Generative AI continues to develop at a blistering pace, making it more important than ever that organizations have access to enterprise-ready capabilities to help them leverage this disruptive technology. 

Harnessing the power of decades of Google’s research, innovation, and investment in AI, Google Cloud continues to make generative AI available with baked-in security, data governance, and scalability across the board. 

To this end, last month, we announced the general availability of Generative AI support on Vertex AI, giving our customers the ability to access powerful foundation models from Google Research and tools for customizing and applying them. 

Today we are announcing the general availability (GA) of four important foundation models for Vertex AI. These include Imagen, PaLM 2 for Chat, Codey, and Chirp. For each of these models, organizations can access APIs on Model Garden and do prompt design and tuning on Generative AI Studio.

• Imagen includes four key features:
  • Image generation for creating studio-grade images at scale
  • Image editing to edit generated or existing images via text prompts 
  • Image captioning for creating captions of images at scale
  • Visual Question & Answering (VQA) for interacting with, analyzing, and explaining images
• PaLM 2 for Chat follows the general availability of PaLM 2 for Text in June 
• Codey supports code generation, completion, and code chat
• Chirp supports multilingual Speech AI 

We’re also announcing Multimodal Embeddings API in preview, which lets customers combine the power of Vertex AI’s generative AI models with their proprietary data, to generate embeddings, or interchangeable vector representations, of their text and image data. These capabilities can enable data science teams to deliver a variety of downstream tasks such as image classification, content recommendations, and visual search. 

In this blog post, we’ll explore what your organization can do with these powerful models and how Vertex AI provides the enterprise-ready capabilities you can use to get up and running with generative AI. 

Helping to drive enterprise value from Generative AI models

Powerful models are the foundation of generative AI, but the software, tools, and infrastructure that surround these models are equally important for enterprise adoption. Organizations face challenges not only accessing these models, but also integrating AI while maintaining protection over intellectual property, adhering to regulations around data security and privacy, and ensuring models and applications are safe to use. Many organizations also want to use generative AI without incurring large costs or managing huge clusters.

We help address these challenges head-on with Vertex AI’s platform capabilities for scalable application integration, purpose-built AI infrastructure, secure and private data customization, and responsible use of this technology. 

Let’s see how each of these pillars can help your organization. 

Access models to build production-ready generative applications 
Vertex AI can make it easy to access foundation models, as today’s model announcements attest. While models are an inextricable part of generative AI, the software that helps enterprises use this technology is equally important—which is why Vertex AI also offers a range of tools for tuning, deploying, monitoring, and maintaining models, so you can build differentiated applications using your own data. 

Turning to today’s announcements, in May we announced Imagen, our foundation model for image generation. Now, we are excited to announce Imagen is generally available with an allowlist (i.e., approved access via your sales representative), letting onboarded customers start using image generation and editing capabilities. Visual Q&A and Captioning for production workloads are also generally available for all customers. Visual Q&A provides new ways to engage with image-based data like retail products or image libraries. This new capability can give you answers to questions about an image, helping you analyze large amounts of data quickly, and it can even help the visually impaired understand images or graphs that they wouldn’t be able to otherwise. Captioning, meanwhile, can make it easy to generate relevant descriptions for your images. Captions can help with indexing and searching, as well as assigning image descriptions to product listings on eCommerce websites. 

“Imagen is beginning to power key capabilities within Omni, Omnicom’s open operating system, that will enable 17,000+ trained and certified users to create audience-driven customized images in minutes. Imagen has been instrumental in offering a scalable platform for image generation and customization. Integrating it into our platform allows us to expand the scope of audience-powered creative inspiration, at a scale that wasn’t previously possible,” said Art Schram, Annalect Chief Product Officer at Omnicom. “We’re starting to adopt the latest features like styles and fine tuning, and engineering data-driven prompts. We look forward to continuing to provide our users relevant visual inspiration in a responsible way.”

“The latest improvements in Imagen’s product preservation capabilities are a perfect match for Typeface’s focus on personalized AI for brands,” explained Vishal Sood, Head of Product at Typeface. “By combining Google Vertex AI’s Imagen with Typeface’s brand-personalized AI, we are able to help enterprises to create 10x personalized content in a fraction of time.”

Google Shopping recently built an application called Product Studio using Imagen on Vertex AI. Product Studio can enable merchants to create rich product images quickly and easily, at a fraction of the time it takes to do professional product photo shoots. “We’re excited about the feedback we’re getting from merchants in our early pilots, who say that Product Studio, which leverages Imagen on Vertex AI, helps them generate and publish lifestyle product photos directly to their product catalogs,” says Jeff Harrell, Google’s Senior Director of Product Management for Merchant Shopping. 

Announced in May, PaLM 2 is a family of models that power dozens of Google products, including Bard and Duet AI in Google Cloud. With the PaLM 2 for Chat model, now generally available, you can leverage Google’s PaLM’s variety of abilities for multi-turn chat applications, such as shopping assistants, customer support agents, and more. 

ThoughtSpot, provider of a widely-adopted business intelligence platform, is using PaLM 2 to build a new feature in ThoughtSpot for Google Sheets called “AI Explain,” which can instantly generate explanations of charts, visuals, and anomalies, and will launch new conversational AI and ML-enabled predictive forecasting capabilities into its analytics platform.

With Codey, your organization’s developers can accelerate a wide variety of coding tasks, helping to empower them to work efficiently and close skills gaps. The model enables not only code completion and code generation capabilities, but also chat to help with debugging, documentation, learning new concepts, and more. Since launching in preview in May, we’ve added additional programming languages including Go, Google Standard SQL, Java, Javascript, Python, and Typescript. We’ve also improved the quality of code responses and increased serving capacity, enabling your developers with the right tools to enter the era of generative engineering.  

“Security and privacy are key to incorporating AI into the software development lifecycle,” said David DeSanto, Chief Product Officer at GitLab. “GitLab leverages Vertex AI to deliver new, AI-powered features with a privacy-first approach, including the ability to run our own models and leverage Codey foundation models built on top of PaLM 2. The GitLab DevSecOps platform empowers organizations to harness the benefits of AI for faster software delivery, while ensuring their data, intellectual property, and source code are protected.”

Originally released in May in preview, Chirp is a version of our 2 billion-parameter speech model, which was trained on millions of hours of audio and supports over 100 languages. Chirp achieves 98% accuracy on English and relative improvement of up to 300% in languages with less than 10 million speakers. Whether the use case involves customer support, transcriptions, or voice control, Chirp can help your organization communicate with customers and constituents inclusively, by engaging audiences in their native languages. 

Last but not least, our Multimodal Embeddings API, now in preview, can unlock an array of new applications, such as image and text-based recommendations, by enabling the processing of text and images interchangeably. This capability complements our Text Embeddings API, which became generally available in June, and remains a recommended choice for those with fully text-based use cases. Multimodal Embeddings API makes it possible to categorize images and text together and can be crucial for use cases like retail recommendation systems that can provide relevant outputs from both images of products and text descriptions.

Match generative AI with infrastructure 
Beyond access to models and tools for building generative AI apps, you need infrastructure to make sure your apps can scale and reliably perform — ideally without running into daunting compute costs or management overhead that distracts your technical talent from building innovative products. Google Cloud offers the choice and power to run smaller models running finite tasks at the lowest latency levels, as well as to run large models capable of cutting-edge experiments. 

As our large language model customers are looking to scale up their projects and applications using our models, they often need assurances that their requests will be serviced with acceptable performance. This is especially critical for delivering real-time applications where customer service is paramount. Starting in August, Vertex AI will support provisioned, dedicated generative AI capacity that can deliver guaranteed throughput. This feature can be especially beneficial to customers who have a high volume of sustained workloads.  

Leverage generative AI while protecting data and privacy 
One capability enabled by Google Cloud is the ability to customize models using your own data. Vertex AI can help customers keep their data protected, secure, and private. When a company tunes a foundation model in Vertex AI, private data, model outputs, and prompts can be kept private, and they are never used in the foundation model training corpus. We recently published a whitepaper, “Adaptation of Large Foundation Models,” which outlines how we help protect customer data. 

Auditability and compliance are essential to helping ensure the security and privacy of customer data. We also engage in comprehensive GDPR privacy efforts, including our transparency commitments for customer data usage and the support for our customer’s Data Protection Impact Assessments (DPIAs). Now, we’re excited to support HIPAA compliance for many of our generally available models on Vertex AI, so that healthcare and life science customers with whom we have a Business Associate Agreement can run workloads with Protected Health Information (PHI) data on Google Cloud. 

Innovate responsibly 
Our AI Principles put beneficial use, user safety, and avoidance of harms above business outcomes and are embedded in how we develop our AI products. We’ve conducted extensive reviews on our generative AI products to identify potential risks and have developed guardrails to mitigate these impacts. For example, to address concerns around safety, we’ve implemented safety filters for bias, toxicity, and other harmful content. We also equip our customers with the tools they need to help reduce risk within their applications and provide recommendations to help navigate responsible AI. 

Bring the power of generative AI to your organization

With both a wide selection of foundation models and extensive, enterprise-grade platform capabilities, Vertex AI continues to unlock ways for your business or organization to access foundation models, tune them on your proprietary data, and leverage them for differentiated apps and digital experiences. To take the next step, visit our product page or reach out to our sales representatives to gain access to our latest capabilities.

Over the past two decades, consumers have become accustomed to receiving personalized recommendations in all facets of their online life. Whether that be recommended products while shopping on Amazon, a curated list of apps in the Google Play store, or relevant videos to watch next on YouTube. In fact, in a Verge article “​​How YouTube perfected the feed: Google Brain gave YouTube new life,” the Google Brain team reveals how their recommendation engine has impacted the platform with “more than 70 percent of the time people spend watching videos on the site being driven by YouTube’s algorithmic recommendations” thereby increasing time spent on the platform by 20X in three years. 

It’s become clear that personalized recommendations are no longer a differentiator for an organization but rather something consumers have come to expect in their day-to-day experiences online. So what should you do if you are behind the curve and want to get started or simply want to improve upon what you already have? While there are all sorts of techniques, from content-based systems to deep learning methods, our goal in this recommender-focused blog series is to demystify three available approaches to building recommendation systems on Google Cloud: Matrix Factorization in BigQuery Machine Learning (BQML),  Recommendations AI, and deep retrieval techniques available via the Two-Tower built-in algorithm.

One of these approaches can be used to meet you where you are in your personalization journey, no matter if you are just starting or if you are well into it. This first blog post will introduce our three approaches and when to use them. 

What is Matrix Factorization and how does it work?

Collaborative filtering is a foundational model for building a recommendation system as the input dataset is simple and the embeddings are learned for you. How does Matrix factorization fit into the mix you might be wondering?  Matrix factorization is simply the model that applies collaborative filtering. BQML enables users to create and execute a matrix factorization model by using standard SQL directly in the data warehouse. 

Collaborative filtering begins by creating an interaction matrix. The interaction matrix represents users as a row and items as columns in your dataset. This interaction matrix often is sparse in nature as not all users will have interacted with many items in your catalog. This is where embeddings come into play. Generating embeddings for users and items not only allows you to collapse many sparse features into a lower dimensional space but they also allow you to derive a similarity measure so that similar users/items fall nearby in the embedding space. These similarity measures are key as collaborative filtering uses similarities between users and items to make the end recommendations. The underlying assumption being that similar users will like similar items whether that be movies or handbags. 

What’s required to get started?

To train a matrix factorization model you need a table that includes three input columns: user(s), item(s), and an implicit or explicit feedback variable (e.g., ratings is an example of explicit feedback). With the base input dataset in place, you can then easily run your model in BigQuery after specifying several hyperparameters in your CREATE MODEL SQL statement. Hyperparameters are available to specify the number of embeddings, the feedback type, the amount of L2 regularization applied and so on.    

Why use this approach and who is it a good fit for?  

As mentioned earlier, Matrix Factorization in BQML is a great way for those new to recommendation systems to get started. Matrix factorization has many benefits: 

• Little ML Expertise: Leveraging SQL to build the model lowers the level of ML expertise needed
• Few Input Features: Data inputs are straightforward, requiring a simple interaction matrix
• Additional Insight: Collaborative filtering is adept at discovering new interests or products for users 

While Matrix Factorization is a great tool for deriving recommendations it does come with additional considerations and potential drawbacks depending upon the use case. 

• Not Amenable to Large Feature Sets: The input table can only contain two feature columns (e.g., user(s), item(s)). If there is a need to include additional features such as contextual signals, Matrix factorization may not be the right method for you.  
• New Items: If an item is not available in the training data, the system can’t create an embedding for it and will have difficulty recommending similar items. While there are some workarounds available to address this cold-start issue, if your item catalog often includes new items, Matrix factorization may not be a good fit.  
• Input Data Limitations: While the input matrix is expected to be sparse, training examples without feedback can cause problems. Filtering for items and users that have at least a handful of feedback (e.g., ratings) examples can improve the model. More information on limitations can be found here. 

In summary, for users with a simplified dataset looking to iterate quickly and develop a baseline recommendation system, Matrix Factorization is a great approach to begin your personalization AI journey. 

What is Recommendations AI and how does it work?

Recommendations AI is a fully managed service which helps organizations deploy scalable recommendation systems that use state-of-the-art deep learning techniques, including cutting-edge architectures such as two-tower encoders, to serve personalized and contextually relevant recommendations throughout the customer journey.

Deep learning models are able to improve the context and relevance of recommendations in part because they can easily address the previously mentioned limitations of Matrix Factorization. They incorporate a wide set of user and item features, and by definition they emphasize learning successive layers of increasingly meaningful representations from these features. This flexibility and expressivity allows them to capture complex relationships like short-lived fashion trends and niche user behaviors. However, this increased relevance comes at a cost, as deep learning recommenders can be difficult to train and expensive to serve at scale. 

Recommendations AI helps organizations take advantage of serving these deep learning models and handles the MLOps required to serve these models globally with low latency. Models are automatically retrained daily and tuned quarterly to capture changes in customer behavior, product assortment, pricing, and promotions. Newly trained models follow a resilient CI/CD routine which validates they are fit to serve and promotes them to production without service interruption. The models achieve low serving latency by using a scalable approximate nearest neighbors (ANN) service for efficient item retrieval at inference time. And, to maintain consistency between online and offline tasks, a scalable feature store is used, preventing common production challenges such as data leakage and training-serving skew.  

What’s required to get started?

To get started with Recommendations AI we first need to ingest product and user data into the API:

• Import product catalog: For large product catalog updates, ingest catalog items in bulk using the catalogItems.import method. Frequent catalog updates can be schedule with Google Merchant Center or BigQuery
• Record user events: User events track actions such as clicking on a product, adding items to cart, or even purchasing an item. These events need to be ingested in real time to reflect the latest user behavior and then joined to items imported in the product catalog 
• Import historical user events: The models need sufficient training data before they can provide accurate predictions. The recommended user event data requirements are different across model types (learn more here)

Once the data requirements are met, we are able to create one or multiple models to serve recommendations: 

• Determine your recommendation types and placements:  The location of the recommendation panel and the objective for that panel impact model training and tuning. Review the available recommendations types, optimization objectives, and other model tuning options to determine the best options for your business objectives.
• Create model(s): Initial model training and tuning can take 2-5 days depending on the number of user events and size of the product catalog 
• Create serving configurations and preview recommendations: After the model is activated, create serving configurations and preview the recommendations to ensure your setup is functioning as expected before serving to production traffic

Once models are ready to serve, consider setting up A/B experiments to understand how newly trained models impact your customer experience before serving them to 100% of your traffic. In the Recommendations AI console, see the Monitoring & Analytics  page for summary and placement-specific metrics (e.g., recommender-engaged revenue, click-through-rate, conversion rate, and more).

Why use this approach and who is it a good fit for?  

Recommendations AI is a great way to engage customers and grow your online presence through personalization. It’s used by teams who lack technical experience with production recommendation systems, as well as customers who have this technical depth but want to allocate their team’s effort towards other priorities and challenges. No matter your team’s technical experience or bandwidth, you can expect several benefits with Recommendations AI: 

• Fully managed service: no need to preprocess data, train or hypertune machine learning models, load balance or manually provision you infrastructure – this is all taken care of for you. The recommendation API also provides a user-friendly console to monitor performance over time. 
• State-of-the-art AI: take advantage of the same modeling techniques used to serve recommendations across Google Ads, Google Search, and YouTube. These models excel in scenarios with long-tail products and cold-starts users and items
• Deliver at any touchpoint: serve high-quality recommendations to both first-time users and loyal customers anywhere in their journey via web, mobile, email, and more
• Deliver globally: serve recommendations in any language anywhere in the world at low-latency with a fully automated global serving infrastructure
• Your data, your models: Your data and models are yours. They’ll never be used for any other Google product nor shown to any other Google customer

For users looking to leverage state of the art AI to fuel their recommendation systems but need an existing solution to get up and running more quickly, Recommendations AI is the right solution for you. 

What are Two Tower encoders and how do they work?

As a reminder, in recommendation system design, our objective is to surface the most relevant set of items for a given user or set of users. The items are usually referred to as the candidate(s) where we might include information about the items such as the title or description of the item, other metadata about the item like language, number of views, or even clicks on the item over time. User(s) are often represented in the form of a query to a recommendation system where we might provide details about the user such as the location of the user, preferred languages, and what they have searched for in the past.   

Let’s start with a common example. Imagine that you are creating a movie recommendation system. The input candidates for such a system would be thousands of movies and the query set can consist of millions of viewers. The goal of the retrieval stage is to select a smaller subset of movies(candidates) for each user and then score and rank order them before presenting the final recommended list to the query/user.

The retrieval stage is able to refine our list of candidates by encoding both the candidate and the query data so they share the same embedding space. A good embedding space will place candidates which are similar to one another closer together and dissimilar items/queries farther apart in the embedding space.

Once we have a database of query and candidate embeddings we can then use an approximate nearest neighbor search method to then generate a list of final “like” candidates, i.e. find a certain number of nearest neighbors for a given query/user and surface final recommendations.

What’s required to get started?

At the most basic level, in order to train a two-tower model you need the following inputs:

• Training Data: Training data is created by combining your query/user data with data about the candidates/items. The data must include matched pairs, cases where both user and item information is available. Data in the training set can include many formats from text, numeric data, or even images. 
• Input Schema: The input schema describes the schema of the combined training data along with any specific feature configurations.

Several services within Vertex AI have come available that complement the existing Two-Tower built-in algorithm and can be leveraged in your execution: 

• Nearest Neighbor (ANN) Service: Vertex AI Matching Engine and  ScANN provide a high-scale and low-latency Approximate Nearest Neighbor (ANN) service so you can more easily identify similar embeddings.
• Hyperparameter Tuning Service: A hyperparameter tuning service such as Vizier can help you identify the optimal hyperparameters such as the number of hidden layers, the size of the hidden layers, and the learning rate in fewer trials. 
• Hardware Accelerators: Specialized hardware, such as GPUs or TPUs, can be valuable in your recommendation system to help accelerate experiments and improve the speed of training cycles. 

Why use this approach and who is it a good fit for?  

The Two-Tower built-in algorithm can be considered the “custom sports car” of recommendation systems and comes with several benefits: 

• Greater Control: While Recommendations AI uses the two-tower architecture as one of the available architectures it doesn’t provide granular control or visibility into model training, example generation, and model validation details. In comparison, the Two-Tower built in algorithm provides a more customizable approach as you are training a model directly in a notebook environment. 

• More Feature Options: The Two Tower approach can handle additional contextual signals ranging from text to images. 
• Cold Start Cases: Leveraging a rich set of features not only enhances performance but also allows the candidate generation to work for new users or new candidates.

While the Two-Tower built in algorithm is an excellent and best-in class solution for deriving   recommendations, it does come with additional considerations and potential drawbacks depending upon the use case.

• Technical ML Expertise Required: Two tower encoders are not a “plug and play” solution like the other approaches mentioned above. In order to effectively leverage this approach, appropriate coding and ML expertise is required. 
• Speed to Insight: Building out a custom solution via two-tower encoders may require additional time as the solution is not pre-built for the user.   

For users looking for greater control, increased flexibility, and have the technical chops to easily work within a managed notebook environment – the two-tower built in algorithm is the right solution for them. 

What’s next?

In this article, we explored three common methods for building recommendation systems on Google Cloud Platform. As you can see thus far, there are alot of considerations to take into account before choosing a final approach. In an effort to help you align more quickly we have distilled the decision criteria down to a few simple steps (see below for more details).

In the next installments of this series, we will dive more deeply into each method, explore how hardware accelerators can play a key role in recommendation system design, and discuss how recommendation systems may be leveraged in key verticals. Stay tuned for future posts in our recommendation systems series. Thank you for reading! Have a question or want to chat? Find authors here – R.E. [Twitter | LinkedIn], Jordan [LinkedIn], and Vaibhav [LinkedIn].

Acknowledgements

Special thanks to Pallav Mehta, Henry Tappen,Abhinav Khushraj, and Nicholas Edelman for helping to review this post. 

References

• Announcing ScaNN: Efficient Vector Similarity Search
• Factorization Machines for Item Recommendation with Implicit Feedback Data

As we think about economic recovery from COVID-19—both inside Google and outside through working with Google Cloud customers—we’ve made many important observations. Among them is the recognition that the ways software developers and IT practitioners work together will shift in the post COVID-19 world. Our economic recovery today will look different than past recoveries, and on a fundamental level, the way we innovate will be different than it’s ever been before.

Right now, we’re entering a new phase of cloud computing, where businesses have shifted from making tactical infrastructure decisions, to making larger IT decisions with an eye towards enabling transformation throughout the company. Data, and what we can do with that data, is key to this transformation. And how companies put data in the hands of every employee to help catalyze transformation and solve the most important and impactful opportunities in their industries is at the core.  

A recent Google-commissioned study by IDG highlighted the role of data analytics and intelligent solutions when it comes to helping businesses separate from their competition. The survey of 2,000 IT leaders across the globe reinforced the notion that the ability to derive insights from data will go a long way towards determining which companies win in this new era.

Data analytics and intelligence were prioritized during COVID-19

The results of the IDG study show a separation amongst those organizations that embrace the capabilities of today’s data analytics and AI/ML tools and those that do not. When COVID-19 hit, many organizations cancelled IT initiatives, with 55% of respondents delaying or cancelling at least one technology project. However, 32% of respondents accelerated or introduced initiatives around building out or improving the use of data analytics and intelligence. IT leaders realize how critical data is to their future success, even when resources are scarce.

Digital-focused companies are faster to embrace advanced intelligence tools

Furthermore, enthusiasm for big data analytics, AI, and ML technologies is highest among companies who are further along in their digital transformation journeys. Fifty-four percent of companies who identify as “Fully digitally transformed” or “Digital native” are using or considering using these tools, vs. the global average of 37%. And, these same organizations are embracing the promise of AI more than their peers. Forty-eight percent felt that “Embedded AI across our full stack of cloud solutions will be critical” vs 39% of digital conservatives. These companies realize these digital tools enable them to be more resilient, agile, and prepared for whatever the future brings.

Companies are turning to cloud to maximize insights from data

As companies tap into the promise of data analytics and AI/ML, they are turning to cloud for help. When considering which cloud providers to work with, 78% of respondents said big data analysis is a “must have” or a “major consideration,” which placed this capability at the top of the list of consideration factors. This is not surprising, as cloud solutions address the most common pain points and barriers to innovation. Three of the respondents’ top four areas impeding innovation are addressed by cloud: Insufficient IT & developer skill sets (1st), security risks and concerns (2nd), and legacy systems and technologies (4th). Plus, cloud makes it easier to quickly launch a project, scale up or scale down, and pay for only what you use.

COVID-19 changed the very nature of business, and of IT. It forced IT leaders to decide where to put their scarce resources and big data analytics and AI/ML were, understandably, at the top of the list. To learn more about the findings, download the IDG report “No turning back: How the pandemic reshaped digital business agendas.”