Monthly Archives: January 2022

Building apps for Android Automotive OS

Posted by Madan Ankapura, Product Manager

Today we’re announcing the availability of version 1.2 beta of the Car App Library, enabling app developers to start building their navigation, parking, and charging apps for Android Automotive OS.

Now, developers can begin building and testing apps for these categories using the Automotive OS emulator across both Android Automotive OS and Android Auto. For the entire list of changes in v1.2 beta, please see the release notes. To start building your app for the car, check out our updated developer documentation, car quality guidelines, and design guidelines.

As announced earlier, drivers of Polestar 2 and Volvo cars can now download charging (ChargePoint, PlugShare), parking (Spothero, Parkwhiz), and navigation (Flitsmeister, Sygic) apps developed with the Car App Library by joining the Google Group and opting-in to each app's beta on the Google Play store, with your Gmail account.

six spp icons

Car App Library apps on Android Automotive OS are automatically rendered to be consistent with the rest of the experience within each car, without additional work needed from developers.. For example,

Polestar 2
Polestar 2 setting with labeled On / Off switches for PlugShare

Polestar 2 setting with labeled On / Off switches for PlugShare

Volvo settings with sliding switches for PlugShare

Volvo settings with sliding switches for PlugShare

Polestar 2 sign-in screen for SpotHero

Polestar 2 sign-in screen for SpotHero

Volvo sign-in screen for SpotHero

Volvo sign-in screen for SpotHero

Example of app customization on Android Automotive OS

Experience for yourself how your app will look within the different systems, by accessing the OEM emulator system images downloadable in Android Studio. You can begin developing your charging, parking and navigation apps for Android Automotive OS today, and we are working to enable you to publish your apps to the Google Play store in the coming months (stay tuned!).

Beyond navigation, rideshare drivers spend a lot of time in their vehicles and will benefit from safer interactions if those apps can be brought to the car’s screen. We are working with Lyft and Kakao Mobility to bring their driver app experiences into the car in the coming months.

image of car screen with gps map and Lyft logo

We are also pleased to announce that we are expanding support to all Points of Interest apps. Beyond charging and parking, this allows any app that will help users discover and search for interesting locations on a map, and optionally enable them to navigate to such points. We are partnering with MochiMochi, Fuelio, Prezzi Benzina, and NAVITIME JAPAN as our early access partners.

If you’re interested in joining our Early Access Program in the future, please fill out this interest form. You can get started with the Android for Cars App Library today, by visiting

Use Directory Sync beta for fast and easy sync of users and groups

What’s changing 

We’re launching a beta for a new Directory Sync solution which can make it quicker and easier to synchronize your Active Directory user and group data with your Google Cloud directory. 

Directory Sync is an alternative to Google Cloud Directory Sync (GCDS). You might want to consider it if you want to sync Microsoft Active Directory LDAP data with your Google cloud directory using a completely cloud-based solution, without the need to manage on-prem hardware and deployments. Please read more about its features and network requirements to learn whether it's right for you. 

Who’s impacted 


Why you’d use it 

The new Directory Sync solution is: 
  • Cloud based: The cloud-based sync process is auto-scheduled to run on a continuous basis in a loop, so there’s no need to install a sync client or on-premises software. 
  • Easy to use: A simple and modern UI integrated with the Admin console makes it easy to use for those with no LDAP knowledge. Plus, there’s no need for Google exclusion rules if you would like to sync from multiple Active Directory sources or manage a subset of users or groups within Google without synching from Active Directory. 
  • Integrated reporting: It offers centralized reporting in the Admin console. You can filter, search, and set custom alerts. 
  • Native multi-directory support: You can sync users and groups from more than one Active Directory source. 

The initial scope for the Directory Sync beta supports user and group sync with Active Directory only, and covers a limited range of attributes. In the future, we’ll add other features, including support for additional attributes, OU mapping to automatically place new user accounts in OUs, and more types of data. 

Additional details 

If your Active Directory server is located on-premises or hosted outside a Google Cloud environment, you’ll require a connection between Google Cloud and the LDAP server using Cloud VPN or Cloud Interconnect. Learn more about system requirements for using Directory Sync and supported network connections for Directory Sync

Getting started 

  • Admins: To use the Directory Sync beta, go to Admin console > Home > Directory > Directory Sync. No beta sign up or registration is required. You can delegate the ability to manage Active Directory with the new Directory Sync admin user role. Use our Help Center to learn more about using the new Directory Sync, and see FAQs about Directory Sync. 
  • End users: No end user impact 

Rollout pace 


  • Available to all Google Workspace customers, as well as G Suite Basic and Business customers and Cloud Identity customers 


New Navigation bar functions in Google Drive

Quick Summary 

From the URL bar in Google Drive, you can now quickly access key pages and functions When navigating into the Google Drive web application from the URL bar by hitting the "Tab" key, you can access buttons like "Skip to main content", "Keyboard shortcuts", and "Accessibility feedback"from the bar at the top of the page. 

Access from the navigation bar

Access from the navigation bar

Getting started 

Rollout pace 


  • Available to all Google Workspace customers, as well as G Suite Basic and Business customers 


Does Your Medical Image Classifier Know What It Doesn’t Know?

Deep machine learning (ML) systems have achieved considerable success in medical image analysis in recent years. One major contributing factor is access to abundant labeled datasets, which are used to train highly effective supervised deep learning models. However, in the real-world, these models may encounter samples exhibiting rare conditions that are individually too infrequent for per-condition classification. Nevertheless, such conditions can be collectively common because they follow a long-tail distribution and when taken together can represent a significant portion of cases — e.g., in a recent deep learning dermatological study, hundreds of rare conditions composed around 20% of cases encountered by the model at test time.

To prevent models from generating erroneous outputs on rare samples at test time, there remains a considerable need for deep learning systems with the ability to recognize when a sample is not a condition it can identify. Detecting previously unseen conditions can be thought of as an out-of-distribution (OOD) detection task. By successfully identifying OOD samples, preventive measures can be taken, like abstaining from prediction or deferring to a human expert.

Traditional computer vision OOD detection benchmarks work to detect dataset distribution shifts. For example, a model may be trained on CIFAR images but be presented with street view house numbers (SVHN) as OOD samples, two datasets with very different semantic meanings. Other benchmarks seek to detect slight differences in semantic information, e.g., between images of a truck and a pickup truck, or two different skin conditions. The semantic distribution shifts in such near-OOD detection problems are more subtle in comparison to dataset distribution shifts, and thus, are harder to detect.

In “Does Your Dermatology Classifier Know What it Doesn’t Know? Detecting the Long-Tail of Unseen Conditions”, published in Medical Image Analysis, we tackle this near-OOD detection task in the application of dermatology image classification. We propose a novel hierarchical outlier detection (HOD) loss, which leverages existing fine-grained labels of rare conditions from the long tail and modifies the loss function to group unseen conditions and improve identification of these near OOD categories. Coupled with various representation learning methods and the diverse ensemble strategy, this approach enables us to achieve better performance for detecting OOD inputs.

The Near-OOD Dermatology Dataset
We curated a near-OOD dermatology dataset that includes 26 inlier conditions, each of which are represented by at least 100 samples, and 199 rare conditions considered to be outliers. Outlier conditions can have as low as one sample per condition. The separation criteria between inlier and outlier conditions can be specified by the user. Here the cutoff sample size between inlier and outlier was 100, consistent with our previous study. The outliers are further split into training, validation, and test sets that are intentionally mutually exclusive to mimic real-world scenarios, where rare conditions shown during test time may have not been seen in training.

Long tail distribution of different dermatological conditions in our dataset. The 26 inlier conditions, with at least 100 samples, (blue) and the remaining 199 rare outlier conditions (orange). Outlier conditions can have as low as one sample per condition.
    Train set  Validation set      Test set
Inlier Outlier Inlier Outlier Inlier Outlier
Number of classes 26 68 26 66 26 65
Number of samples 8854 1111 1251 1082 1192 937
Inlier and outlier conditions in our benchmark dataset and detailed dataset split statistics. The outliers are further split into mutually exclusive train, validation, and test sets.

Hierarchical Outlier Detection Loss
We propose to use “known outlier” samples during training that are leveraged to aid detection of “unknown outlier” samples during test time. Our novel hierarchical outlier detection (HOD) loss performs a fine-grained classification of individual classes for all inlier or outlier classes and, in parallel, a coarse-grained binary classification of inliers vs. outliers in a hierarchical setup (see the figure below). Our experiments confirmed that HOD is more effective than performing a coarse-grained classification followed by a fine-grained classification, as this could result in a bottleneck that impacted the performance of the fine-grained classifier.

We use the sum of the predictive probabilities of the outlier classes as the OOD score. As a primary OOD detection metric we use the area under receiver operating characteristics (AUROC) curve, which ranges between 0 and 1 and gives us a measure of separability between inliers and outliers. A perfect OOD detector, which separates all inliers from outliers, is assigned an AUROC score of 1. A popular baseline method, called reject bucket, separates each inlier individually from the outliers, which are grouped into a dedicated single abstention class. In addition to a fine-grained classification for each individual inlier and outlier classes, the HOD loss–based approach separates the inliers collectively from the outliers with a coarse-grained prediction loss, resulting in better generalization. While similar, we demonstrate that our HOD loss–based approach outperforms other baseline methods that leverage outlier data during training, achieving an AUROC score of 79.4% on the benchmark, a significant improvement over that of reject bucket, which achieves 75.6%.

Our model architecture and the HOD loss. The encoder (green) represents the wide ResNet 101x3 model pre-trained with different representation learning models (ImageNet, BiT, SimCLR, and MICLe; see below). The output of the encoder is sent to the HOD loss where fine-grained and coarse-grained predictions for inliers (blue) and outliers (orange) are obtained. The coarse predictions are obtained by summing over the fine-grained probabilities as indicated in the figure. The OOD score is defined as the sum of the probabilities of outlier classes.

Representation Learning and the Diverse Ensemble Strategy
We also investigate how different types of representation learning help in OOD detection in conjunction with HOD by pretraining on ImageNet, BiT-L, SimCLR and MICLe models. We observe that including HOD loss improves OOD performance compared to the reject bucket baseline method for all four representation learning methods.

Representation Learning
OOD detection metric (AUROC %)
With reject bucket With HOD loss
ImageNet 74.7% 77%
BiT-L 75.6% 79.4%
SimCLR 75.2% 77.2%
MICLe 76.7% 78.8%
OOD detection performance for different representation learning models with reject bucket and with HOD loss.

Another orthogonal approach for improving OOD detection performance and accuracy is deep ensemble, which aggregates outputs from multiple independently trained models to provide a final prediction. We build upon deep ensemble, but instead of using a fixed architecture with a fixed pre-training, we combine different representation learning architectures (ImageNet, BiT-L, SimCLR and MICLe) and introduce objective loss functions (HOD and reject bucket). We call this a diverse ensemble strategy, which we demonstrate outperforms the deep ensemble for OOD performance and inlier accuracy.

Downstream Clinical Trust Analysis
While we mainly focus on improving the performance for OOD detection, the ultimate goal for our dermatology model is to have high accuracy in predicting inlier and outlier conditions. We go beyond traditional performance metrics and introduce a “penalty” matrix that jointly evaluates inlier and outlier predictions for model trust analysis to approximate downstream impact. For a fixed confidence threshold, we count the following types of mistakes: (i) incorrect inlier predictions (i.e., mistaking inlier condition A as inlier condition B); (ii) incorrect abstention of inliers (i.e., abstaining from making a prediction for an inlier); and (iii) incorrect prediction for outliers as one of the inlier classes.

To account for the asymmetrical consequences of the different types of mistakes, penalties can be 0, 0.5, or 1. Both incorrect inlier and outlier-as-inlier predictions can potentially erode user trust in the model and were penalized with a score of 1. Incorrect abstention of an inlier as an outlier was penalized with a score of 0.5, indicating that potential model users should seek additional guidance given the model-expressed uncertainty or abstention. For correct decisions no cost is incurred, indicated by a score of 0.

                  Action of the Model
Prediction as Inlier Abstain
Inlier 0 (Correct)

1 (Incorrect, mistakes
that may erode trust)

0.5 (Incorrect,
abstains inliers)
Outlier     1 (Incorrect, mistakes
that may erode trust)
0 (Correct)
The penalty matrix is designed to capture the potential impact of different types of model errors.

Because real-world scenarios are more complex and contain a variety of unknown variables, the numbers used here represent simplifications to enable qualitative approximations for the downstream impact on user trust of outlier detection models, which we refer to as “cost”. We use the penalty matrix to estimate a downstream cost on the test set and compare our method against the baseline, thereby making a stronger case for its effectiveness in real-world scenarios. As shown in the plot below, our proposed solution incurs a much lower estimated cost in comparison to baseline over all possible operating points.

Trust analysis comparing our proposed method to the baseline (reject bucket) for a range of outlier recall rates, indicated by ?. We show that our method reduces downstream estimated cost, potentially reflecting improved downstream impact.

In real-world deployment, medical ML models may encounter conditions that were not seen in training, and it’s important that they accurately identify when they do not know a specific condition. Detecting those OOD inputs is an important step to improving safety. We develop an HOD loss that leverages outlier data during training, and combine it with pre-trained representation learning models and a diverse ensemble to further boost performance, significantly outperforming the baseline approach on our new dermatology benchmark dataset. We believe that our approach, aligned with our AI Principles, can aid successful translation of ML algorithms into real-world scenarios. Although we have primarily focused on OOD detection for dermatology, most of our contributions are fairly generic and can be easily incorporated into OOD detection for other applications.

We would like to thank Shekoofeh Azizi, Aaron Loh, Vivek Natarajan, Basil Mustafa, Nick Pawlowski, Jan Freyberg, Yuan Liu, Zach Beaver, Nam Vo, Peggy Bui, Samantha Winter, Patricia MacWilliams, Greg S. Corrado, Umesh Telang, Yun Liu, Taylan Cemgil, Alan Karthikesalingam, Balaji Lakshminarayanan, and Jim Winkens for their contributions. We would also like to thank Tom Small for creating the post animation.

Source: Google AI Blog

How a student leader cultivates on-campus diversity and leadership in Australia

Posted by Matthew Ranocchiari, Developer Relations Community Manager, Australia & New Zealand

Banner image shows Milindi Kodikara smiling with text that reads How a student leader cultivates on-campus diversity and leadership in Australia

To get familiar with her local community, International student Milindi Kodikara, originally from Sri Lanka, joined almost every tech club at RMIT University in Melbourne, Australia. She immediately identified an opportunity to establish a more diverse and inclusive tech club to prepare students to work in the tech industry. Using both her instincts and passion for community building, she set out on a mission to re-establish a Google Developer Student Club (GDSC) at RMIT University.

“I applied to become the GDSC Lead at RMIT University to create a better, brighter, and stronger community where our students (regardless of age, gender, sexuality, or race) can thrive, learn new skills, and enjoy their time with new friends,” Kodikara says.

Preparing students for tech careers

After starting the club, Milindi sought the advice of Matthew Ranocchiari (Google Developer Relations Community Manager for the region) who then suggested Milindi form a team, set goals, and plan activities for the semester. Milindi assembled a team of seven students hailing from Malaysia, the Philippines, Indonesia, Bangladesh, and Sri Lanka. They brought a mix of technical and non-technical backgrounds. Each shared an eagerness to learn about technology, help other students build projects, and develop leadership skills.

“ I was looking for a team of passionate individuals to make a change at RMIT University.”

Screenshot from a Google Video Chat shows the GDSC core team at RMIT University. They are all smiling at the camera

Social media officer Chaamudi Kodikara, helped set up the club’s various channels, which currently engage 400+ students every day. Design experts Isaac (Yi Jie) Chuah, Jacqueline Ann Lim, and Andrea Gocheco; event managers Sheryl Mantik and Kowsar Rahman Sadit; and a tech expert Prottay Karim each round out the group.

Planning a “Study Jam-packed” semester

When asking students what they wanted from the new club, most students said they were seeking an opportunity to learn how to land a tech job through networking, get a sense of what tech jobs are like day-to-day, and explore various technologies. Most of all, they wanted to feel empowered. With these goals in mind, the team planned multiple events per month, including an inaugural virtual Games Night, a series of lightning talks (called Geeky Google Tech Talks), and Cloud Study Jams.

“As Cloud is already a field that is in high demand, we thought our community could benefit from learning this for their own career success,” explains Milindi.

Embracing school spirit and multiple learning opportunities

To ensure students would be prepared for workshops on Artificial Intelligence, Machine Learning, and robotics, the club held beginner-friendly programs that introduced students to the topics. They launched these in collaboration with RMIT's Society for Women in Information Technology, RMIT's Programming Club and additional partners and sponsors. The club’s HackVision hackathon, featured games, mentoring sessions, and workshops on topics such as public speaking, pitching ideas, ideation, UX/UI design, and software architecture.

With a wide variety of workshops that drew many attendees, the events and facilitators helped students with their technical queries throughout the semester. Some facilitators such as industry professional Thomas Frantz, Google Chrome Engineer, Jakub Młynarczyk, and Patrick Haralabidis of the Melbourne TensorFlow User Group stay connected with the students. Their connections and willingness to help solidifies the importance of mentorship. For example, one instructor led a machine learning workshop, while another one walked students through the complex algorithm-related questions asked during technical job interviews. Milindi herself led a live robotics demo where she programmed Nao Bots to dance to music and showed other students how to program the robots.

“These workshops helped them expand their tech skills to unexplored areas and help them feel confident at interviews. Our speakers showed students what is possible with these new technologies and machines. The time and effort spent on this event was truly rewarding seeing how much all the participants learned and experienced,” Milindi says.

Collage style image made up of images of GDSC RMIT members, memes, emojis, and graphics created by the group

Representation and professional development

Professional development remains important too, notes Milindi. “We here at GDSC RMIT have quite a diverse community, and we wanted to make sure our community was well prepared for the hardships of finding success.” She says, "as a female international student, I've had to apply to hundreds of companies when I was looking for internships, so I understand the struggle of applying for jobs, especially the frustration that comes with being rejected, in most cases, for reasons beyond your control, like visa issues.”

In this spirit, the club collaborated with RMIT's Society for Women in Information Technology to hold an event called "ROAR!” to empower female and underrepresented students. The event included an #IamRemarkable workshop with Googlers. Additionally, the club encouraged students to develop their resumes and LinkedIn profiles and offered a Personal Branding workshop, giving students multiple opportunities to showcase their talents and abilities.

Preparing a diverse student body for tech careers

Milindi expresses gratitude for having the opportunity to reinvigorate the community around RMIT University’s Google Developer Student Club. She relishes the ability to bring diverse students from many countries and gender identities together in the service of developing new skills, learning from each other, and ultimately landing jobs in the fast-growing tech industry.

"My amazing Core Team and I have given GDSC at RMIT University a new chapter and made a safe space for tech nerds of all ages, races, genders, and sexualities to feel welcome to be themselves and strive to be successful," Milindi says. "I am grateful for this fantastic opportunity to serve our community and help make the world a better place.”

For more information on this GDSC chapter visit the site here. If you’re a student and would like to join a Google Developer Student Club community, look for a chapter near you here, or visit the program page to learn more about starting one in your area.

Beta Channel Update for Desktop

The Beta channel has been updated to 98.0.4758.74 for Mac ,Windows and Linux.

A full list of changes in this build is available in the log. Interested in switching release channels? Find out how here. If you find a new issues, please let us know by filing a bug. The community help forum is also a great place to reach out for help or learn about common issues.

Srinivas SistaGoogle Chrome

Offers on Google Play: a new destination to find great deals

Since 2012, Google Play has been a one-stop shop for discovering and enjoying your favorite apps, games and digital content. This week we’re launching “Offers” — a new tab in the Google Play Store app to help you discover deals in games and apps across travel, shopping, media & entertainment, fitness, and more. The rollout is underway and it will be available to more people in the United States, India and Indonesia over the coming weeks, and more countries later in 2022.

Sections like “Offers for apps you might like” help you easily find deals that are relevant to you. We’re partnering with developers of some of the top apps and games on Google Play to add new, fresh deals every day. As Allison Boyd of Strava explains, “Offers is a win-win. We get an additional touchpoint with people to share information about a valuable promotion or update, and people can easily redeem the offer by opening the Strava app from the Offers tab.”

Find limited time deals, in-app rewards and more

Here are a few of the deal types that you can expect to find over time:

  • Sales on games and in-game items: find limited-time deals on magic orbs, tokens, and more.
  • Rewards and bundled offers: see what apps are offering you free delivery, free rides, and other rewards.
  • Discounts on movies and books: find the latest sales on movies and books to rent or buy.
  • Try something new: browse apps that are offering 30 days free, and other extended trials at no cost.

When you see Offers in the bottom bar of the Google Play Store app on your Android mobile device, be sure to check it out.

How Divya believed and bet on herself to get to Google

Welcome to the latest edition of “My Path to Google,” where we talk to Googlers, interns and alumni about how they got to Google, what their roles are like and even some tips on how to prepare for interviews.

Today’s post is all about Divya Gonnabathula, a Client Success Acceleration manager based in Hyderabad, India who’s always gravitated toward new experiences — including here at Google!

What do you do at Google?

I’m a Client Success Acceleration manager leading a team of digital marketing strategists. We collaborate with sales teams in the UK to help our customers succeed using Google Ads. As a manager, I also make sure I spend time with my reports in team and individual meetings — I’m passionate about leading with respect.

What’s something about you that might surprise us?

I consider myself a nomad. Over the years, I’ve lived in many different parts of India. I was born in Vijayawada, went to school in Kolkata, attended college in Bhubaneswar and worked in many cities around the country. Thanks to these experiences, I can easily connect with people from different cultural backgrounds and have a keen interest in languages — I speak six and I’m picking up a seventh!

Why did you apply to Google?

Like so many others, Google’s products and services are part of my daily routine. I can’t imagine a day without searching for things online, watching YouTube or checking emails. The opportunity to create such impact at scale inspired me to look for open roles on the Google Careers website. Once I found one that matched my career goals, I applied.

What was the journey to your current role like?

I’m a sales and marketing professional, and I’ve spent most of my career in the Indian consumer goods industry. A few years ago, I jumped to consumer tech — which led me to Google. By the time I started, I was eight months pregnant. Joining a new organization always comes with a lot of change — but the thought of going on maternity leave in just one month, combined with onboarding remotely, was pretty overwhelming. Luckily, Google made things easy. My recruiter, manager and onboarding “buddy” helped me ease into the company, creating a custom onboarding plan and encouraging me to build a support network on my team.

How did you prepare for your interviews?

I started by brainstorming questions the interviewer might ask and practicing my responses. I also reflected on my previous roles, including my strengths, management experiences and leadership philosophies. I brushed up on my knowledge of Google Ads products, too.

What’s one of your favorite things about working at Google?

My teammates! The opportunity to work with some of the greatest minds in the industry is a very big motivator, because I believe we learn so much from the people around us. In order to bring my best self to work, it’s important that I have colleagues and managers who inspire me.

Any advice for your past self?

Don’t doubt yourself. I was in a marketing role when I applied for this sales position at Google, so I was worried that I wouldn’t be a good fit. But I realized through the hiring process that having diverse skills and experiences is actually really valuable.

Any tips for aspiring Googlers?

Throughout my Google interview process, I was determined to be myself. I made sure to show my personality in all of my interviews and approached them as candid conversations with future colleagues and managers. This gave me a better understanding of what I was signing up for, and helped my interviewers see how I would really fit into the role.

Get ready for Smart Shopping and Local campaign upgrades

New year, same trend: Consumers continue to tap into an ever-growing number of places to get inspiration and find what they need. This past holiday season, 54% of shoppers used five or more channels, like video and social media, to shop over a two-day period.[ec0303]

Performance Max uses Google’s automation to help advertisers reach shoppers as they browse for things they love — online and in stores.[053cc9]It builds on Smart Shopping and Local campaigns to deliver the same foundational features, while adding brand new inventory and automation insights. Starting in April, you can begin upgrading your Smart Shopping and Local campaigns to Performance Max to access additional inventory and formats across YouTube, Search text ads and Discover.[d34fff]Based on early testing, advertisers who upgrade Smart Shopping campaigns to Performance Max see an average increase of 12% in conversion value at the same or better return on ad spend (ROAS).[7cb4f9]

Performance Max allows you to grow online, offline or omnichannel sales by unlocking all of Google’s ad inventory from a single campaign with a product feed. For example, you can transform your Video ads into your digital storefront and highlight your top products on YouTube — right where valuable customers are watching relevant video content every day.

Learn how to upgrade Smart Shopping and Local campaigns to Performance Max

Here’s a preview of the upgrade process so you know what to expect and how to plan ahead to set your campaigns up for success.[c7f14c]

Keep these key milestones in mind as you prepare for the upgrade:

  • Create new campaigns with Performance Max: You can continue using your existing Smart Shopping and Local campaigns until you’ve upgraded them to Performance Max later this year. For any new campaigns you create in Google Ads, start using Performance Max. You’ll access ad inventory that’s already available in Smart Shopping and Local campaigns plus new inventory and formats — including across YouTube, Search text ads and Discover.[d34fff]
  • Upgrade with a one-click tool: Starting in April, you can easily upgrade your Smart Shopping campaigns with a new “one-click” tool in Google Ads. You’ll be able to upgrade your Local campaigns with the tool starting in June. The tool gives you flexibility to upgrade specific campaigns or all of your campaigns at once.[076e64]Learnings from your existing campaigns will be used in your new Performance Max campaigns to maintain consistent performance.
  • Upgrade automatically: From July through September, your Smart Shopping campaigns will be automatically upgraded for you. Your Local campaigns will be automatically upgraded from August through September. You will no longer be able to create new Smart Shopping and Local campaigns once your existing campaigns are automatically upgraded. The automatic upgrade process will conclude by the end of September to ensure you’re well prepared to use Performance Max for the 2022 holiday season.

Leading e-commerce partners like Shopify, WooCommerce and BigCommerce are already working to support Performance Max so more merchants can benefit from new inventory and formats to connect with their customers. All merchants and partners will be able to upgrade their Smart Shopping and Local campaigns through the Google Ads API later this year. Learn more about the process in our developer blog.

Prepare for campaign upgrades with new resources

Tune in to our upcoming webinar in February, where product experts will demonstrate the one-click upgrade tool and dive into best practices for creating new Performance Max campaigns. They’ll also explain the new technology used in your campaign upgrades to help you get better results, faster.

Meanwhile, our updated Performance Max best practices guide has new recommendations geared towards achieving your online and offline goals. You can also explore the infographic below to understand how to use Performance Max with other Google Ads campaigns.

Infographic of how to use Performance Max together with other campaign types based on advertiser objectives: Lead Generation, Online Sales, Offline Sales (omnichannel) or Offline Sales (offline-only). The guidance is for advertisers to maximize performance in their Search campaigns first. At the bottom, it shows three options for how to grow performance across other channels. Advertisers can test Performance Max to access all Google Ads inventory from a single campaign, upgrade to Performance Max later this year if they’re currently using Smart Shopping or Local campaigns, and/or use Video, Display or Discovery campaigns if they want to increase performance on specific channels or reach specific audiences.

In the coming months, we’ll share tools and tips to help you at each milestone and make your upgrade process as smooth as possible. In the meantime, share your questions and comments and we’ll address them in our upcoming webinar. You can also follow @AdsLiasion to stay updated throughout the process.

Hash Code 2022 returns with a new look

Registration is officially open for Hash Code, Google’s team-based programming competition! On February 24, developers around the globe will team up to test their coding skills and have some fun. Whether you’re new to coding or a seasoned expert, there’s something for everyone. Here’s what you can expect from Hash Code 2022:

Compete on a new platform

Hash Code is excited to announce that participants will compete on a new platform — the same one used by Code Jam and Kick Start — so participants will all get the same experience across Google's Coding Competitions. You can test it out after you and your team register for the competition and try out the practice problem available at

Work like a software engineer

Hash Code problems are written by Google software engineers and are based on real engineering problems. Hash Code is a great way to build your engineering skills and learn something new. Because there is no right answer, your team can submit multiple solutions during a round. — which is exactly how our engineers work on a daily basis at Google.

Want a taste of what Hash Code problems look like? You can find all problems from previous years of Hash Code in the archive.

Team up with friends (or meet some new ones)!

In order to compete in the Qualification Round, you must be on a team of 2-4 people. If you don't have a team, don't worry! You can still register today and find teammates later using our Facebook group.

Amp up your resume

Hash Code is a great way to grow your coding skills and experience how real software engineers solve real problems. Every team that submits a solution and scores one point in the Qualification Round will receive a certificate in their Coding Competitions profile. This certificate can be posted to social media, added to a job profile or used for your resume.

Hash Code 2022 kicks off with the Qualification Round on Thursday, February 24 at 9:30 a.m. PT and wraps up with the World Finals in April. Register at by February 23 and spread the word!