AI in Manufacturing, Google Vertex AI, and Session-Based Recommendations
An increasing number of manufacturers are implementing AI solutions in their core processes, especially in Europe.
Context
As the access to Artificial Intelligence solutions grows, industries across the globe are increasingly adopting these novel technological solutions. While there is still quite a way to go for some high-impact use-cases due to ethical, governance, and systems considerations, some industries are massively adopting straight-forward models that have already proven their value.
What's new
According to research from CapGemini, some 51% of European manufacturers are implementing AI solutions. When compared to 30% in Japan and only 28% in the United States, it seems that Europe is significantly ahead in AI Adoption.
The report's author, Pascal Brosset explains the importance of this adoption by stating that "AI in manufacturing is a game-changer". The reason behind this is three-fold: (1) the technology applied to manufacturing use-cases is extremely mature, (2) there are no important ethical problems as the algorithms don't use personal (or even any human) information, and (3) the production floor is a somewhat controlled environment with clear and concise production objectives.
Hence, it comes as no surprise that the three top use-cases in the industry are:
- Intelligent predictive maintenance
- Product quality control
- Demand planning
All of these applications combine a golden combo of AI-maturity:
- Clear business benefits
- Relatively straight-forward models
- Data availability
- Potential interpretability
Why it matters
Manufacturing data is a great fit for Artificial Intelligence applications. The data usually arrives in predictable quantities at predictable times as it is in essence generated by the manufacturing processes themselves. Above that, models do not go stale as quickly as in other applications such as Recommender Systems due to the fact that the data distribution is inherently bound by the production processes themselves, not by the behavior or preferences of human customers.
Moreover, the impact AI use-cases can have in manufacturing are astounding. By automating some processed and assisting shop workers with others, companies are able to save an immense amount of money, all while increasing the efficiency and quality of their production lines. The key is using computer-aided algorithms that are able to detect patterns in data undetectable by humans, and implementing these at the right steps of your production process.
What's next
As the technology matures and regulations become a reality, we expect these trends to spread to other industries.
Are you keeping up with this industry trend? Our engineers are democratizing Artificial Intelligence for the good of society and businesses. Their high technical expertise and business know-how is helping companies across Switzerland and Europe adopt AI systems more intelligently. Specifically, in the manufacturing industry, our ML solutions have helped clients increase monthly production line running time by 10%, and significantly reduced need for manual machine check-ups and thus operational costs.
For more information about a past Predictive Maintenance project performed by Visium for Nestlé, click here.
Google Cloud launches Vertex AI, a cloud platform simplifying deployment and maintenance of production Machine Learning models
Context
Most ML projects start off with a Proof-of-Concept or Pilot phase. A scaled-down implementation of the desired application is developed and tested. This phase is essential to evaluate the technical feasibility, different potential models, business insights, and strategic elements of the solution.
As this phase often entails lower technical requirements than production-grade systems, it can be very difficult to scale these systems for small and large companies alike. In fact, bringing Machine Learning systems into production is an extremely complex task. The ML Operations space contains a vast landscape of multiple and diverse tooling software. For the uninitiated, the road between experimentation to production is a long a perilous one.
What's new
Google Cloud has recently launched Vertex AI, a unified and fully-managed ML platform. By providing pre-trained ML models developed by Google Research and MLOps tooling for managing data and models at scale, Google Cloud believes Vertex AI can help you deploy models with 80% less code than custom solutions.
One of the key features of Vertex AI is its unified UI and API. From training and comparing models using AutoML to integrating with open source frameworks such as TensorFlow, PyTorch, and scikit-learn, Vertex AI brings together all ML services in the same place.
Above that, Vertex AI users are able to utilize Google's pretrained APIs for vision, video, and natural language. Use-cases such as speech-to-text, translation, and document OCR can be handled directly on the Google Cloud platform.
A statement by Andrew Moore, VP of Cloud AI at Google Cloud:
We had two guiding lights while building Vertex AI: get data scientists and engineers out of the orchestration weeds, and create an industry-wide shift that would make everyone get serious about moving AI out of pilot purgatory and into full-scale production.
We are very proud of what we came up with in this platform, as it enables serious deployments for a new generation of AI that will empower data scientists and engineers to do fulfilling and creative work.
For more information about what you can do with Vertex AI, Google has posted an informative video. Additionally, several use cases such as Data readiness, Model serving, Model monitoring, Hyperparameter tuning, and more are available here. Finally, to have a look at Vertex AI's extensive list of features, click here.
Why it matters
Several high-profile companies have already been given access to the platform. Among them is Essence, a media agency that is part of London-based global advertising and communications giant WPP. Essence uses Vertex AI to update their models to keep pace with the rapidly-changing world of human behaviors and channel content.
Another early-access company is ModiFace, a part of L’Oréal that uses AR and AI in the beauty industry. ModiFace is currently using Vertex to train it's ML models for all of its services. For example, the company’s skin diagnostic service is trained on thousands of images from L’Oréal’s Research & Innovation arm using Vertex AI.
Given Google's customer base across the globe, companies of all sizes are likely going to start adopting this new ML platform in the coming months.
What's next
Vertex AI is indubitably going to make waves in the industry. In fact, many companies are already using the Google Cloud platform for renting compute, training models, and much more. For them, the shift to Vertex AI will be quite straightforward.
However, Vertex AI is a new player in an existing market containing a plethora of ML tooling solutions. Only the future will tell exactly how it will fit into existing pipelines of more established and data-mature companies.
An applied research report from Cloudera Fast Forward tackles a recent hot topic, session-based recommender systems
Context
As a modern-day owner or manager of a digital marketplace, being able to recommend a product of interest to a customer based on their personal preferences is essential. However, due to privacy restrictions and recent GDPR regulations, it is more difficult than ever to identify a user that has not explicitly logged in. How do you stay afloat in an extremely competitive space if you don't have access to user profiles and their historical preferences?
What's new
Cloudera, an enterprise data platform built on open source technology, has recently published an applied research report about session-based recommender systems. This type of recommendation is solely based on the customer's interactions in an ongoing digital shopping session. It should come as no surprise that companies around the world are using this technology to better guide their customers when browsing.
If you're not convinced, I urge you to (1) open an incognito window, (2) go on Amazon, (3) search for and click on a couple of desk chairs, and (4) reload the front-page and scroll down to the "Gift ideas inspired by your shopping history" section. As you can see, recommendations galore about office supplies such as desk mats, monitors, and of course desk chairs.
In the large scheme of things, recommender systems can be split into two categories, content-based and collaborative filtering-bases methods. While the former uses the user's preferences and the product's content (i.e. features) to find the best match, collaborative filtering relies on the concept that similar users will most likely enjoy similar products.
Unfortunately, you may not have access to the information required to make these predictions, notably the user preferences. Above that, user preferences change all the time, and depending on the specific use-case, the trends may have shifted too much, making your recommendations useless.
This is where session-based recommender systems come in. These systems rely on customer's interactions in the current session (i.e. from the moment they open your digital marketplace) to adapt the marketplace's recommendations in real-time.
These types of recommendations have risen in popularity over recent years. Particular use-cases include recommendations for rentals, music, products, apartment listings, and so on.
For more information about session-based recommender systems including the technical Machine Learning details and some experiments on a UK boutique's e-commerce dataset, navigate to Cloudera's article. The code is evidently available, right here.
Why it matters
The big advantage of session-based recommendations is their independence of historical user data. This is extremely useful as privacy regulations are becoming more stringent and internet browsers are becoming increasingly wary about giving their information away (as they should be).
Moreover, the widespread presence of data and the rise of real-time Machine Learning applications respectively allow for the efficient development and deployment of such solutions.
What's next
The adoption of session-bases recommender systems is bound to increase in the near future. Not only does this technique respect customer's privacy, it also helps the customer navigate your digital marketplace with more ease.
Visium's vast expertise with recommendation systems and digital marketplaces makes it the ideal partner for the development and deployment of your Session-based Recommender System. For a detailed list of Visium's services, click here.
