Examples of AI use cases for climate action and clean energy and sustainable industry and agriculture.

Examples of AI use cases for transport & mobility and environment & oceans.

Semantic search from a business point of view and how it can bring value to your business processes.

Deploying a Scikit-learn Pipeline + LightGBM Classifier Model on Vertex AI for Explainable Online Predictions

We’ll look at different approaches to optimize transformer models for CPU inference.

Common hybrid NLP design patterns and look at example situations of when you should opt for which pattern.

A guide on how to choose the right speaker diarization framework for your use case.

How to approach uncertainty in machine learning : forecasting, regression and classification models.

ML6 advised and supported GfK on setting up an MLOps way-of-working, enabling GFK to bring insights to production in weeks instead of months, at much lower costs.

Identified gaps in the architectures for SLR/SLT by considering and exploiting existing state-of-the-art architectures.

Applications of using VQGAN combined with Discrete Absorbing Diffusion models.

Our overall thoughts & takeways on the summit and why we’re excited about Beam’s future.

We review and compare the previous SOTA generative models, from autoencoders to autoregressive models, and work our way to the discrete absorbing diffusion model.

Developing a DALL-E-based logo generator.

Detecting errors in transformers generated abstractive summaries

How we tackle information management at ML6 and how our tech radar helps us do this.

A blueprint for developing Vertex AI pipelines.

In this blogpost we will discover the complexity of pooling that hides behind its apparent simplicity.

Learn more about the similarities and differences between transformers and sentence-transformers.

What do we do though if we don’t have any or enough usable data yet to get started? One obvious option is to collect (more) data. However, today we want to go beyond that. We will look at three other dimensions that can be relevant to unblock ML use cases: data protection, external data, and synthetic data.

When deploying a Transformer model, the complexity increases since besides the model one also needs to consider how to deploy the tokenizer. Is it best to deploy the tokenizer on the server or to have the end users handle it and what is the best way to deploy the tokenizer along with the model on the server so that inference latency and costs are minimized?

An introduction of semantic search systems, why we need a domain transfer in dense retrieval search engines, the concept of generative pseudo labeling and how you can use generative pseudo labeling to adapt a dense retrieval model on a new domain for non-English texts.

Methods to tackle the notion of similarity for multivariate time series without making assumptions about the data itself.

Insights in how to run your Transformer in the cloud in a cost-effective way.

How to maximise the value of data to drive better performance of ML models while minimising the cost of doing so.

Together with SWISS KRONO, ML6 identified the different requirements for AI use cases. Few key requirements are a solid data architecture and infrastructure, multi-disciplinary teams and management/organisation support.

Building an AI driven model as a powerful tool in assisting journalists in their writing work by speeding up the summary creation as well as suggest alternative wordings. In that way, human creativity and AI can enhance each other’s complementary strengths and produce high quality results.

The premium feauture TelerouteRecommends allows customers to get instant recommendations of trade-lanes and solve problems carriers and freight forwarders might have when looking for truck-capacity or loads.

Abbreviated version of the AI act proposal : what you need to know.

A practical overview on information retrieval, in particular Semantic Search.

A practical guide to generating artificial data using Unity Perception.

Why we created an Ethical AI unit at ML6 and what we learned so far.

In this blogpost we take a look at what data augmentation is and how you can efficiently discover good data augmentation strategies for your model and your use case.

In light of a partnership with the organizations Reef Support and Fruitpunch AI, we designed an image segmentation algorithm. The main challenge here was the way the data is labelled, containing sparse pixel annotations created by domain experts. The blogpost covers on a high level how we solved this challenge, and how the challenge will be going forward.

A guide on how to integrate Darknet model training into your Vertex AI pipelines.

How we enabled consultative selling and increasing sales effectiveness by sending consultants to the right company at the right time, driven by the right information, so that we are more relevant to our customers’ and prospects’ needs.”

In this blogpost, we look at object detection using 3D ‘depth’ sensor data. We explain how two recent approaches, VoteNet and 3DETR, tackle the problem and benchmark their performance out of the box and after fine-tuning.

We will guide you through the technical architecture of TalkNet to provide you with a basic technical understanding. Then, we will show you how you can train this yourself based on your own data.

Funke Mediengruppe collaborated with ML6 to explore how to leverage intelligent technology for working with content, in order to enable new ideas and products.

Introduction blogpost to JAX : how it works, how it is different from Tensorflow/Pytorch and why we think every machine learning engineer should add it to its arsenal of tools.

This blopost explains the solution on how you can share TensorBoards securely with others in an easy way.

How to use SHAP, an open-source explainability toolkit, in your latest Huggingface Transformers pipeline.

A look at the new Vertex AI managed TensorBoard pricing model and provides an alternative, serverless and cost-efficient solution on GCP

Proper data augmentation is useful to increase the model performance to annotate a large amount of training data.

We explore how the connexion library facilitates API-first design with Python andtake a look at what the advantages are of API-first and compare with the opposite approach of “code-first”. Based on a benchmarking exercise and our values, we will dive deeper into why we are helping to maintain connexion.

We turn the wheel and make the most unproductive week, the most productive week of the year with our XMas Projects.

Log4Shell got security teams around the world on their toes In this blog post, we take a look at how and why we built a live threatmap on top of Google Cloud to detect and visualize cyber attacks and log4j exploits.

Learn about three ways that knowledge graphs and machine learning reinforce each other.

We accelerated the adoption of AI in Keypoint's platforms by automating the processing of their repair offers and doing fraud detection on claims images and weather data.

We recently tested a fancy deep learning model for tabular data and test the model now to predict the value of a certain house.

Our Head of Ethical AI , Caroline , takes a look at why explainability is important when developing AI solutions, and how to think about implementing Explainable AI within the entire ML project cycle.

Businesses in industries such as media, finance, jurisdiction and many more, tend to accumulate a large quantity of medium to long documents. Processing these documents can be challenging. We'll deep dive into the three main trends here.

Together with Plan international Belgium, we looked into a way to leverage the potential of AI to advance their mission of creating a better future for children. As an NGO, Plan international Belgium is reliant on funding, coming mainly in the form of either donations or child sponsorship.

Learn how to structure your data with knowledge graphs in this introduction blogpost to knowledge graphs and its business applications.

An easy way of serving a model in Vertex AI is by deploying a TensorFlow SavedModel using the pre-built containers. However, the lack of an official TensorFlow implementation for YOLO prevents us from directly using this methodology. In this blogpost, our ML Engineer Jérémy explains step by step one of the possible solutions for tackling this problem, which involves converting your Darknet weights to a TensorFlow SavedModel.

We had the honour to welcome Niels Rogge, ML Engineer at Huggingface. He shared with us how he started contributing to the Huggingface transformers repo, the lessons he learned along the way, and how anyone can start making contributions in various ways.

In this blogpost we’ll have a look at Anthos, a platform created by Google to easily manage multi and hybrid-cloud environments running Kubernetes clusters.

We are often asked how to detect very small objects in high resolution, i.e. very large images. A good example is finding objects in aerial images. The goal of this blog post is to demonstrate a practical approach to this problem using Slicing Aided Hyper Inference (SAHI).

From the ML6 community to the ML6 Maffia, our alumni network. How our mutual passion for technology, connects us for life.

In this blogpost, we explain our multi-article summarization approach, where we try to partially eliminate the black-box and make the summarization process more accessible to the user.Our summarization pipeline relies on the idea that sentences are the basic building blocks of any article.

ML6 helped Accolade Wines to implement a ML process to reduce wine loss.

In this blogpost, our ML engineer Juta will dive into the world of multimodal AI. Multimodal AI is able to reason about different modalities, such as text and images, at the same time and leads to more general understanding across domains.

Data Engineer Liam takes a look at how Pipelines can be deployed on the new and improved GCP product, Vertex AI Pipelines. The recently released tool takes a more managed approach to Pipelines in GCP, which promises to make running MLOps pipelines in the cloud easier than ever.

What is exactly that DNA that not only differentiates ML6 but also defines us?We thought it was time to explain this concept using examples and thus asked our HR Business Partner, Kaat van Doren, to give you some more insight into this.

Global warming is one of the biggest problems humanity is currently facing. Despite governmental efforts to reduce the output of greenhouse gasses, the emissions still rise every day. More and more, technology seems to be a promising tool to battle this issue. That is why we at ML6, decided to dive deep into some real-life use cases from four different domains where machine learning can be applied to help the planet.

The goal of this project was to increase the efficiency in notary offices by digitizing the wealth of knowledge inside their archives. These archives contain both text-layered documents as well as numerous scans of paperwork.

In our last blogpost, we have talked about what Trustworthy and Ethical AI is, and how we at ML6 help our clients build trust with their customers and employees through our Ethical AI Risk Assessment. In this post, we want to make it more practical, and demonstrate which dimensions and questions we might consider during such an assessment.

ML6 intern, Amir, studied ways in which to implement a drone that can perform follow-me behavior. The approach that he decided to investigate, is the possibility of using Reinforcement Learning (RL) as a means to allow an agent to train itself to perform the desired behavior.

Decision Forests are very powerful tools to be used as they are simple algorithms, and easy to understand. Recently, TensorFlow made a huge step by incorporating these models into their code stack. Machine learning expert, Pieter Coussement, shares in his latest blogpost how the TensorFlow Serving code was modified to accept TensorFlow Decision Forest models, and how to use this code as a custom docker image to serve the TF-DF models in three easy steps.

Our Machine Learning engineer, Jeremy, came across an interesting flaw that occurs when using Google Kubernetes Engine in combination with Terraform. Interested in finding out the flaw that Jeremy encountered and how we ended up fixing it? Read it in his latest Medium blogpost and learn how default settings can save money and trees.

We often get the question to forecast sales demand for customers. Recently we’ve done this successfully using Prophet. A new package called NeuralProphet builds upon this: it adds some neural networks to the mix. In this blog post we compare both time series forecasting models on a real-world dataset. We're predicting B2C sales volumes in retail to find out both methods' strengths and weaknesses.

To prove our dedication towards information security, our security program is certified against the ISO/IEC 27001:2017 standard. This internationally accepted standard provides requirements for establishing, implementing, maintaining and continually improving information security within an organization. ‍In this blogpost, we explain the why, what and how behind our ISO 27001 certification.

We take a look at how models can be deployed and used for prediction on the new and improved GCP product, Vertex AI. The platform was launched a couple of weeks ago and aside from adding additional features, existing features such as serving models for predictions have been updated and improved. We take a look at how endpoints are now a central part of model predictions and how containers have made deploying custom models very straightforward.

We explore the effectiveness of recently released face anonymization algorithms

March - a proptech player in commercial real estate - uses AI to disrupt the traditional ways of working in the commercial real estate market. The March Matching Engine uses a wealth of data to generate high-impact leads, allowing March’s real estate professionals to focus most of their time on the customer.

In this blogpost, we take a look at a promising new piece of hardware: the openCV AI Kit. An AI capable edge camera device with a sturdy build and solid 4K 12MP image and video quality. Exploring just how easy it can be to deploy a custom model on it, by showing you how to do just that. We look at the code and setup required to deploy a background segmentation model onto the device. Turning it into a webcam with background replacement built in, to give an example of the workflow.

Google just launched Vertex AI and our own ML in production experts are excited. We're showing you the story of GCP AI tooling and why Google decided to release Vertex now and where it's coming from. All of Google Cloud's AI and ML tools are now unified under a single banner and we saw some brand new products pop up as well. We're summarising the new Vertex platform tools and will follow up with more in-depth blogposts in the coming weeks. Check out what this means for your business, whether you're already GCP customer, interested to migrate to the new stack or just an enthusiast. This update is a big one, and we hope you'll follow us in exploring this new bastion of MLOps that Google Cloud has supposedly created!

In this blogpost, our Head of Labs, Jens Bontinck, outlines the benefits of multi-cloud, looks at the risks and complexity and shares his top three tips that make multi- cloud feasible.

In this blogpost we’ll set up a docker container to run an NVIDIA deepstream pipeline on a GPU or a Jetson and show you some tricks to get the most out of it! Deepstream’s documentation, guides and sample projects are few and far between, so this article aims to be a reference to get you from Deepstream zero to Deepstream-hey-I-can-make -a-pipeline-now and beyond!

In our latest interview, we discuss Ethical AI and why you should care about ethics. Our Chief Strategy Officer and AI Expert discuss some of the major challenges and how we tackle these at ML6.

In this blogpost, we outline four major technology trends we notice the AI R&D communities and industry are actively working towards sustainability in AI. We also share ML6’s key principles for AI development as we believe that implementing AI can be done sustainable. We actively follow and iterate on these development principles to push us and our customers to build responsible and sustainable technology.

For organizations that struggle to adopt AI. We help your business to create a common understanding of AI and its benefits.

Understanding and evaluating the ethical risks of an AI solution or idea through our unique approach.

AI can make a significant contribution to manufacturing by faster action and decision making, innovative product, services and material designs, improved efficiency, higher precision and lower costs. It is time for a change. You can enable AI without the need for high capital investments or major process redesigns. You just need to look for the right opportunities to be more productive and sustainable.

In this blog, we will uncover some pressing challenges within the Life Science manufacturing industry and how breakthroughs in Machine Learning techniques offer measurable returns.

In this blogpost, we tackle the challenge of picking the right camera for your computer vision projects by outlining a few key camera characteristics and features that will help you wit your camera decision.

The customer story of ASML and ML6 using Google Cloud to on-premises machine learning solutions.

In this blogpost, we’ll explain how AI can help solve typical challenges in the commercial model of life sciences companies.

What is Ethical and Trustworthy AI is and how do we at ML6 help our clients unlock the full value of AI and build trust with their customers, employees and society as a whole.

How we develop a secure development procedure to include security checks in multiple steps of the development process.

In this blogpost we cover the most important aspects of tackling a computer vision problem by labeling your own data.

MLOps is a trending topic within the field of AI. Yet, for someone new to the topic it’s quite a difficult term to grasp. This post explains MLOps in simple terms and outlines why you should consider deploying it in your organization.

With our MLOps Advisory services, we collaborate with our clients, their business and IT teams to integrate MLOps best practices into their organization to increase the success rate of their machine learning projects and successfully unlock the value of AI.

As we often get questions about our MLOps best practices and the tech stack we use, we gladly give a short overview of how we apply MLOps at ML6 below.

Let’s look into a few pitfalls that we’ve encountered in our MLOps advisory services and ML in production projects if you scale AI without the right team, budget and MLOps. 

After transformers disrupted natural language processing and computer vision, they now set their eyes on the biggest data type out there: tabular data. In this blog post, we will have a 10 000 feet (3048 m) view at TabTransformer, a paper by Amazon Web Services. First, we'll answer why we can apply transformers to tabular data. Then, we'll see how they rock tabular data.

ML6 worked together with CNH to accelerate the integration of their AI algorithms into existing embedded hardware. The goal was to incorporate functionality so that the combine can automatically alter its internal parameters to optimise harvest quality. The combines are equipped with a camera and hardware to detect the quality of the crop that is being harvested by running image segmentation on the hardware and images taken by the camera.

With our Advisory services, we support multiple clients in speeding up the adoption of AI and Advanced Analytics by giving concrete recommendations for their data & cloud architecture and application landscape. In Koen's latest blogpost, he explains our view on the modern enterprise team and shares five tips and tricks on how to make the most of your data and organize for success.

Data is the “fuel” for ML. Yet, in many projects, it is still challenging to get the data right due to a lack of documentation, data quality issues, missing historical data, scalability issues with data platforms and overall lack of ownership.

In this blogpost, we will use a new, user friendly technique for time series data mining: calculating the matrix profile. Unfortunately, we can’t reveal the real data or application, so we’ll substitute the original problem with a new one. To conclude we will lay out the pros and cons of gthe matrix profile.