Frequently Asked Questions
Frequently Asked Questions
Jump to a question:
What is rTLBO?
rTLBO is our flagship IP, enabling Pharma to robustly learn from their historical experimental data.
What can rTLBO be used for?
rTLBO can be used for any kind of experimentation where you have access to historical experimental data. (This usually is the case in most laboratories pursuing long-term development goals)
What is the impact of rTLBO?
Most simply, it's an insurance for unreliable historical data. That data might be unreliable for several reasons:
Human risk: Data entry was faulty
Hardware risk: Often, hardware degrades unnoticed, e.g. such as a pipetting arm or a measurement sensor
Modeling risk: The historical data is not as relevant to the current experiment as hoped
We put together a impact simulator to estimate the savings you can achieve with rTLBO.
What is the science behind rTLBO?
As rTLBO is a pending patent, we are limited on how much we can publicly state on the science that powers it. Given appropriate NDAs, we are happy to share it's inner workings. Alternatively, we will also link the patent application here, once it's public.
rTLBO was developed with Pharma's experimentation challenges in mind. It is the result of multi-year R&D investment in Machine Learning Research, pulling together a few key ideas of Statistics that allow reliable assessment of historical data.
Who else is using rTLBO?
While it's main use is with Pharma companies that have an abundance of relevant historical experimental data and the ability to implement rTLBO, it has also been used in smaller laboratories and also for applications outside of Pharma, Chemistry and Materials.
What is MFBO?
MFBO allows scientist to combine and pool data from different experimental sources. For example, there could be a target experiment, such as a production facility, as well as some pilot plants.
The production plant is expensive to experiment on, while the pilot plant can produce 10 samples a day with a much smaller budget than production. By pooling these two data sources, we can get to the optimal operation parameters faster than with just the production plant. See our resarch for more details.
Our expertise lies in understanding the limitations of MFBO and how to avoid common modeling pitfalls that lead to suboptimal, often worse than SFBO, performance.
What can MFBO be used for?
MFBO applies to any expermental setup with more two or more different ways to do the same experiment with similar outcomes.
Some of those experiments can be very precise, but costly because of that. (i.e. high fidelity)
Others can be rather approximate, but cheap and/or fast. (i.e. low fidelity)
What is the impact of MFBO?
MFBO has been shown to find an optimum faster than experimenting with only a single fidelity.
The below plot shows how MFBO improves faster than a single fidelity approach, up to a certain budget level.
Who else is using MFBO?
MFBO has found widespread adoption in the Pharma, Chemicals and Materials development. A quick Google Scholar search yields about 7500 papers on the subject.
Earlier work includes "Multi-fidelity machine learning models for accurate bandgap predictions of solids". More recent work includes "Multi-fidelity Bayesian Optimisation for Syngas Fermentation Simulators".
Licensing Plans
How can I learn more about the trial/full-access licensing plan?
We have briefly summarised the license packages on our product page.
Further details are most easily discussed with our experts, who'd be happy to learn more about your specific problems and how our IP can help solve those.
How much does it cost?
As a general principle, our costing is entirely dependent on the value our IP provides. We do not intend to overcharge you for low-to moderate use of our IP.
In the same way, we expect appropraite compenstation for applications where our IP delivers significant value.
Most simply, get in touch with our experts or alternatively, build intuition with our business impact simulator, e.g. for rTLBO.
What sort of support will I get throughout the licensing period?
IP support depends on the licensing model you choose. We aim to provide full support packages as part of our full-access license models.
For the trial packages, we have more defined expectations of the support we provide as the engagement is primarily focused on delivering value, fast, to your problems.
For education, we are limited in the support we provide, but aim to be swift in sharing references and further material to accelerate your learning.
How will your method be implemented and integrated with my data systems?
We provide our IP in most shapes and forms supported by your organisation.
This usually ranges from simple software packages build on top of your current stack, to guiding you through a complete software implementation roadmap for Bayesian Optimisation.
Put simply, implementation is smaller issue compared to the concrete scientific problems and how you can make our IP work in your organisation.
What are your data privacy/protection policies?
Since we do not store or track any data, but only provide IP, we are not at risk of privacy and protection breaches.
What happens when my Trial-Access ends?
For Trial-Access, it is our utmost priority to provide the maximal value of our IP to solving your concrete optimisation problem.
Part of that is also the preparation of transition to Full-Access. By design, this will be an organic transition and part of end-of-trial milestone assessment.
Any further questions?
Interested in learning more about our ML IP tools, licensing models and research consulting services?