Foundation Model Assessment

Foundation models are popping up everywhere – do you need one for your proprietary image dataset?

A domain-specific foundation model trained with self-supervised learning on your unique data could bring a number of benefits:

Less reliance on labeled data
Improved generalizability to domain shifts
The ability to use all channels of multispectral or multiplex images
Faster prototyping of new models

But they can be time-consuming and costly to train. And do you even have enough data to train one? How long will it take? Do you have the resources available?

What if you knew exactly what the ROI would be?

And which benefits your team is likely to see. What resources it will take to build one. Whether it is something to prioritize now versus next year. And if there is a quicker way to get started.

Get a clear perspective on whether you can benefit from a foundation model

Though this assessment, I will answer the following questions:

Would you benefit from a foundation model trained on your proprietary images?
Are there publicly available models you could try before training your own?
Do you have the capability to train one yourselves?
What are some of the factors you’d need to consider when training one?

Here’s how it works:

You’ll start by filling out a questionnaire to give me some insights into your goals, data, and relevant team members. I’ll interview some of your leadership, domain expert, and machine learning team members to discover your overall goals, the nuances of the image datasets you have available, and the capabilities of your team. With this knowledge, I’ll deliver a report with recommendations to answer the questions above. A follow up call after report delivery is also included to ensure that you have all the information you need to move forward.

Results you can expect from this assessment:

Save time and expenses by not pursuing a foundation model if it is unlikely to help you achieve your goals.
A clear understanding of how a foundation model can help you achieve your goals.
Recommendations on publicly available models to try before building your own.
Know whether your current team is capable of building a foundation model – or what expertise they’re lacking.
Gain some insights into the nuances and complexity of training a good foundation model so that you can better prepare for the task ahead.

Ready to get started?

Are you ready to get a clear perspective on the ROI of a foundation model for your proprietary image dataset? Apply for a Foundation Model Assessment by clicking the button below. The price is $5,000, which could save your team many months of experimentation if you choose the wrong path – and hundreds of thousands of dollars.

Apply Now

Don’t just take my word for it…

Heather has accelerated the development of our machine learning projects by challenging the status quo and introducing the team to new approaches. She helped us get out of the echo chamber and brought an outside perspective with very relevant alternative approaches that would have taken a lot longer to do on our own.

-- Joe Sturonas, Ancera, VP of Systems Development

Heather has deep understanding of digital pathology and machine learning and their application to whole slide images. Her in-depth review of the current state of the art research has enabled Gestalt to rapidly focus our machine learning efforts on approaches that are yielding value for our pathologists and their patients.

-- Brian Napora, Gestalt Diagnostics, VP of Sales & Product Management

Heather provided practical feedback and guidance, which allowed us to quickly take steps in the right direction and improve the performance of our models.

-- Darragh Maguire, Deciphex, Artificial Intelligence Engineer

Still have questions?

What happens after I apply?

If needed, we’ll schedule a short call to confirm that this package is suitable for your goals. Then I’ll send over a contract. Once you submit your payment, you’ll receive an email with a link to a questionnaire that I’ll use to get some background information and begin scheduling calls with a few of your team members.

Will you sign an NDA before we start?

The contract will include confidentiality and non-disclosure provisions.

What are some of the benefits we might realize with a domain-specific foundation model?

Check out this article in which I elaborate on the benefits: Unlock the Potential of Your Proprietary Images with a Domain-Specific Foundation Model.

Can’t we just use one of the many publicly available foundation models?

If there is one trained on images similar to yours, definitely, yes! But you may still benefit from a foundation model trained on your own imagery. And if your images are much different than those used to train publicly available models – perhaps they’re even multispectral or multiplex – you still might be able to use transfer learning, but you’ll get less benefit than a model trained on your unique imagery.

I don’t get it. How can you train a model without labeled data?

Most foundation models are built using an approach called self-supervised learning, which uses a pretext task to learn a representation that can benefit downstream tasks. The pretext task could involve predicting cluster labels, making nearby patch encodings similar and far away ones different, reconstructing some missing parts of an image, or some other domain-specific task like predicting the magnification level. In order to complete the task, the model must learn how to represent the images and the patterns within them. Once a model has been trained with one of these pretext tasks, it can be transferred with or without finetuning to another task on similar imagery.

So a foundation model just involves training a self-supervised model on our dataset?

Yes, that’s a fundamental component. But training a good foundation model that is generalizable to many downstream tasks requires more than that. It requires understanding the data in order to balance the inputs presented to the model, adapt to any domain shifts or batch effects, apply image augmentation appropirately, and select a suitable pretext task. While contrastive and reconstruction pretext tasks are most common, they may not be the best objectives for every dataset. Selecting an appropriate pretext task for your data requires understanding some of the nuances of your image dataset.

I don’t think we have enough compute power to train a foundation model.

That might be the case. The goal of this assessment is to evaluate the pros and cons of building a foundation model on your image dataset. Please don’t jump to conclusions until we have all the facts.

I don’t think we have a large enough dataset to train a foundation model.

If your dataset is small, you won’t be able to train a foundation model with 100 million parameters. But you still could train a smaller one that enables you to apply deep learning to downstream tasks that you’ve consistently overfit before. Please don’t jump to conclusions on whether a foundation model is suitable until we have all the facts.

What’s next after the assessment? Can you help us with implementation?

The recommendations from this assessment will determine what next steps might be suitable. If we decide that a foundation model would be beneficial for you and that your team has the capabilities to build it, I can guide you through this process as part of my monthly Advisory Services. If your team does not have the capabilities – and depending on the size of your dataset and expected complexity of the project – we might be able to scope a custom project in which I build the foundation model for you.

Still not sure if this is the right package for you?

Schedule a free Discovery Call, and I’ll help you determine if this assessment is suitable for you.

Availability is limited

I only offer two Foundation Model Assessments per month. Scheduling is first come, first served. The sooner you apply, the sooner you will have a clear perspective on foundation models.