cloud products, including Cloud Healthcare API. Once complete, this model is expected to be trained, deployed, and maintained in Google’s Vertex AI, allowing our providers to be more productive as they make clinical decisions with AI support. As the model is proven over time, we plan to make the predictive services accessible to other healthcare organizations.

With Google, we’re able to achieve a unified architecture for storing data as well as training and deploying AI models, which enable our staff to work more efficiently and securely from anywhere. While I may not be able to predict the future as accurately as AI can, I foresee our continued partnership with Google as a key part of HMH’s improved provider and patient outcomes.

You are what you eat. We’ve all been told this, but the truth is what we eat is often more complex than we are – genetically at least. Take a grain of rice. The plant that produces rice has 40,000 to 50,000 genes, double that of humans, yet we know far more about the composition of the human genome than of plant life. We need to close this knowledge gap quickly if we are to answer the urgent challenge of feeding 8 billion people, especially as food security around the globe is likely to worsen with climate change.

For this reason, AI company InstaDeep has teamed up with Google Cloud to train a large AI model with more than 20 billion parameters on a dataset of reference genomes for cereal crops and edible vegetables, using the latest generation of Google’s Tensor Processing Units (Cloud TPU v4), which is particularly suited for training efficiency at scale. Our aim is to improve food security and sustainable agriculture by creating a tool that can analyze and predict plants’ agronomic traits from genomic sequences. This will help identify which genes make some crops more nutritious, more efficient to grow, and more resilient and resistant to pests, disease and drought.

Genomic language models for sustainable agriculture

Ever since farming began, we have been, directly or indirectly, trying to breed better crops with higher yields, better resilience and, if we’re lucky, better taste too. For thousands of years, this was done by trial and error, growing crops year-on-year while trying to identify and retain only the most beneficial traits as they naturally arise from evolutionary mutations. Now that we have access to the genomic sequences of plants, we hope to directly identify beneficial genes and predict the effect of novel mutations.

However, the complexity of plant genomes often makes it difficult to identify which variants are beneficial. Revolutionary advances in machine learning (ML) can help to understand the link between DNA sequences and molecular phenotypes. This means we now have precise and cost-effective prediction methods to help us close the gap between genetic information and observable traits. These predictions can help identify functional variants and accelerate our understanding of which genes link to which traits – so we can make better crop selections.

Moreover, thanks to the vast library of available crop genetic sequences, training large models on hundreds of plant genomes means we can transfer the knowledge from thoroughly-studied species to those that are less understood but important for food production – especially in developing countries. And by doing this digitally, AI can quickly map and annotate the genomes of both common and rare crop variants.

One of the major limitations of traditional ML methods for plant genomics has been they mostly rely on supervised learning techniques. They need labeled data. Such data is scarce and expensive to collect, severely limiting these methods. Recent advances in natural language processing (NLP), such as Transformer architectures and BERT-style training (Bidirectional Encoder Representations from Transformers), allow scientists to train massive language models on raw text data to learn meaningful representations. This unsupervised learning technique changes the game. Once learned, the representations can be leveraged to solve complex regression or classification tasks – even when there is a lack of labeled data.

InstaDeep partners with Google Cloud to train the new generation of AI models for genomics on TPUs

Researchers have demonstrated that large language models can be especially effective in proteomics. To understand how this works, imagine reading amino acids as words and proteins as sentences. The treasure trove of raw genomics data – in sequence form – inspired InstaDeep and Google Cloud to apply similar technologies on nucleotides, this time reading them as words and chunks of genomes as sentences.

Moreover, the representations that the system learned improved in line with the size of the models and datasets, NLP research studies showed. This finding led InstaDeep researchers to train a set of increasingly larger language models on genomics datasets ranging from 1 billion to 20 billion parameters.

• Models of 1 billion and 5 billion parameters were trained on a dataset comprising the reference genomes for several edible plants, including fruit, cereal and vegetables for a total of 75 billion nucleotides.
• The training dataset must increase in the same proportion as the model capacity, recent work has shown. Thus, we created a larger dataset gathering all reference genomes available on the National Center for Biotechnology Information (NCBI) database including human, animal, non-edible plant and bacteria genomes. This dataset, which we used to train a 20-billion-parameter Transformer model, comprised 700 billion tokens, exceeding the size of most datasets typically used for NLP applications, such as the Common Crawl or Wikipedia dataset.
• Both teams announced that the 1 billion-parameter model will be shared with the scientific community to further accelerate plant genomics research.

The compact and meaningful representations of nucleotide sequences learned by these models can be used to tackle molecular phenotype prediction problems. To showcase their ability, we trained a model to predict the gene function and gene ontology (i.e. a gene’s attribute) for different edible plant species.

Early results have demonstrated that this model can predict these characteristics with high accuracy – encouraging us to look deeper at what these models can tell us. Based on these results, we decided to annotate the genomes of three plant species with considerable importance for many developing countries: cassava, sweet potato, and yam. We are working on making these annotations freely available to the scientific community and hope that these will be used to further guide and accelerate new genomic research.

Overcoming scaling challenges with massive models and datasets with Cloud TPUs

The compute requirement for training our 20 billion-parameter model with billions of tokens is massive. While modern accelerators offer impressive peak performance per chip, to utilize this performance often requires tightly coupled hardware and software optimizations. Moreover, maintaining this efficiency when scaling to hundreds of chips presents additional system design challenges. The Cloud TPU’s tightly-coupled hardware and software stack is especially well suited to such challenges. The Cloud TPU software stack is based on the XLA Compiler which offers out-of-the-box optimizations (such as compute and communication overlap) and an easy programming model for expressing parallelism.

We successfully trained our large models for genomics by leveraging Google Tensor Processing Units (TPUv4). Our code is implemented with the JAX framework. JAX provides a functional programming-based approach to express computations as functions that can be easily parallelized using JAX APIs powered by XLA. This helped us to scale from a single host (four chips) configuration to a multi-host configuration without having to tackle any of the system design challenges. The TPU’s cost-effective inter- and intra-communication capabilities led to an almost linear scaling between the number of chips and training time. This allowed us to train the models quickly and efficiently on a grid of 1024 TPUv4 cores (512 chips).

Conclusion

Ultimately, our hope is that the functional characterization of genomic variants predicted by deep learning models will be critical to the next era in agriculture, which will largely depend on genome editing and analysis. We envisage that novel approaches, such as in-silico mutagenesis – the assessment of all possible changes in a genomic region by a computer model – will be invaluable in prioritizing mutations that improve plant fitness and guiding crop improvements. Attempting similar work in wet-lab experiments would be difficult to scale and nearly impossible in nature. By making our current and future annotations available to the research community, we also hope to help democratize breeding technologies so that they can benefit all of global agriculture.

Further Reading

To learn more about the unique features of Cloud TPU v4 hardware and software stack we encourage readers to explore Cloud TPU v4 announcement. To learn more about scaling characteristics, please see this benchmark and finally we recommend reading PJIT Introduction to get started with JAX and SPMD parallelism on Cloud TPU.

This research was made possible thanks to the support of Google’s TPU Research Cloud (TRC) Program which enabled us to use the Cloud TPUv4 chips that were critical to this work.

In our conversations with technology leaders about data-driven transformation using Google Data Cloud –  industry’s leading unified data and AI solution – , one important topic is incorporating continuous intelligence to move from answering questions such as “What has happened? to questions like “What is happening?” and “What might happen?”. The core to this evolution is the need for an underlying data processing that not only provides powerful real-time capabilities for events happening close to origination, but also brings together existing data sources under one unified data platform to enable organizations to draw insights and take actions holistically. Dataflow, Google’s cloud-native data processing and streaming analytics platform, is a key component of any modern data and AI architecture and data transformation journey, along with BigQuery, Google’s internet-scale warehouse with built-in streaming, BI engine and ML; Pub/Sub, a global no-ops event delivery service; and Looker, a modern BI and embedded analytics platform. One of the key evaluation factors is potential economic value of Dataflow to their organization, particularly in the context of engaging other stakeholders is key for many of the leaders that we engage with. So we commissioned Forrester Consulting to conduct a comprehensive study on the impact that Dataflow had on their organization by interviewing actual customers . 

Today we’re excited to share our commissioned study conducted by Forrester Consulting, the Total Economic Impact™ of Google Cloud Dataflow, which allows data leaders to understand and quantify the benefits of Dataflow, and use cases it enables. Forrester conducted interviews with Dataflow customers to evaluate the benefits, costs, and risks of investing in Dataflow across an organization. Based on their interviews, Forrester identified major financial benefits across four different areas: business growth, infrastructure cost savings, data engineer productivity, and administration efficiency. In fact, Forrester found that customers adopting Dataflow can achieve a 55% boost in developer productivity and a 50% reduction in infrastructure costs. In fact, Forrester projects that customers adopting Dataflow can achieve a range of up to 171% Return on Investment (ROI) and a less than six months payback period. Customers can now use figures in the report to compute their own Return on Investment (ROI) and payback period.

“Dataflow is integral to accelerating time-to-market, decreasing time-to-production, reducing time to figure out how to use data for use cases, focusing time on value-add tasks, streamlining ingestion, and reducing total cost of ownership.” – Lead technical architect, CPG

Let’s take a deeper look at the ways that Forrester found that Dataflow can help you achieve your goals and unlock your business potential. 

Benefit #1: Increase data engineer productivity by 55%

Developers can choose among a variety of programming languages to define and execute data workflows. Dataflow also seamlessly integrates with other Google Cloud Platform and open source technologies to maximize value and applicability to a wide variety of use cases. Dataflow streamlined workflows with code reusability,dynamic templates, and the simplicity of a managed service. Engineers trusted pipelines to run correctly and adhere to governance. Data engineers avoided laborious issue-monitoring and remediation tasks that were common in the legacy environments such as poor performance, lack of availability, and failed jobs. Teams valued the language flexibility and open source base.

“Dataflow provided us with ETL replacement that opened limitless potential use cases and enabled us to do smarter data enhancement while data remains in motion.” — Director of data projects, financial services

Benefit #2: Reduce infrastructure costs by up-to 50% for batch and streaming workloads 

Dataflow’s serverless autoscaling and discrete control of job needs, scheduling, and regions eliminated overhead and optimized technology spending. Consolidating global data processing solutions to Dataflow further eliminated excess costs while ensuring performance, resilience, and governance across environments. Dataflow’s unified streaming and batch data platform gives organizations the flexibility to define either workload in the same programming model, run it on the same infrastructure, and manage it from a single operational management tool. 

“Our costs with our cloud data platform using Dataflow are just a fraction of the costs we faced before. Now we only pay for cloud infrastructure consumption because the open source base helps us avoid licensing costs. We spend about $120,000 per year with Dataflow, but we’d be spending millions with our old technologies.” – Lead technical architect, CPG

Benefit #3: Increase top-line revenue by improving customer experience and retention with payback time of < 6 months

Streaming analytics is an essential capability in today’s digital world to gain real-time actionable insights. Likewise, organizations must also have flexible, high- performance batch environments to analyze historical data for building machine learning models, business intelligence, and advanced analytics. Dataflow enabled real-time streaming use cases, improved data enrichment, encouraged data exploration,improved performance and resiliency, reduced errors, increased trust, and eliminated barriers to scale. As a result, organizations provided customers with more accurate, relevant, and in-the-moment data-backed services and insights — boosting customer experience, creating new revenue streams, and improving acquisition, retention, and enrichment.

“It’s already been proven that we are getting more business [with Dataflow] because we can turn around results faster for customers.” – VP of technology, financial services technology

“When we provide data to our customers and partners with Dataflow, we are much more confident in those numbers and can provide accurate data within a minute. Our customers and partners have taken note and commented on this. It’s reduced complaints and prevented churn.” – Senior software engineer, media

Other benefits 

Eliminated administrative overhead and toil

As a cloud-native managed service, all administration tasks such as provisioning, scaling, and updates are automatically handled by Google Cloud. Teams no longer need to manage servers and related software for legacy data processing solutions. Admins also streamlined processes for setting up data sources, adding pipelines, and enforcing governance.

Saved business operations costs for support teams and data end users

Dataflow improved the speed, quality, reliability, and ease of access to data for insights for general business users, saving time and empowering users to drive better data-backed outcomes. It also reduced support inquiry volume while automating manual job creation.

What’s next?

Download the Forrester Total Economic Impact study today to dive deep into the economic impact Dataflow can deliver your organization. We would love to partner with you to explore the potential Dataflow can unlock in your teams. Please reach out to our sales team to start a conversation about your data transformation with Google Cloud.

Volkswagen strives to design beautiful, performant, and energy efficient vehicles. This entails an iterative process where designers go through many design drafts, evaluating each, integrating the feedback, and refining.

For example, a vehicle’s drag coefficient—its resistance to air—is one of the most important factors of energy efficiency. Thus, getting estimates of the drag coefficient for several designs helps the designers experiment and converge toward more energy-efficient solutions. The cheaper and faster this feedback loop is, the more it enables the designers.

Unfortunately, estimating drag coefficient is an expensive and time-consuming operation that involves either a physical wind tunnel or a computationally intensive simulation. This can be a bottleneck in the feedback cycle.

For this reason, Volkswagen and Google Cloud decided to collaborate on a joint research project to investigate using machine learning (ML) to get fast and inexpensive estimates of the drag coefficient. In this post, we’ll explore the challenges and approaches undertaken in this project.

The core principles of the project were simple. First, we needed to collect a dataset of existing car designs and their respective drag coefficients. Then, we needed to create a representation of the various cars that would be suitable for ML. The next step was to train a deep learning model to predict the drag coefficient, and then, finally, we would use that model to efficiently estimate drag for any new design.

Representing three-dimensional car designs

Design software recreates a physical object as a three-dimensional triangle mesh made up of three types of objects—faces, edges, and vertices. Figure 1, below, shows such a mesh for an Audi S6. Faces are flat surfaces, such as the window in a car door. An edge is where two faces meet (e.g., the side of the door), and a vertex is where two or more edges meet, such as the corner of the door.

Car bodies, however, come in all shapes and sizes. A Volkswagen Golf economy model is very different from a Tiguan SUV, and a single vehicle can have both large smooth surfaces as well as areas with delicately designed features. Consequently, there can be a huge variety from one polygonal mesh to the next.

ML models need consistent representation in order to form robust generalized rules. With such a dramatic variance between each polygonal mesh, the models would be compromised and the results could have huge margins of error.

We needed to find a way to create simple meshes that capture the shape of the car but are still suited for ML models.

Representing a car with digital shrink wrapping

Rather than building a representation of each car from the ground up, we applied a “shrink wrapping” method for the 3D meshes. The principle is very similar to vacuum-sealing a cucumber. The cucumber is placed in a plastic bag and the air is then gradually removed until the bag fits tightly around it, capturing its shape.

Our approach works similarly: we start with a base mesh, a simple shape that corresponds to the plastic bag, and we deform it until it captures the shape of the target mesh. For our purposes, the base mesh is a simplified representation of a car and the target mesh is the particular car we are designing for at that moment. Such meshes can be defined, managed, and presented to ML models for training using the Tensorflow Graphics and trimesh libraries.

Our “shrink wrapping” method mainly works by iteratively minimizing a measure of distance (e.g., chamfer distance) between the two meshes. Additionally we can regularize our mesh to preserve certain qualities, like smoothness, in the resulting mesh. This iterative optimization is analogous to the vacuum pump, gradually shrinking and fitting the vertices of the mesh as closely as possible to the complex shape of the car. With shrink-wrapping, we are able to produce cleaner meshes that are more suitable to our estimation task. An example of such a procedure is shown in Figure 2.

How to train a model

Shrink-wrapping the 3D car designs was an important first step, but the work was far from over. Our next challenge was to build and test the machine learning algorithms.

We wanted our algorithms to estimate the drag coefficient as accurately and quickly as possible each time it looked at a new design. To do so, we had to train the ML models on existing data.

From publicly available datasets, we calculated the drag coefficients for 800 different car meshes, which we trained the models on. Then, we evaluated the trained models on a further 100 meshes, seeing how accurate their estimates were on new data.

As we worked through this training, we refined our approach. Initially, we tested models based on convolutional neural networks – similar to PointNet – that observed only the vertices, i.e., the fixed points in each mesh. But when we tested mesh-convolutional models – similar to FeastNet – we found a slightly different focus improved the accuracy of the estimates. Rather than focusing on vertices alone, these models looked at a mesh of vertices and how they relate to each other. These models placed each vertex in a richer context, leading to more accurate estimates when air-flow hit particularly subtle design features.

Working in parallel and at scale
To collaborate across time zones and two organizations, we’ve used the Google Cloud Vertex AI platform.

Vertex AI Workbench serves as a central hub to interact with other services and infrastructure on the Vertex AI platform. It enables quick experiments and preparation of training packages for resource-intensive ML model training jobs, all in a Python notebook environment for immediate execution of code. The notebook environments allow code-based interaction with other services on Google Cloud and ML tools such as Vertex AI Training and Vertex AI Pipelines.

The process of training a new model is a seamless one. First, a dataset is prepared and stored in Google Cloud Storage, usually with the help of Tensorflow Datasets. Then, for every ML model we want to test, we package and store the training code as a container image with Google Cloud Build and Container Registry. This ensures that every job is fully documented, including the provided parameters, training code package, logs from the training task, and resulting artifacts such as metrics and model files.

From there, we submit the model to the Vertex AI Training service, which provides easy access to large scale infrastructure and hardware accelerators, such as GPUs and TPUs, by simply defining resource needs when submitting a job. By using Vertex AI Training’s hyperparameter tuning feature, we can run experiments in parallel with multiple neural networks to find the right one for our purposes.

With Vertex AI Tensorboard, we can capture metrics and visualize the results of our experiments. These are readily available to anyone in the team, wherever they are in the world, for a wider discussion.

The first milestone

This joint research effort between Volkswagen and Google has produced promising results with the help of the Vertex AI platform. In this first milestone, the team was able to successfully bring recent AI research results a step closer to practical application for car design. This first iteration of the algorithm can produce a drag coefficient estimate with an average error of just 4%, within a second.

An average error of 4%, while not quite as accurate as a physical wind tunnel test, can be used to narrow a large selection of design candidates to a small shortlist. And given how quickly the estimates appear, we have made a substantial improvement on the existing methods that take days or weeks. With the algorithm that we have developed, designers can run more efficiency tests, submit more candidates, and iterate towards richer, more effective designs in just a small fraction of the time previously required.

Going forward, faster and more accurate estimates could even enable more automated searching for efficient designs, which would help both engineers and designers to hone in on the areas of the vehicle body where they could have the most impact. An important next step will be integrating the results into 3D design software to let designers benefit from the output and provide feedback.

As we continue, our focus is on improving the accuracy of the models. Firstly, we will build a larger, better quality dataset. Secondly, we will improve our shrink-wrapping algorithm to capture more details. Finally, we will enhance our existing models by experimenting with Vertex AI Neural Architecture Search to explore and experiment with different neural architecture options.

Moreover, we believe that our results for drag coefficient estimation is only a starting point for further exploration. There could potentially be numerous use cases in the space of physical simulations and assessments where cost and time savings could be achieved through ML-based estimators.

Acknowledgements

This work wouldn’t have been possible without the contributions from Volkswagen Data:Lab, Google Research, and Google Cloud. Thanks to Ahmed Ayyad, Dr. Andrii Kleshchonok, Dr. Daniel Weimer, Gülce Cesur, Henrik Bohlke, Andreas Müller from Volkswagen, Ameesh Makadia, Ph.D., and Carlos Esteves, Ph.D., from Google Research, and Daniel Holgate, Holger Speh, and Dr. Michael Menzel from Google Cloud.