Train state-of-the-art AI models
Using your own data, you can train state-of-the-art AI models based on deep learning by following a simple wizard in Criterion AI. You get to select from a wide range of preimplemented algorithms hosted in our model zoo. These algorithms can train image classifiers, segmentation models, and anomaly detectors to perform at superhuman levels—just to name a few.
You can explore all of the models that are available to you by creating a free account on our Starter plan. Here are three of our featured models that many of our customers use.
Ever since the convolutional neural network AlexNet won the ImageNet Large Scale Visual Recognition Competition in 2012, researchers have made incredible improvements in algorithms to solve tasks related to image recognition. This means that we can now offer superhuman performance levels at classification tasks, which you can take advantage of directly from Criterion AI.
Image segmentation is the task of dividing an image into meaningful segments. Just like algorithms for image classification have improved exponentially since 2012, image segmentation algorithms (such as U-Net) have advanced tremendously and can now be applied to relief many pharmaceutical workloads such as analysis of agar plates, Petri dishes, and microscopic images.
Anomaly detection is the task of finding anomalies in large amounts of datasets or streams. Using so-called generative models, algorithms can now be trained to detect anomalies in high-resolution images, videos, NIR and Raman spectra and more. Most of them work by learning a distribution of the data in order to detect when a sample lying far away from the mean is observed.
Train with GPUs in the cloud
Criterion AI is based on Google Cloud Platform where we can leverage the availability graphics processing units (GPUs) as well as tensor processing units (TPUs)—i.e., very powerful hardware, which we need to train our AI models. The great thing about being based on Google’s cloud infrastructure is that we require no up-front investments in expensive computing equipment. After creating an account in Criterion AI, you can be training state-of-the-art models on some of the world’s most advanced hardware in literally a matter of minutes. 💪
Though, there may be some scenarios where customers want to train their models locally and we do support that. Please get in touch with us if that is a requirement you have.
Build your own custom models
If you deal with a scenario that can’t be solved by one of the many algorithms in our model zoo, you can easily build your own. We work with all of the open-source tools that the vast majority of data scientists work with—i.e., Python, Keras, TensorFlow, scikit-learn and related frameworks. We have made it easy for you to build your own algorithm in Python, wrap it into a package and run it on Criterion AI.
We have an entire section in our documentation dedicated to explaining how custom models in Criterion AI work. If you are interested in getting started with building your own model, we would love to help you out. Please send us a message with a description of your use case and we will happily set aside resources to support you in the development process. 🤝