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.

The game of baseball has no shortage of statistics — from batting average to exit velocity, strikeouts to wins above replacement. Among all sports, Major League Baseball (MLB) arguably contains the most analytical and data-driven participants and fan base. Subconsciously or viscerally, players and managers on the field and those following from anywhere are constantly assessing and making decisions based off of game play trends and expectations — whether a batter will come through with a hit in an important situation, when a pitcher should be pulled. Less analyzed, however, is what leads fans to become engaged with certain players or teams, and what factors drive their love of the game. This is the motivation behind the problem being posed by Major League Baseball in their Kaggle competition for Player Digital Engagement Forecasting. Can you use machine learning to deconstruct baseball fandom?

This competition asks you to predict measures of digital engagement for each active player on a daily basis during the MLB season. So, how large was the surge in fan interest after Joe Musgrove threw the first no-hitter in Padres history? Is Shohei Ohtani’s engagement higher when he pitches well, when he hits a monster home run…or when he does both? You’re provided a wealth of game, team and player information – detailed stats, awards, rosters, and transaction information – as well as social and digital engagement data as your inputs. Data scientists will recognize this as an exciting forecasting problem with both traditional regression and time series components, where having this input data just prior to the prediction date is critical to determining which players will receive the most engagement.

With so many variables in the game, there are an endless number of vectors which could possibly influence fan engagement. Eleven-time All-Star Miguel Cabrera delighted fans by hitting the first home run of the season – in the snow! Occasionally a lesser-known player like Musgrove or Carlos Rodón “wins the day” with an unlikely no-hitter. And sometimes just getting traded to an iconic franchise like the Yankees generates a ton of fan interest, like it did for Rougned Odor in early April.

As these examples show, a player’s digital engagement can be pretty dynamic during the season, with many different potential contributors to who is “trending” on a given day. How can you use data to uncover which factors are the most influential of engagement with each player’s digital content?

Ready to play ball? Check out the competition on Kaggle for all the details. $50,000 in prizes is up for grabs in two prize categories. The code competition puts your machine learning skills to the test, to see who can build the most accurate forecasting models to predict daily digital engagement for every active player. You’ll have until July 31st to build your models and then be evaluated on a future time frame, which will determine the winners. For data visualization and exploration experts out there, the explainability prizes give you an opportunity to analyze more broadly which factors, even those outside of what we’re providing directly, most influence digital engagement. You’ll be evaluated on how well you can use what the data is telling you to support your findings.

And if you’re looking to get started, we’ve provided an introductory video and some notebook tutorials, including a starting point for harnessing the power of Vertex AI through tools including Cloud Notebooks, Explainable AI, and Vizier.

With the second half of the season upon us, it’s an exciting time to be an MLB fan. With this Kaggle competition, it’s also a perfect opportunity to use data science to help understand baseball fandom and potentially earn some of your own accolades in the process. Step up to the plate!

Major League Baseball trademarks and copyrights are used with permission of Major League Baseball. Visit MLB.com.

Communication service providers (CSPs) are at an inflection point. From stagnating revenues, to network strain in meeting the demands of 5G, to challenges in delivering innovative customer experiences, there’s enormous pressure on the telecommunications industry to transform.

Over the past few years, CSPs around the globe have turned to artificial intelligence (AI) to address some of these challenges, but the lion’s share of an operator’s operational expenses are still spent on infrastructure and data management. This has limited their ability to capitalize on core data assets and develop differentiated customer experiences that meet individual needs. 

Enter generative AI, a type of machine intelligence that’s received enormous attention as of late. We’ve all marveled over its ability to generate text that reads like it was written by a person, to create new images, and to build even musical scores. It’s a riveting addition to the AI toolset — and one that complements machine learning (ML) and its ability to identify patterns to make predictions, spot efficiencies, or interpret large data sets. 

But while there’s a lot of hype around generative AI, at Google Cloud we view it through a much more practical lens for the telecommunications industry. Generative AI can accelerate the transformation already underway, with its potential to streamline many of the tools and processes that CSPs engage with daily, bringing a new level of natural interaction between people and computers, and enabling machines to be programmed to carry out an action with a spoken request, and respond in natural, interactive ways. 

Generative AI builds on existing Google Cloud data, AI, and ML services. For example, Contact Center AI, with human-like interactions between callers and computers, has been successfully adopted by CSPs for many years, increasing the satisfaction of both customers and call center workers. As we add generative AI to this technology, CSPs and their customers will see even greater capabilities and impact, for example, virtual agents that not only provide helpful information, but also let customers make payments and execute other transactions. With generative AI, CSPs will be able to harness customer call summaries to better understand customer sentiment and identify cross-sell and up-sell opportunities. CSPs could also easily and quickly build and deploy virtual agents informed by customer conversations that enable more innovative and personalized customer interactions. And that’s just the start.

Three key areas

The contact center is just one of the areas where practical and generative AI will help drive new value. When reflecting on the key challenges facing CSPs today, three areas in particular stand out where generative AI may be transformative:

1. Personalized experiences: In addition to further improvements in customer call center interactions, generative AI can deliver improved personalization in ecommerce interactions — a big factor in helping customers sort through their choices of phones and calling plans. Personalization is also important for lowering churn, offering relevant new services, and managing the customer lifecycle. For example, generative AI could enable CSPs to produce marketing campaign content customized for select themes, and target individual customers with customized text and images. 
2. Autonomous networks: Generative AI will also help to pave the way for autonomous networks by connecting multiple complex AI/ML models used across network planning and operations with large language models (LLMs) that can understand network behaviors and create action plans in areas including network capacity planning and performance. For example, generative AI will enable CSPs to train models with customer experience and sentiment data to build better prediction capabilities. Importantly, the customer data sets used to tune these models are not public, but curated internal customer data — significantly enhancing privacy, factuality, and relevance, while protecting intellectual property. In addition, generative AI will be able to help network planning and design, which requires high levels of reporting and analysis.
3. Streamlined operations: Both operations center uptime and field service efficiency are crucial for managing costs and improving customer satisfaction. In particular, applying generative AI to field service devices can speed up diagnostics and analysis, and can even help with installation, parts, and troubleshooting, and minimizing the number of times companies have to send out trucks and improve field-service training. Generative AI will also provide a productivity boost to IT development processes, enabling code generation and troubleshooting to deliver reliable software products and services.

Data security and reliability

An under-discussed area in generative AI is the importance of data quality and data security in building and training the LLMs that power the technology. Many CSPs are rightly concerned about intellectual property leaking both into and out of LLMs, risking the security of their systems and their intellectual property. We’ve long offered industry-leading data security and privacy technologies, and with generative AI integrated with Vertex AI, we can ensure all data is secured within the CSP’s environment.

Meanwhile, to ensure that their LLMs generate accurate information, CSPs are building out scenarios and use cases for training on smaller, controlled amounts of their own data, sometimes accompanied by highly trusted sources from partners and others. Google Cloud also provides tools including Prompt Engineering, Tuning, and Reinforcement Learning from Human Feedback to further ensure data factuality and reliability. This will likely result in the first generative AI applications targeted to smaller, high-impact problems, like optimizing network topologies. 

The human element

Of course, people are critical to the success of generative AI, whether that means solving problems in call centers, field service workers combining AI information with their own knowhow, marketing and creative teams brainstorming with generative AI to make new presentations and marketing materials, or operations engineers augmenting and approving an AI suggestion. We’ve built a lot of amazing technology to help augment what people cannot do: synthesize millions, if not billions, of records and sources to help energize new workflows and productivity.

Telecommunications is a fast-changing industry that is tech savvy and hungry to learn and deploy the best possible new technologies, and that includes generative AI. Each meeting with a CSP inspires new ideas, sparks more use cases, and leads to more industry-changing initiatives. It’s exciting to see this pace of change, and we are only just getting started.

We look forward to showing you more soon as we deep dive into some of the exciting CSP industry use cases in upcoming blog posts. We also invite you to find out more about how Google Cloud is partnering with CSPs around the world to deliver a holistic cloud transformation.

Global life science supply chains are lengthy and complex with many moving parts. One small disruption can create serious delays and affect your ability to deliver therapeutics for patients.

Supply chain disruptors

Over the last few years, healthcare organizations have encountered a range of obstacles, from both internal and external factors, that have resulted in supply networks failing to get drugs and medical devices to where they need to be on time. These obstacles include:

• Labor and supply shortages
• Rising material costs
• Raw material constraints
• Geo-political events
• Unpredictable weather

How do you overcome supply chain disruptors that are out of your control?

The intelligent healthcare supply chain

While many organizations have already implemented data-driven supply chains, organizations are still faced with the challenges of static, siloed, and different functional supply chain applications; limited data exchange with key trading partners across upstream and downstream operations; and the inability to effectively leverage relevant external data.

At Google Cloud, we believe the key to meaningful and effective change is a data-driven supply chain that allows you to achieve visibility, flexibility, and innovation.

Our solutions help you prepare for the unpredictable and enhance the value of your data. By unlocking AI-driven insights, you can strengthen distribution networks and optimize your workflows and supply chains to become more reliable, intelligent, and sustainable. Some of the business challenges we address include:

• Predicting demand with Vertex AI Forecast
• Visual inspection for quality and predictive maintenance with pre-built ML models
• Automating and optimizing pickup and delivery operations with Cloud Fleet Routing API
• Real time and holistic inventory visibility with Supply Chain Twin

Make sure you’re prepared for the unpredictable with real-time visibility over your distribution networks. Learn how you can harness the power of AI and analytics and gain actionable insights that enhance your supply chain.