This concept was informed by personal ML experiences,
and the result of a rapid design jam.
The concept is currently being tested.
Screen Reader Only Text:
From AI Explorer to ML Expert
ML Designer makes browsing and creating Machine Learning models simple for both AI hobbyists and deep learning engineers. Streamlined data management, powerful visualization tools, and the ability to test and iterate models in real time means creating with Machine Learning has never been easier.
Powerful but approachable.
ML Designer allows everyone to quickly create and preview Machine Learning (ML) models, without a single line of code. However, for the curious, and the experts, the inner workings of models are always accessible. Models can be fully customized, from blurring image data to the size of an individual pooling layer.
Whether you’re looking to label whale sounds, or want to dance around with the latest skeletal tracking model, we have the right template to get you started. You can even import your own neural network architecture or a pre-trained model.
Create data anywhere, from anything.
Whether you’re photographing plants in the Daintree Forest, sketching dance routines, or recording penguin calls in Antarctica, you can create training data whevere you are. ML Designer is created to run on mobile devices, letting you import files or use the various sensors on your devices to create your own model on the go.
Seven popular categories of data are currently supported. A custom datatype category also exists, supporting community extensions like HTML and DICOM.
To ensure you have a good quality model, the most important aspects of your data are presented with tips for when a dataset might need adjustments. Advanced options like pre-processing* and data-splitting* are automated, but can always be customized.
Use our AutoML models, inspect famous models, or create your own.
By default, a model is autonomatically suggested based on analysis of your dataset. Whether you have 30,000 motion recordings, or 10 sketches, you’ll always have a starting point. You can also optimize your model depending on if your application needs to prioritize speed or accuracy.
ML Designer offers the ability to dive deep into models. Users can browse the latent space and embeddings of models, freeze certain layers for training, or train from scratch. Model architectures can be edited either visually, or through code.
ML Designer provides full transparency on the estimated cost associated with the training optinos you’ve selected. From the time expended, to the economical and environmental costs. Training parametres, like the maximum number of iterations, can alter these costs in real time. Training can be done locally on device, on a nearby computer, or on a server.
Evaluate a model and learn how to improve it.
Features like Version History, and Metrics like precision and recall reveals how your model performs over time. Evaluation Reports can also be shared. These summarize your data, biases, and model’s performance in an easy-to-understand manner.
Intelligent tips, confusion matrices, similarity indices, activation visualizations, and 3D UMAP embedding visualizations, help combat data biases by highlighting areas of your model that may require data augmentation.
View your model live, improve your model, and share it with the world.
Test your model by importing test data or using sensors to for live interence. Turn your camera, microphone, or the motion sensors in your Apple Watch, into a source of test data.
Easily annotate and re-add test data into your dataset to improve your model further. This rapid feedback loop allows you to quickly improve the accuracy of your model.
Once you have a good model, share it with the world! Models can be exported to iOS, Android, and the web. Models can also be exported with code wrappers created just for your target platform, so you always have a template to start creating something amazing!