🗺 ❓What is the current ML value chain landscape? Help us shape it!
Today we have a very special project for you – as we know you are really plugged into the ML world. We offer you to shape an objective landscape of the ML Value Chain. You've probably seen some AI/ML companies’ landscapes before. They are typically assembled by either analyst firms (e.g. CB Insights 100 AI Companies), or media (e.g. Forbes Top-50 AI Companies), or VC firms (e.g. FirstMark’s Machine Learning, AI and Data (MAD) Landscape). But we trust that TheSequence’s audience only can shape an accurate landscape of the ML Value Chain. Participate and be the first to receive this super useful research shaped by you!
Some context
The AI market is booming more now than ever, and the demand for environments that can support ML processes is on the rise. These processes tend to fall into six distinct stages of the ML value chain:
Data collection:
This is the beginning of the ML lifecycle. Several qualitatively different solutions that fulfill different needs are available on the market. Simply put, this stage is about obtaining raw and/or unstructured data.
Data processing:
This stage is about preparing raw data for labeling or testing. It may involve data cleaning, filtering, visualization, and searching for data insights through any number of methodologies.
Data annotation:
Considering that 80% of all AI project time is spent managing data, this is one of the most decisive stages. Human-handled, automatic, and hybrid solutions exist.
Model training and evaluation:
Model training under supervised learning is about getting your model to learn a function by fitting an input to an output using an input-output example (this is where the labeled data comes in). The training model has to accurately predict the output. During evaluation, you assess the model’s correctness by using a separate dataset that wasn’t used in training.
Model deployment:
Sadly, much research indicates that many if not most ML models never make it into production, because of using inferior data that renders training models and AI products inadequate.
Model monitoring:
This stage is about keeping a close eye on your ML model in order to avoid such things as model degradation, data drift or concept drift. It comes down to having your model perform at a satisfactory level in the long term.
Proposal
Many companies that offer ML services tend to offer them in some but not all stages of the chain; for example, some do only annotation, while others focus primarily on evaluation, deployment, and monitoring. Since most ML projects fail before deployment, we feel that there’s an unfulfilled need for a comprehensive, easy-to-navigate map that will showcase market solutions valued by the end user, making it easier for ML teams to reach their goals quickly and effectively.
We’ve made the first version of the ML value chain landscape specifically for this project (it’s a compilation of existing landscapes), which is meant to serve as a starting point and help you shape your ideal landscape.
It’s important to keep in mind that the landscape above is the compilation of existing research: we count on you to judge if it's accurate and help us shape a more precise picture. This version will change as more information and recommendations come from you. Share your insights and shape the best possible version of this landscape that reflects today’s ML value chain processes and needs fairly and accurately.
Please do not share this compilation with anybody yet. With your help, we hope to make it more accurate. After analyzing all the responses and making adjustments, you will be the first to receive the accurate landscape of the ML Value Chain with a lot of useful insights.
Thank you! 💜