We start with a free call to discuss your goals, your data, and your efforts so far. We then typically require samples of your images and data in order to begin research into possible solutions. Investigating whether an out of the box solution or easy to use package already exists for your task is always our first step. If not - which is typical for the tasks we work on - we scour the research literature for related problems and insights. From our research, we write a custom report for you detailing possible solutions to achieve your goals with your data.

We can then discuss options for achieving your goals on this project. You may be interested in implementing a solution in-house using our Advising services to guide you, or we can take on the challenge with our Custom Solutions services.

Schedule a free call to discuss how we can accelerate your project with machine learning.

Do you already have labeled images and wonder whether two or more groups of these images can be separated based on their appearance? Perhaps this is a very challenging task for human experts. Or maybe the best trained expert cannot distinguish these classes based on the images alone. Let us give it a shot with our in-house suite of deep learning software. This powerful form of image representation captures complex and abstract properties beyond the capability of the best-trained human experts.

We will provide you with an analysis of classification results and options for further investigation. If the groups of images were successfully distinguished, this provides you with a lower bound on the classification accuracy to expect. We will suggest further options to improve performance. If the classes were not differentiable with a reasonable accuracy, we may be able to suggest more advanced methods to try or data gathering and labeling strategies to improve your results.

We've had previous success in predicting molecular properties of tumors that are too complex for expert pathologists to assess from images alone.

Schedule a free call to discuss your image data set and the challenges you currently face.

We support technical teams as they develop and fine-tune their own in-house data-driven image analysis solution. At the beginning of a project, we provide guidance on methods to try and how to structure data. Regular meetings with the key players on your team keep us up-to-date on how the project is progressing. We provide feedback on next steps: algorithm modifications, pointers to software packages or research literature, performance metrics to track progress, data requirements. We work with you to solve your image-based task with machine learning.

Schedule a free call to discuss how our advising services can help guide you to a successful data-driven solution.

We design and build software for unique challenges that generic software packages currently cannot solve. Our proven process includes working with you to understand the application area and desired result, integrating and refining cutting-edge research, prototyping a functional solution, analyzing algorithm performance, and ongoing support.

Schedule a free call to discuss how we can accelerate your project with machine learning.

1) Discovery Call

Our process begins with a discovery call to discuss your goals, your data, and your efforts so far. We provide further information about our expertise and answer any questions. The goal is for us to understand your problem and to determine if we have a mutual fit. This call is always free of charge.

2) Evaluation

Next, we begin a comprehensive analysis in which we review your existing data, annotations, and requirements. We review the scientific literature and make a plan for a functional prototype. Through this process we gain insight into whether sufficient training data is available, more data should be gathered, existing data should be further annotated, or if we can supplement with a public or synthetically-generated data set. We deliver a detailed report for a proposed algorithmic solution. This might include a progression of model components so that simpler techniques can be tried first.

3) Prototype

We implement the proposed solution, often as a series of functional prototypes to reduce the risk and demonstrate the capability of each model. We run experiments in parallel to develop an optimal data-driven solution. Performance analysis enables us to understand algorithm accuracy and failure modes. We iterate as needed, perhaps to gather more data, more annotations, stronger annotations, or by improving our algorithm. We provide regular status updates, and we deliver a final report and code for a functional prototype.

4) Ongoing Support

We transition the functional prototype to your in-house team for integration into your product or service, and we provide support to your team as needed.